Authors: Yi Liu
Abstract: We discover that large language models exhibit \emph{spectral phase transitions} in their hidden activation spaces when engaging in reasoning versus factual recall. Through systematic spectral analysis across \textbf{11 models} spanning \textbf{5 architecture families} (Qwen, Pythia, Phi, Llama, DeepSeek-R1), we identify \textbf{seven} core phenomena: (1)~\textbf{Reasoning Spectral Compression} -- 9/11 models show significantly lower $\alpha$ for reasoning ($p < 0.05$), with larger effects in stronger models; (2)~\textbf{Instruction Tuning Spectral Reversal} -- base models show reasoning $\alpha < $ factual $\alpha$, while instruction-tuned models reverse this relationship; (3)~\textbf{Architecture-Dependent Generation Taxonomy} -- prompt-to-response shifts partition into expansion, compression, and equilibrium regimes; (4)~\textbf{Spectral Scaling Law} -- $\alpha_\text{reasoning} \propto -0.074 \ln N$ across 4 Qwen base models ($R^2 = 0.46$); (5)~\textbf{Token-Level Spectral Cascade} -- per-token alpha tracking reveals local synchronization that decays exponentially with layer distance, and is weaker for reasoning than factual tasks; (6)~\textbf{Reasoning Step Spectral Punctuation} -- phase-transition signatures align with reasoning step boundaries; and (7)~\textbf{Spectral Correctness Prediction} -- spectral $\alpha$ alone achieves AUC $= 1.000$ (Qwen2.5-7B, late layers) and mean AUC $= 0.893$ across 6 models in predicting correctness \emph{before} the final answer is generated. Together, these findings establish a comprehensive \emph{spectral theory of reasoning} in transformers, revealing that the geometry of thought is universal in direction, architecture-specific in dynamics, and predictive of outcome.
Authors: Abdulmalek Saket
Abstract: Low-Rank Adaptation (LoRA) has become the dominant parameter-efficient fine-tuning method for large language models, yet standard practice applies LoRA adapters uniformly to all transformer layers regardless of their relevance to the downstream task. We introduce Aletheia, a gradient-guided layer selection method that identifies the most task-relevant layers via a lightweight gradient probe and applies LoRA adapters only to those layers with asymmetric rank allocation. Across 81 experiment rows covering 14 successful models from 8 architecture families (0.5B-72B parameters, including dense and Mixture-of-Experts architectures), with one additional documented failed Pythia/GPT-NeoX attempt in Campaign 2, Aletheia achieves a 15-28% training speedup (mean 23.1%, p < 0.001) with bounded extra forgetting and broadly matched downstream behavior on the evaluated MMLU, GSM8K, and HumanEval benchmark pack. Across the tested families and scales, Campaign 1 shows a 100% per-model speed win rate and Campaign 2 shows broadly preserved downstream behavior within a bounded-degradation framing. Together these results support a practical model-economics claim: intelligent layer selection can make LoRA fine-tuning materially more efficient without introducing major downstream damage on the evaluated set.
Authors: Gregory Magarshak
Abstract: Recent work on KV cache quantization, culminating in TurboQuant, has approached the Shannon entropy limit for per-vector compression of transformer key-value caches. We observe that this limit applies to a strictly weaker problem than the one that actually matters: compressing the KV cache as a sequence. The tokens stored in a KV cache are not arbitrary floating-point data -- they are samples from the exact formal language the model was trained on, and the model is by construction a near-optimal predictor of that language. We introduce sequential KV compression, a two-layer architecture that exploits this structure. The first layer, probabilistic prefix deduplication, identifies semantically equivalent shared prefixes across sessions using the trie metric d_T(s, s') = -log_2 P_M(s ^ s') from Probabilistic Language Tries (PLTs). The second layer, predictive delta coding, stores only the residual of each new KV vector from the model's own prediction of it, achieving a per-token entropy bound of H(KV_{i+1} | KV_{<=i}) <= H(token_{i+1} | token_{<=i}). We prove that at typical language model perplexity -- approximately 10-20 for fluent English text -- this bound is 3.3-4.3 bits on average per token position, compared to TurboQuant's 3 bits per vector component (with typical attention heads having 64-128 components). The theoretical compression ratio over TurboQuant is approximately 914,000x at the Shannon limit. Even at 1000x above the entropy floor -- a deliberately pessimistic worst-case overhead, two orders of magnitude above the 2-5x typical of practical source coders -- the ratio remains approximately 914x over TurboQuant, with compression improving rather than degrading as context length grows. The two layers are orthogonal and compose with existing per-vector quantization methods including TurboQuant.
Authors: Jaime de Miguel Rodriguez, Artjom Vargunin, Brigitta Robin Raudne, David Solis Martin, Yaroslava Mykhailenko, Kaarel Oja
Abstract: This study presents a triadic analysis of energy storage operation under multi-stage model predictive control, investigating the interplay between data characteristics, forecast uncertainty, planning horizon, and battery c-rate. Synthetic datasets are generated to systematically explore variations in data profiles and uncertainty, enabling parametrization and the construction of relationships that map these characteristics to optimal horizon length. Results reveal the presence of an effective horizon, defined as the look-ahead length beyond which additional forecast information provides limited operational benefit. Accounting for this horizon can reduce computational costs while maintaining optimal performance. The study provides optimal horizon lengths across a broad range of combinations of battery types, uncertainty levels, and data profiles, offering practical guidance for industrial storage operation. It also quantifies revenue losses due to forecast uncertainty, showing that errors can impact performance even for fast batteries. Finally, the framework lays the groundwork for future machine learning approaches that map dataset parametrization to optimal horizons, supporting continuous optimization in industrial settings without heavy computation.
Authors: Sanjeev Panta, Rhett M Morvant, Xu Yuan, Li Chen, Nian-Feng Tzeng
Abstract: Accurate and timely rainfall nowcasting is crucial for disaster mitigation and water resource management. Despite recent advances in deep learning, precipitation prediction remains challenging due to limitations in effectively leveraging diverse multimedia data sources. We introduce M3R, a Meteorology-informed MultiModal attention-based architecture for direct Rainfall prediction that synergistically combines visual NEXRAD radar imagery with numerical Personal Weather Station (PWS) measurements, using a comprehensive pipeline for temporal alignment of heterogeneous meteorological data. With specialized multimodal attention mechanisms, M3R novelly leverages weather station time series as queries to selectively attend to spatial radar features, enabling focused extraction of precipitation signatures. Experimental results for three spatial areas of 100 km * 100 km centered at NEXRAD radar stations demonstrate that M3R outperforms existing approaches, achieving substantial improvements in accuracy, efficiency, and precipitation detection capabilities. Our work establishes new benchmarks for multimedia-based precipitation nowcasting and provides practical tools for operational weather prediction systems. The source code is available at https://github.com/Sanjeev97/M3Rain
Authors: Kang An, Chenhao Si, Shiqian Ma, Ming Yan
Abstract: Physics-Informed Neural Networks (PINNs) often suffer from slow convergence, training instability, and reduced accuracy on challenging partial differential equations due to the anisotropic and rapidly varying geometry of their loss landscapes. We propose a lightweight curvature-aware optimization framework that augments existing first-order optimizers with an adaptive predictive correction based on secant information. Consecutive gradient differences are used as a cheap proxy for local geometric change, together with a step-normalized secant curvature indicator to control the correction strength. The framework is plug-and-play, computationally efficient, and broadly compatible with existing optimizers, without explicitly forming second-order matrices. Experiments on diverse PDE benchmarks show consistent improvements in convergence speed, training stability, and solution accuracy over standard optimizers and strong baselines, including on the high-dimensional heat equation, Gray--Scott system, Belousov--Zhabotinsky system, and 2D Kuramoto--Sivashinsky system.
Authors: Tomasz S{\l}u\.zalec, Marcin {\L}o\'s, Askold Vilkha, Maciej Paszy\'nski
Abstract: We explore the possibility of solving Partial Differential Equations (PDEs) using discrete weak formulations. We propose a programming environment for defining a discrete computational domain, introducing discrete functions defined over a set of points, constructing discrete inner products, and introducing discrete weak formulations employing Kronecker delta test functions. Building on this setup, we propose a discrete neural network representation, training the solution function defined over a discrete set of points and employing discrete finite difference derivatives in the automatic differentiation procedures. As a challenging computational model example, we focus on Stokes equations in two-dimensions, defined over a discrete set of points. We train the solution using the discrete weak residual and the Adamax algorithm with discrete automatic differentiation of the discrete gradients. Despite introducing the python environment, we also provide a rigorous mathematical formulation based on discrete weak formulations, proving the well-posedness and robustness of the loss function. The solution of the discrete weak formulations is based on neural network training employing a robust loss function that is related to the true error. In this way, we have a robust control of the numerical error during the training of the neural networks. Besides the Stokes formulation, we also explain the functionality of the proposed library using the Laplace problem formulation.
Authors: G. Aytug Akarlar
Abstract: We present causal evidence that hallucination in autoregressive language models is an early trajectory commitment governed by asymmetric attractor dynamics. Using same-prompt bifurcation, in which we repeatedly sample identical inputs to observe spontaneous divergence, we isolate trajectory dynamics from prompt-level confounds. On Qwen2.5-1.5B across 61 prompts spanning six categories, 27 prompts (44.3%) bifurcate with factual and hallucinated trajectories diverging at the first generated token (KL = 0 at step 0, KL > 1.0 at step 1). Activation patching across 28 layers reveals a pronounced causal asymmetry: injecting a hallucinated activation into a correct trajectory corrupts output in 87.5% of trials (layer 20), while the reverse recovers only 33.3% (layer 24); both exceed the 10.4% baseline (p = 0.025) and 12.5% random-patch control. Window patching shows correction requires sustained multi-step intervention, whereas corruption needs only a single perturbation. Probing the prompt encoding itself, step-0 residual states predict per-prompt hallucination rate at Pearson r = 0.776 at layer 15 (p < 0.001 against a 1000-permutation null); unsupervised clustering identifies five regime-like groups (eta^2 = 0.55) whose saddle-adjacent cluster concentrates 12 of the 13 bifurcating false-premise prompts, indicating that the basin structure is organized around regime commitments fixed at prompt encoding. These findings characterize hallucination as a locally stable attractor basin: entry is probabilistic and rapid, exit demands coordinated intervention across layers and steps, and the relevant basins are selected by clusterable regimes already discernible at step 0.
Authors: Saif Mahmoud, Ahmad Almasri
Abstract: Token pruning methods for Vision Transformers (ViTs) promise quadratic reductions in attention FLOPs by dropping uninformative patches. Yet when pruned sequences are executed with state-of-the-art variable-length attention APIs -- including FlashAttention-2's varlen and PyTorch's NestedTensor SDPA-the wall-clock attention latency doesn't scale accordingly. We trace this to a dispatch-overhead bottleneck: at the short, post-pruning sequence lengths typical of ViTs (<=197 tokens), actual matrix arithmetic completes in single-digit microseconds while the host-side dispatch path consumes 60-90 us. We present a lightweight, bidirectional Triton attention kernel whose dispatch floor is 40 us roughly 1.5x lower than FlashAttention-2 varlen-allowing pruning savings to become more visible in wall-clock time. Integrated into a complete pack-attend-unpack pipeline, our system achieves up to 2.24x end-to-end throughput over padded PyTorch SDPA consistently across four pruning algorithms (Threshold-L2, DynamicViT, EViT, ATS), scales across DeiT-T/S/B, and maintains bit-exact classification predictions with <0.007 max absolute logit difference.
Authors: Ranjith Chodavarapu, Lei Xu
Abstract: KV caching is a ubiquitous optimization in autoregressive transformer inference, long presumed to be numerically equivalent to cache-free computation. This assumption fails under standard FP16 precision: cache-ON and cache-OFF execution paths employ different floating-point accumulation orderings which, due to FP16 non-associativity, produce a deterministic divergence in decoded token sequences. Across three open-weight models (LLaMA-2-7B, Mistral-7B-v0.3, Gemma-2-2B) evaluated on GSM8K, we observe a 100\% token divergence rate across all sampling strategies, including greedy decoding, which rules out sampling randomness as a cause, and also with cache-ON yielding higher accuracy in 8 of 9 conditions, where the accuracy difference serves as an indicator that the divergence direction is systematic rather than random. Controlled FP32 falsification reduces divergence by eight orders of magnitude, eliminates token flips, and drops the flip rate to exactly 0.0\%, confirming FP16 non-associativity as the sole causal driver. Layer-wise drift profiling reveals architecturally predictable propagation patterns: models using Grouped-Query Attention exhibit sharp divergence at the first layer, while Gemma's larger head dimension and sliding window attention produce uniform accumulation across all layers. Finally, activation patching of the entire residual stream fails to recover the cache-free trajectory, localizing the causal variable to the stateful KV cache. These findings establish that FP16 KV cache inference is fundamentally non-equivalent to recomputation and provide a mechanistic framework for understanding numerical instability in modern LLM inference systems.
Authors: Tingjia Miao, Wenkai Jin, Muhua Zhang, Jinxin Tan, Yuelin Hu, Tu Guo, Jiejun Zhang, Yuhan Wang, Wenbo Li, Yinuo Gao, Shuo Chen, Weiqi Jiang, Yayun Hu, Zixing Lei, Xianghe Pang, Zexi Liu, Yuzhi Zhang, Linfeng Zhang, Kun Chen, Wei Wang, Weinan E, Siheng Chen
Abstract: The paradigm of agentic science requires AI systems to conduct robust reasoning and engage in long-horizon, autonomous exploration. However, current scientific benchmarks remain confined to domain knowledge comprehension and complex reasoning, failing to evaluate the exploratory nature and procedural complexity of real-world research. In this work, we present research-oriented evaluations in theoretical and computational physics, a natural testbed with comprehensive domain knowledge, complex reasoning, and verifiable end-to-end workflows without reliance on experiments. Here we introduce PRL-Bench (Physics Research by LLMs), a benchmark designed to systematically map the capability boundaries of LLMs in executing end-to-end physics research. Constructed from 100 curated papers from the latest issues of Physical Review Letters since August 2025 and validated by domain experts, PRL-Bench covers five major theory- and computation-intensive subfields of modern physics: astrophysics, condensed matter physics, high-energy physics, quantum information, and statistical physics. Each task in the benchmark is designed to replicate the core properties of authentic scientific research, including exploration-oriented formulation, long-horizon workflows, and objective verifiability, thereby reconstructing the essential reasoning processes and research workflows of real physics research. Evaluation across frontier models shows that performance remains limited, with the best overall score below 50, revealing a pronounced gap between current LLM capabilities and the demands of real scientific research. PRL-Bench serves a reliable testbed for accessing next generation AI scientists advancing AI systems toward autonomous scientific discovery.
Authors: Lute Lillo, Nick Cheney
Abstract: Continual reinforcement learning must balance retention with adaptation, yet many methods still rely on \emph{single-model preservation}, committing to one evolving policy as the main reusable solution across tasks. Even when a previously successful policy is retained, it may no longer provide a reliable starting point for rapid adaptation after interference, reflecting a form of \emph{loss of plasticity} that single-policy preservation cannot address. Inspired by quality-diversity methods, we introduce \textsc{TeLAPA} (Transfer-Enabled Latent-Aligned Policy Archives), a continual RL framework that organizes behaviorally diverse policy neighborhoods into per-task archives and maintains a shared latent space so that archived policies remain comparable and reusable under non-stationary drift. This perspective shifts continual RL from retaining isolated solutions to maintaining \emph{skill-aligned neighborhoods} with competent and behaviorally related policies that support future relearning. In our MiniGrid CL setting, \textsc{TeLAPA} learns more tasks successfully, recovers competence faster on revisited tasks after interference, and retains higher performance across a sequence of tasks. Our analyses show that source-optimal policies are often not transfer-optimal, even within a local competent neighborhood, and that effective reuse depends on retaining and selecting among multiple nearby alternatives rather than collapsing them to one representative. Together, these results reframe continual RL around reusable and competent policy neighborhoods, providing a route beyond single-model preservation toward more plastic lifelong agents.
Authors: Dingzhi Yu, Rui Pan, Yuxing Liu, Tong Zhang
Abstract: Sign-based optimization algorithms, such as SignSGD, have garnered significant attention for their remarkable performance in distributed learning and training large foundation models. Despite their empirical superiority, SignSGD is known to diverge on non-smooth objectives, which are ubiquitous in modern machine learning due to ReLUs, max-pools, and mixture-of-experts. To overcome this fundamental limitation, we propose \textbf{StoSignSGD}, an algorithm that injects structural stochasticity into the sign operator while maintaining an unbiased update step. In the regime of (online) convex optimization, our theoretical analysis shows that StoSignSGD rigorously resolves the non-convergence issues of SignSGD, achieving a sharp convergence rate matching the lower bound. For the more challenging non-convex non-smooth optimization, we introduce generalized stationary measures that encompass prior definitions, proving that StoSignSGD improves upon the best-known complexity bounds by dimensional factors. Empirically, StoSignSGD exhibits robust stability and superior efficiency across diverse large language model (LLM) training regimes. Notably, in low-precision FP8 pretraining -- a setting where AdamW fails catastrophically -- StoSignSGD remains highly stable and yields a remarkable 1.44$\times$ to 2.14$\times$ speedup relative to established baselines. Furthermore, when fine-tuning 7B LLMs on mathematical reasoning tasks, StoSignSGD delivers substantial performance gains over both AdamW and SignSGD. Finally, to dissect the mechanisms driving its success, we develop a sign conversion framework capable of transforming any general optimizer into its unbiased, sign-based counterpart. Utilizing this framework, we deconstruct the core components of StoSignSGD and present a comprehensive ablation study to empirically validate our algorithmic design choices.
Authors: Koyena Pal, Serdar Kadioglu
Abstract: Foundational optimization embeddings have recently emerged as powerful pre-trained representations for mixed-integer programming (MIP) problems. These embeddings were shown to enable cross-domain transfer and reduce reliance on solver-generated labels. In this work, we investigate whether such representations generalize beyond optimization to decision problems, focusing on Boolean satisfiability (SAT). We adapt the foundational optimization architecture to SAT by mapping CNF formulas into the same bipartite constraint-variable graph representation used for MIPs. This allows direct reuse of the pre-trained embedding model without architectural changes or supervised fine-tuning. Our results show that these embeddings capture structural regularities in SAT instances and support unsupervised tasks such as instance clustering and distribution identification. We demonstrate, for the first time, that foundational optimization embeddings can transfer to constraint satisfaction domains. Our findings is a step toward a unified representational framework for both optimization and decision problems.
Authors: Nassima M. Bouzid, Dehao Yuan, Nam H. Nguyen, Mayana Pereira
Abstract: LLM-based simulators offer a promising path for generating complex synthetic data where traditional differentially private (DP) methods struggle with high-dimensional user profiles. But can LLMs faithfully reproduce statistical distributions from DP-protected inputs? We evaluate this using PersonaLedger, an agentic financial simulator, seeded with DP synthetic personas derived from real user statistics. We find that PersonaLedger achieves promising fraud detection utility (AUC 0.70 at epsilon=1) but exhibits significant distribution drift due to systematic LLM biases--learned priors overriding input statistics for temporal and demographic features. These failure modes must be addressed before LLM-based methods can handle the richer user representations where they might otherwise excel.
Authors: Yisheng Zhong, Sijia Liu, Zhuangdi Zhu
Abstract: Large Language Models (LLMs) unlearning is crucial for removing hazardous or privacy-leaking information from the model. Practical LLM unlearning demands satisfying multiple challenging objectives simultaneously: removing undesirable knowledge, preserving general utility, avoiding over-refusal of neighboring concepts, and, crucially, ensuring robustness against adversarial probing attacks. However, existing unlearning methods primarily focus on a limited subset of these goals, typically unlearning efficacy and utility preservation while overlooking robustness and boundary behaviors. Naively extending these methods to multi-objective settings may lead to unlearning task interference. We propose a novel multi-objective unlearning framework that harmonizes multiple unlearning objectives through a data and optimization co-design: We standardize training corpora into a unified data representation to reduce the domain gap, and then introduce a bidirectional distillation method that simultaneously elicits desired behavior from a context-instructed teacher while suppressing undesirable behavior in the student model. Theoretical and empirical analyses show that our method aligns domain distributions and converts seemingly irrelevant unlearning tasks into cooperative optimization. Evaluation demonstrates state-of-the-art performance, which enables balanced and reliable unlearning across diverse, challenging requirements.
Authors: Physical Intelligence, Bo Ai, Ali Amin, Raichelle Aniceto, Ashwin Balakrishna, Greg Balke, Kevin Black, George Bokinsky, Shihao Cao, Thomas Charbonnier, Vedant Choudhary, Foster Collins, Ken Conley, Grace Connors, James Darpinian, Karan Dhabalia, Maitrayee Dhaka, Jared DiCarlo, Danny Driess, Michael Equi, Adnan Esmail, Yunhao Fang, Chelsea Finn, Catherine Glossop, Thomas Godden, Ivan Goryachev, Lachlan Groom, Haroun Habeeb, Hunter Hancock, Karol Hausman, Gashon Hussein, Victor Hwang, Brian Ichter, Connor Jacobsen, Szymon Jakubczak, Rowan Jen, Tim Jones, Gregg Kammerer, Ben Katz, Liyiming Ke, Mairbek Khadikov, Chandra Kuchi, Marinda Lamb, Devin LeBlanc, Brendon LeCount, Sergey Levine, Xinyu Li, Adrian Li-Bell, Vladislav Lialin, Zhonglin Liang, Wallace Lim, Yao Lu, Enyu Luo, Vishnu Mano, Nandan Marwaha, Aikys Mongush, Liam Murphy, Suraj Nair, Tyler Patterson, Karl Pertsch, Allen Z. Ren, Gavin Schelske, Charvi Sharma, Baifeng Shi, Lucy Xiaoyang Shi, Laura Smith, Jost Tobias Springenberg, Kyle Stachowicz, Will Stoeckle, Jiaming Tang, Jimmy Tanner, Shalom Tekeste, Marcel Torne, Kyle Vedder, Quan Vuong, Anna Walling, Haohuan Wang, Jason Wang, XuDong Wang, Chris Whalen, Samuel Whitmore, Blake Williams, Charles Xu, Sukwon Yoo, Lili Yu, Wuming Zhang, Zhuoyang Zhang, Ury Zhilinsky
Abstract: We present a new robotic foundation model, called ${\pi}_{0.7}$, that can enable strong out-of-the-box performance in a wide range of scenarios. ${\pi}_{0.7}$ can follow diverse language instructions in unseen environments, including multi-stage tasks with various kitchen appliances, provide zero-shot cross-embodiment generalization, for example enabling a robot to fold laundry without seeing the task before, and perform challenging tasks such as operating an espresso machine out of the box at a level of performance that matches much more specialized RL-finetuned models. The main idea behind ${\pi}_{0.7}$ is to use diverse context conditioning during training. This conditioning information, contained in the prompt, makes it possible to steer the model precisely to perform many tasks with different strategies. It is conditioned not just on a language command that describes what it should do, but on additional multimodal information that also describes the manner or strategy in which it should do it, including metadata about task performance and subgoal images. This enables ${\pi}_{0.7}$ to use very diverse data, including demonstrations, potentially suboptimal (autonomous) data including failures, and data from non-robot sources. Our experiments evaluate ${\pi}_{0.7}$ across numerous tasks with multiple robot platforms, on tasks that require speed and dexterity, language following, and compositional task generalization.
Authors: Zixuan Weng, Jinghuai Zhang, Kunlin Cai, Ying Li, Peiran Wang, Yuan Tian
Abstract: Large language models (LLMs) often exhibit undesirable behaviors, such as safety violations and hallucinations. Although inference-time steering offers a cost-effective way to adjust model behavior without updating its parameters, existing methods often fail to be simultaneously effective, utility-preserving, and training-efficient due to their rigid, one-size-fits-all designs and limited adaptability. In this work, we present FineSteer, a novel steering framework that decomposes inference-time steering into two complementary stages: conditional steering and fine-grained vector synthesis, allowing fine-grained control over when and how to steer internal representations. In the first stage, we introduce a Subspace-guided Conditional Steering (SCS) mechanism that preserves model utility by avoiding unnecessary steering. In the second stage, we propose a Mixture-of-Steering-Experts (MoSE) mechanism that captures the multimodal nature of desired steering behaviors and generates query-specific steering vectors for improved effectiveness. Through tailored designs in both SCS and MoSE, FineSteer maintains robust performance on general queries while adaptively optimizing steering vectors for targeted inputs in a training-efficient manner. Extensive experiments on safety and truthfulness benchmarks show that FineSteer outperforms state-of-the-art methods in overall performance, achieving stronger steering performance with minimal utility loss. Code is available at https://github.com/YukinoAsuna/FineSteer
Authors: Mohammad Mahdi Abootorabi, Parvin Mousavi, Purang Abolmaesumi, Evan Shelhamer
Abstract: Deep networks that rely on prototypes-interpretable representations that can be related to the model input-have gained significant attention for balancing high accuracy with inherent interpretability, which makes them suitable for critical domains such as healthcare. However, these models are limited by their reliance on training data, which hampers their robustness to distribution shifts. While test-time adaptation (TTA) improves the robustness of deep networks by updating parameters and statistics, the prototypes of interpretable models have not been explored for this purpose. We introduce ProtoTTA, a general framework for prototypical models that leverages intermediate prototype signals rather than relying solely on model outputs. ProtoTTA minimizes the entropy of the prototype-similarity distribution to encourage more confident and prototype-specific activations on shifted data. To maintain stability, we employ geometric filtering to restrict updates to samples with reliable prototype activations, regularized by prototype-importance weights and model-confidence scores. Experiments across four prototypical backbones on four diverse benchmarks spanning fine-grained vision, histopathology, and NLP demonstrate that ProtoTTA improves robustness over standard output entropy minimization while restoring correct semantic focus in prototype activations. We also introduce novel interpretability metrics and a vision-language model (VLM) evaluation framework to explain TTA dynamics, confirming ProtoTTA restores human-aligned semantic focus and correlates reliably with VLM-rated reasoning quality. Code is available at: https://github.com/DeepRCL/ProtoTTA.
Authors: Tuan Nguyen, Ting He
Abstract: We consider the problem of minimizing the convergence time for decentralized federated learning (DFL) in wireless networks under broadcast communications, with focus on mixing matrix design. The mixing matrix is a critical hyperparameter for DFL that simultaneously controls the convergence rate across iterations and the communication demand per iteration, both strongly influencing the convergence time. Although the problem has been studied previously, existing solutions are mostly designed for decentralized parallel stochastic gradient descent (D-PSGD), which requires the mixing matrix to be symmetric and doubly stochastic. These constraints confine the activated communication graph to undirected (i.e., bidirected) graphs, which limits design flexibility. In contrast, we consider mixing matrix design for stochastic gradient push (SGP), which allows asymmetric mixing matrices and hence directed communication graphs. By analyzing how the convergence rate of SGP depends on the mixing matrices, we extract an objective function that explicitly depends on graph-theoretic parameters of the activated communication graph, based on which we develop an efficient design algorithm with performance guarantees. Our evaluations based on real data show that the proposed solution can notably reduce the convergence time compared to the state of the art without compromising the quality of the trained model.
Authors: Anthony Nouy, Agust\'in Somacal
Abstract: We consider the problem of approximating a function by an element of a nonlinear manifold which admits a differentiable parametrization, typical examples being neural networks with differentiable activation functions or tensor networks. Natural gradient descent (NGD) for the optimization of a loss function can be seen as a preconditioned gradient descent where updates in the parameter space are driven by a functional perspective. In a spirit similar to Newton's method, a NGD step uses, instead of the Hessian, the Gram matrix of the generating system of the tangent space to the approximation manifold at the current iterate, with respect to a suitable metric. This corresponds to a locally optimal update in function space, following a projected gradient onto the tangent space to the manifold. Still, both gradient and natural gradient descent methods get stuck in local minima. Furthermore, when the model class is a nonlinear manifold or the loss function is not ideally conditioned (e.g., the KL-divergence for density estimation, or a norm of the residual of a partial differential equation in physics informed learning), even the natural gradient might yield non-optimal directions at each step. This work introduces a natural version of classical inertial dynamic methods like Heavy-Ball or Nesterov and show how it can improve the learning process when working with nonlinear model classes.
Authors: Oriel Savir, Zhenghan Fang, Jeremias Sulam
Abstract: Proximal operators are fundamental across many applications in signal processing and machine learning, including solving ill-posed inverse problems. Recent work has introduced Learned Proximal Networks (LPNs), providing parametric functions that compute exact proximals for data-driven and potentially non-convex regularizers. However, in many settings it is important to include additional structure to these regularizers--and their corresponding proximals--such as shift and scale equivariance. In this work, we show how to obtain learned functions parametrized by neural networks that provably compute exact proximal operators while being equivariant to shifts and scaling, which we dub Affine-Equivariant Learned Proximal Networks (AE-LPNs). We demonstrate our results on synthetic, constructive examples, and then on real data via denoising in out-of-distribution settings. Our equivariant learned proximals enhance robustness to noise distributions and affine shifts far beyond training distributions, improving the practical utility of learned proximal operators
Authors: Jayadev Billa
Abstract: Steering vectors work for some concepts and layers but fail for others, and practitioners have no way to predict which setting applies before running an intervention. We introduce the Linear Accessibility Profile (LAP), a per-layer diagnostic that repurposes the logit lens as a predictor of steering vector effectiveness. The key measure, $A_{\mathrm{lin}}$, applies the model's unembedding matrix to intermediate hidden states, requiring no training. Across 24 controlled binary concept families on five models (Pythia-2.8B to Llama-8B), peak $A_{\mathrm{lin}}$ predicts steering effectiveness at $\rho = +0.86$ to $+0.91$ and layer selection at $\rho = +0.63$ to $+0.92$. A three-regime framework explains when difference-of-means steering works, when nonlinear methods are needed, and when no method can work. An entity-steering demo confirms the prediction end-to-end: steering at the LAP-recommended layer redirects completions on Gemma-2-2B and OLMo-2-1B-Instruct, while the middle layer (the standard heuristic) has no effect on either model.
Authors: Alexander Peysakhovich, William Berman
Abstract: Consider an auto-regressive model that produces outputs x (e.g., answers to questions, molecules) each of which can be summarized by an attribute vector y (e.g., helpfulness vs. harmlessness, or bio-availability vs. lipophilicity). An arbitrary reward function r(y) encodes tradeoffs between these properties. Typically, tilting the model's sampling distribution to increase this reward is done at training time via reinforcement learning. However, if the reward function changes, re-alignment requires re-training. In this paper, we show that a reward weighted classifier-free guidance (RCFG) can act as a policy improvement operator in this setting, approximating tilting the sampling distribution by the Q function. We apply RCFG to molecular generation, demonstrating that it can optimize novel reward functions at test time. Finally, we show that using RCFG as a teacher and distilling into the base policy to serve as a warm start significantly speeds up convergence for standard RL.
Authors: Ethan Tang, Hasan Davulcu, Jia Zou, Zhongju Zhang
Abstract: Predicting the relative value of any given chess piece in a position remains an open challenge, as a piece's contribution depends on its spatial relationships with every other piece on the board. We demonstrate that incorporating the state of the full chess board via latent position representations derived using a CNN-based autoencoder significantly improves accuracy for MLP-based piece value prediction architectures. Using a dataset of over 12 million piece-value pairs gathered from Grandmaster-level games, with ground-truth labels generated by Stockfish 17, our enhanced piece value predictor significantly outperforms context-independent MLP-based systems, reducing validation mean absolute error by 16% and predicting relative piece value within approximately 0.65 pawns. More generally, our findings suggest that encoding the full problem state as context provides useful inductive bias for predicting the contribution of any individual component.
Authors: Yunbei Zhang, Shuaicheng Niu, Chengyi Cai, Feng Liu, Jihun Hamm
Abstract: Test-Time Adaptation (TTA) for black-box models accessible only via APIs remains a largely unexplored challenge. Existing approaches such as post-hoc output refinement offer limited adaptive capacity, while Zeroth-Order Optimization (ZOO) enables input-space adaptation but faces high query costs and optimization challenges in the unsupervised TTA setting. We introduce BETA (Black-box Efficient Test-time Adaptation), a framework that addresses these limitations by employing a lightweight, local white-box steering model to create a tractable gradient pathway. Through a prediction harmonization technique combined with consistency regularization and prompt learning-oriented filtering, BETA enables stable adaptation with no additional API calls and negligible latency beyond standard inference. On ImageNet-C, BETA achieves a +7.1% accuracy gain on ViT-B/16 and +3.4% on CLIP, surpassing strong white-box and gray-box methods including TENT and TPT. On a commercial API, BETA achieves comparable performance to ZOO at 250x lower cost while maintaining real-time inference speed, establishing it as a practical and efficient solution for real-world black-box TTA.
Authors: Wladimir Silva
Abstract: We present VoodooNet, a non-iterative neural architecture that replaces the stochastic gradient descent (SGD) paradigm with a closed-form analytic solution via Galactic Expansion. By projecting input manifolds into a high-dimensional, high-entropy "Galactic" space ($d \gg 784$), we demonstrate that complex features can be untangled without the thermodynamic cost of backpropagation. Utilizing the Moore-Penrose pseudoinverse to solve for the output layer in a single step, VoodooNet achieves a classification accuracy of \textbf{98.10\% on MNIST} and \textbf{86.63\% on Fashion-MNIST}. Notably, our results on Fashion-MNIST surpass a 10-epoch SGD baseline (84.41\%) while reducing the training time by orders of magnitude. We observe a near-logarithmic scaling law between dimensionality and accuracy, suggesting that performance is a function of "Galactic" volume rather than iterative refinement. This "Magic Hat" approach offers a new frontier for real-time Edge AI, where the traditional training phase is bypassed in favor of instantaneous manifold discovery.
Authors: Taisuke Kobayashi
Abstract: This paper proposes a novel method that incorporates empowerment when reasoning actions in reinforcement learning (RL), thereby achieving the flexibility of exploration-exploitation dilemma (EED). In previous methods, empowerment for promoting exploration has been provided as a bonus term to the task-specific reward function as an intrinsically-motivated RL. However, this approach introduces a delay until the policy that accounts for empowerment is learned, making it difficult to adjust the emphasis on exploration as needed. On the other hand, a trick devised for fine-tuning recent foundation models at reasoning, so-called best-of-N (BoN) sampling, allows for the implicit acquisition of modified policies without explicitly learning them. It is expected that applying this trick to exploration-promoting terms, such as empowerment, will enable more flexible adjustment of EED. Therefore, this paper investigates BoN sampling for empowerment. Furthermore, to adjust the degree of policy modification in a generalizable manner while maintaining computational cost, this paper proposes a novel BoN sampling method extended by Tsalis statistics. Through toy problems, the proposed method's cability to balance EED is verified. In addition, it is demonstrated that the proposed method improves RL performance to solve complex locomotion tasks.
Authors: Tim Launer, Jonas H\"ubotter, Marco Bagatella, Ido Hakimi, Andreas Krause
Abstract: We investigate Functional Majority Voting (FMV), a method based on functional consensus for code generation with Large Language Models, which identifies a representative solution from multiple generations using their runtime execution signatures on test inputs. We find that FMV is an effective test-time inference strategy, substantially boosting performance on LiveCodeBench without a large compute overhead. Furthermore, we extend the utility of functional consensus and apply it as an aggregation strategy for label-free Test-Time Reinforcement Learning. We demonstrate that this increases pass@1 on holdout tasks, but find no evidence of self-improvement beyond the base model's performance ceiling.
Authors: Shimon Pisnoy, Hemanth Chandravamsi, Ziv Chen, Aaron Goldgewert, Gal Shaviner, Boris Shragner, Steven H. Frankel
Abstract: We present PINNACLE, an open-source computational framework for physics-informed neural networks (PINNs) that integrates modern training strategies, multi-GPU acceleration, and hybrid quantum-classical architectures within a unified modular workflow. The framework enables systematic evaluation of PINN performance across benchmark problems including 1D hyperbolic conservation laws, incompressible flows, and electromagnetic wave propagation. It supports a range of architectural and training enhancements, including Fourier feature embeddings, random weight factorization, strict boundary condition enforcement, adaptive loss balancing, curriculum training, and second-order optimization strategies, with extensibility to additional methods. We provide a comprehensive benchmark study quantifying the impact of these methods on convergence, accuracy, and computational cost, and analyze distributed data parallel scaling in terms of runtime and memory efficiency. In addition, we extend the framework to hybrid quantum-classical PINNs and derive a formal estimate for circuit-evaluation complexity under parameter-shift differentiation. Results highlight the sensitivity of PINNs to architectural and training choices, confirm their high computational cost relative to classical solvers, and identify regimes where hybrid quantum models offer improved parameter efficiency. PINNACLE provides a foundation for benchmarking physics-informed learning methods and guiding future developments through quantitative assessment of their trade-offs.
Authors: Xinge Liu, Terry Jingchen Zhang, Bernhard Sch\"olkopf, Zhijing Jin, Kristen Menou
Abstract: The rise of autonomous AI agents suggests that dynamic benchmark environments with built-in feedback on scientifically grounded tasks are needed to evaluate the capabilities of these agents in research work. We introduce Stargazer, a scalable environment for evaluating AI agents on dynamic, iterative physics-grounded model-fitting tasks using inference on radial-velocity (RV) time series data. Stargazer comprises 120 tasks across three difficulty tiers, including 20 real archival cases, covering diverse scenarios ranging from high-SNR single-planet systems to complex multi-planetary configurations requiring involved low-SNR analysis. Our evaluation of eight frontier agents reveals a gap between numerical optimization and adherence to physical constraints: although agents often achieve a good statistical fit, they frequently fail to recover correct physical system parameters, a limitation that persists even when agents are equipped with vanilla skills. Furthermore, increasing test-time compute yields only marginal gains, with excessive token usage often reflecting recursive failure loops rather than meaningful exploration. Stargazer presents an opportunity to train, evaluate, scaffold, and scale strategies on a model-fitting problem of practical research relevance today. Our methodology to design a simulation-driven environment for AI agents presumably generalizes to many other model-fitting problems across scientific domains. Source code and the project website are available at https://github.com/Gudmorning2025/Stargazer and https://gudmorning2025.github.io/Stargazer, respectively.
URLs: https://github.com/Gudmorning2025/Stargazer, https://gudmorning2025.github.io/Stargazer,
Authors: Zehao Wang, Lanjun Wang
Abstract: Graph anomaly detection aims to identify irregular patterns in graph-structured data. Most unsupervised GNN-based methods rely on the homophily assumption that connected nodes share similar attributes. However, real-world graphs often exhibit attribute-level heterophily, where connected nodes have dissimilar attributes. Our analysis of attribute-level heterophily graphs reveals two phenomena indicating that current approaches are not practical for unsupervised graph anomaly detection: 1) attribute similarities between connected nodes show nearly identical distributions across different connected node pair types, and 2) anomalies cause consistent variation trends between the graph with and without anomalous edges in the low- and high-frequency components of the spectral energy distributions, while the mid-part exhibits more erratic variations. Based on these observations, we propose NK-GAD, a neighbor knowledge-enhanced unsupervised graph anomaly detection framework. NK-GAD integrates a joint encoder capturing both similar and dissimilar neighbor features, a neighbor reconstruction module modeling normal distributions, a center aggregation module refining node features, and dual decoders for reconstructing attributes and structures. Experiments on seven datasets show NK-GAD achieves an average 3.29\% AUC improvement.
Authors: Yahya Emara, Mauricio Barba da Costa, Chi-Chih Chang, Cameron Freer, Tim Vieira, Ryan Cotterell, Mohamed S. Abdelfattah
Abstract: Speculative decoding (SD) accelerates language model inference by drafting tokens from a cheap proposal model and verifying them against an expensive target model via rejection sampling. Because rejection truncates the draft block at the first error, throughput degrades when draft and target diverge. Rather than rejecting draft tokens outright, we propose to reweight them. To this end, we introduce sequential Monte Carlo speculative decoding (SMC-SD), which replaces token-level rejection with importance-weighted resampling over a population of draft particles. SMC-SD is a principled approximate inference scheme that trades exactness for additional speed, while preserving theoretical bounds on its per-step approximation error. Because LLM inference is memory bandwidth-bound, the arithmetic needed to draft particles and to score them in parallel comes nearly for free -- SMC-SD uses idle compute to turn verification into a vectorized, fixed-size operation with no rollback. Empirically, SMC-SD achieves 2.36x speed-up over speculative decoding and a 5.2x speed-up over autoregressive decoding, while remaining within 3% of the target model's accuracy on reasoning, instruction-following, and coding benchmarks.
Authors: Prashant Rangarajan, Rajesh P. N. Rao
Abstract: Active inference, a neurally-inspired model for inferring actions based on the free energy principle (FEP), has been proposed as a unifying framework for understanding perception, action, and learning in the brain. Active inference has previously been used to model ecologically important tasks such as navigation and planning, but scaling it to solve complex large-scale problems in real-world environments has remained a challenge. Inspired by the existence of multi-scale hierarchical representations in the brain, we propose a model for planning of actions based on hierarchical active inference. Our approach combines a hierarchical model of the environment with successor representations for efficient planning. We present results demonstrating (1) how lower-level successor representations can be used to learn higher-level abstract states, (2) how planning based on active inference at the lower-level can be used to bootstrap and learn higher-level abstract actions, and (3) how these learned higher-level abstract states and actions can facilitate efficient planning. We illustrate the performance of the approach on several planning and reinforcement learning (RL) problems including a variant of the well-known four rooms task, a key-based navigation task, a partially observable planning problem, the Mountain Car problem, and PointMaze, a family of navigation tasks with continuous state and action spaces. Our results represent, to our knowledge, the first application of learned hierarchical state and action abstractions to active inference in FEP-based theories of brain function.
Authors: Jingyuan Li, Xiaoyi Jiang, Fukang Wen, Wei Liu, Renqian Luo, Yi Zhu, Zuoqiang Shi, Pipi Hu
Abstract: Discrete diffusion models based on continuous-time Markov chains (CTMCs) have shown strong performance on language and discrete data generation, yet existing approaches typically parameterize the reverse rate matrix as a single object -- via concrete scores, clean-data predictions ($x_0$-parameterization), or denoising distributions -- rather than aligning the parameterization with the intrinsic CTMC decomposition into jump timing and jump direction. Since a CTMC is fundamentally a Poisson process fully determined by these two quantities, decomposing along this structure is closer to first principles and naturally leads to our formulation. We propose \textbf{Neural CTMC}, which separately parameterizes the reverse process through an \emph{exit rate} (when to jump) and a \emph{jump distribution} (where to jump) using two dedicated network heads. We show that the evidence lower bound (ELBO) differs from a path-space KL divergence between the true and learned reverse processes by a $\theta$-independent constant, so that the training objective is fully governed by the exit rate and jump distribution we parameterize. Moreover, this KL factorizes into a Poisson KL for timing and a categorical KL for direction. We further show that the tractable conditional surrogate preserves the gradients and minimizers of the corresponding marginal reverse-process objective under standard regularity assumptions. Our theoretical framework also covers masked and GIDD-style noise schedules. Empirically, while the uniform forward process has been explored in prior work, our model, to our best of the knowledge, is the first pure-uniform method to outperform mask-based methods on the OpenWebText dataset.To facilitate reproducibility, we release our pretrained weights at https://huggingface.co/Jiangxy1117/Neural-CTMC.
Authors: Haojie Li, Mengjiao Zhang, Guanfeng Liu, Qiang Hu, Yan Wang, Junwei Du
Abstract: Graph self-supervised learning can reduce the need for labeled graph data and has been widely used in recommendation, social networks, and other web applications. However, existing methods often underuse high-frequency signals and may overfit to specific local patterns, which limits representation quality and generalization. We propose Frequency-Corrupt Based Graph Self-Supervised Learning (FC-GSSL), a method that builds corrupted graphs biased toward high-frequency information by corrupting nodes and edges according to their low-frequency contributions. These corrupted graphs are used as inputs to an autoencoder, while low-frequency and general features are reconstructed as supervision targets, forcing the model to fuse information from multiple frequency bands. We further design multiple sampling strategies and generate diverse corrupted graphs from the intersections and unions of the sampling results. By aligning node representations from these views, the model can discover useful frequency combinations, reduce reliance on specific high-frequency components, and improve robustness. Experiments on 14 datasets across node classification, graph prediction, and transfer learning show that FC-GSSL consistently improves performance and generalization.
Authors: Xiaoyu Yang, En Yu, Wei Duan, Jie Lu
Abstract: Reinforcement Fine-Tuning (RFT) has established itself as a critical paradigm for the alignment of Multi-modal Large Language Models (MLLMs) with complex human values and domain-specific requirements. Nevertheless, current research primarily focuses on mitigating exogenous distribution shifts arising from data-centric factors, the non-stationarity inherent in the endogenous reasoning remains largely unexplored. In this work, a critical vulnerability is revealed within MLLMs: they are highly susceptible to endogenous reasoning drift, across both thinking and perception perspectives. It manifests as unpredictable distribution changes that emerge spontaneously during the autoregressive generation process, independent of external environmental perturbations. To adapt it, we first theoretically define endogenous reasoning drift within the RFT of MLLMs as the multi-modal concept drift. In this context, this paper proposes Counterfactual Preference Optimization ++ (CPO++), a comprehensive and autonomous framework adapted to the multi-modal concept drift. It integrates counterfactual reasoning with domain knowledge to execute controlled perturbations across thinking and perception, employing preference optimization to disentangle spurious correlations. Extensive empirical evaluations across two highly dynamic and safety-critical domains: medical diagnosis and autonomous driving. They demonstrate that the proposed framework achieves superior performance in reasoning coherence, decision-making precision, and inherent robustness against extreme interference. The methodology also exhibits exceptional zero-shot cross-domain generalization, providing a principled foundation for reliable multi-modal reasoning in safety-critical applications.
Authors: Zehao Wang, Lanjun Wang
Abstract: Large Reasoning Models (LRMs) have demonstrated strong capabilities in generating step-by-step reasoning chains alongside final answers, enabling their deployment in high-stakes domains such as healthcare and education. While prior jailbreak attack studies have focused on the safety of final answers, little attention has been given to the safety of the reasoning process. In this work, we identify a novel problem that injects harmful content into the reasoning steps while preserving unchanged answers. This type of attack presents two key challenges: 1) manipulating the input instructions may inadvertently alter the LRM's final answer, and 2) the diversity of input questions makes it difficult to consistently bypass the LRM's safety alignment mechanisms and embed harmful content into its reasoning process. To address these challenges, we propose the Psychology-based Reasoning-targeted Jailbreak Attack (PRJA) Framework, which integrates a Semantic-based Trigger Selection module and a Psychology-based Instruction Generation module. Specifically, the proposed PRJA automatically selects manipulative reasoning triggers via semantic analysis and leverages psychological theories of obedience to authority and moral disengagement to generate adaptive instructions for enhancing the LRM's compliance with harmful content generation. Extensive experiments on five question-answering datasets demonstrate that PRJA achieves an average attack success rate of 83.6\% against several commercial LRMs, including DeepSeek R1, Qwen2.5-Max, and OpenAI o4-mini.
Authors: Ibne Farabi Shihab, Sanjeda Akter, Anuj Sharma
Abstract: Chromatic Correlation Clustering (CCC) extends Correlation Clustering by assigning semantic colors to edges and requiring each cluster to receive a single color label. Unlike standard CC, whose LP relaxation has integrality gap 2 on complete graphs and admits a 2.06-approximation, the analogous LP for CCC has a strict lower bound of 2.11, and the best known LP-rounding algorithm achieves 2.15. We explain this gap by isolating the source of difficulty: cross-edge chromatic interference. Neutral edges, whose color does not match the candidate cluster color, create an irreducible cost absent from standard CC and force any color-independent rounding scheme to pay an additional mismatch penalty. We make four contributions. First, we prove a Global Integrality Gap Decomposition Theorem showing that the gap of any color-independent CCC rounding algorithm equals the standard CC gap plus an irreducible chromatic penalty Delta(L) > 0. Second, we solve the associated min-max problem and derive the staircase formula Delta(L) = ((L-1)/L) Delta_infinity, where Delta_infinity is approximately 0.0734. In particular, the two-color gap is 2.0967, separating CCC from standard CC already at L = 2. Third, we introduce Color-Coupled Correlation Clustering (C4). Adding the valid global constraint sum_c x_uv^c >= L-1 and a correlated interval-packing rounding scheme makes neutral edges behave like classical negative edges, recovering the optimal 2.06 approximation and bypassing the 2.11 lower bound for the uncoupled LP. Fourth, experiments on extremal instances, real multi-relational networks, and fairness benchmarks validate the theory: empirical LP gaps follow the predicted staircase, and C4 matches the unconstrained approximation ratio under fairness constraints.
Authors: Hidetoshi Kawase, Toshihiro Ota
Abstract: In finite-width deep neural networks, the empirical kernel $G$ evolves stochastically across layers. We develop a collective kernel effective field theory (EFT) for pre-activation ResNets based on a $G$-only closure hierarchy and diagnose its finite validity window. Exploiting the exact conditional Gaussianity of residual increments, we derive an exact stochastic recursion for $G$. Applying Gaussian approximations systematically yields a continuous-depth ODE system for the mean kernel $K_0$, the kernel covariance $V_4$, and the $1/n$ mean correction $K_{1,\mathrm{EFT}}$, which emerges diagrammatically as a one-loop tadpole correction. Numerically, $K_0$ remains accurate at all depths. However, the $V_4$ equation residual accumulates to an $O(1)$ error at finite time, primarily driven by approximation errors in the $G$-only transport term. Furthermore, $K_{1,\mathrm{EFT}}$ fails due to the breakdown of the source closure, which exhibits a systematic mismatch even at initialization. These findings highlight the limitations of $G$-only state-space reduction and suggest extending the state space to incorporate the sigma-kernel.
Authors: Xiang Xia, Wuyang Zhang, Jiazheng Liu, Cheng Yan, Yanyong Zhang
Abstract: Diffusion language models (DLMs) have emerged as a promising alternative to autoregressive language generation due to their potential for parallel decoding and global refinement of the entire sequence. To unlock this potential, DLM inference must carefully balance generation quality and decoding speed. Recent block-wise DLM decoding methods improve this trade-off by performing diffusion-based decoding sequentially in blocks. However, existing methods typically rely on fixed block schedules or current-step local signals to determine block boundaries, and use conservative confidence-based parallel decoding to avoid conflicts, limiting the quality-speed trade-off. In this paper, we argue that block-wise DLM inference requires more suitable signals for its two core decisions: cross-step signals for determining block boundaries, and token-level conflict signals for parallel decoding. Based on this view, we propose DepCap, a training-free framework for efficient block-wise DLM inference. Specifically, DepCap instantiates the cross-step signal as the influence of the last decoded block and uses it to adaptively determine how far the next block should extend, while identifying a conflict-free subset of tokens for safe parallel decoding within each block, enabling substantial inference acceleration with negligible quality degradation. DepCap is a plug-and-play method applicable to various DLMs, and compatible with existing KV-cache strategies for block-wise DLM. An information-theoretic analysis further suggests that the cumulative last-block influence on a candidate block is approximately additive across tokens, supporting the proposed block-partitioning criterion. Experimental results show that DepCap achieves favorable speed-quality trade-offs across multiple DLM backbones and reasoning and coding benchmarks, with up to 5.63$\times$ speedup without significant performance degradation.
Authors: Peter Vamplew, Cameron Foale
Abstract: This research note identifies a previously overlooked distinction between multi-objective reinforcement learning (MORL), and more conventional single-objective reinforcement learning (RL). It has previously been noted that the optimal policy for an MORL agent with a non-linear utility function is required to be conditioned on both the current environmental state and on some measure of the previously accrued reward. This is generally implemented by concatenating the observed state of the environment with the discounted sum of previous rewards to create an augmented state. While augmented states have been widely-used in the MORL literature, one implication of their use has not previously been reported -- namely that they require the agent to have continued access to the reward signal (or a proxy thereof) after deployment, even if no further learning is required. This note explains why this is the case, and considers the practical repercussions of this requirement.
Authors: Huan Lin, Lianghui Ding
Abstract: Large-scale Unmanned Aerial Vehicle (UAV) failures can split an unmanned aerial vehicle swarm network into disconnected sub-networks, making decentralized recovery both urgent and difficult. Centralized recovery methods depend on global topology information and become communication-heavy after severe fragmentation. Decentralized heuristics and multi-agent reinforcement learning methods are easier to deploy, but their performance often degrades when the swarm scale and damage severity vary. We present Physics-informed Graph Adversarial Imitation Learning algorithm (PhyGAIL) that adopts centralized training with decentralized execution. PhyGAIL builds bounded local interaction graphs from heterogeneous observations, and uses physics-informed graph neural network to encode directional local interactions as gated message passing with explicit attraction and repulsion. This gives the policy a physically grounded coordination bias while keeping local observations scale-invariant. It also uses scenario-adaptive imitation learning to improve training under fragmented topologies and variable-length recovery episodes. Our analysis establishes bounded local graph amplification, bounded interaction dynamics, and controlled variance of the terminal success signal. A policy trained on 20-UAV swarms transfers directly to swarms of up to 500 UAVs without fine-tuning, and achieves better performance across reconnection reliability, recovery speed, motion safety, and runtime efficiency than representative baselines.
Authors: Dongxin Guo, Jikun Wu, Siu Ming Yiu
Abstract: Early-exit neural networks enable adaptive computation by allowing confident predictions to exit at intermediate layers, achieving 2-8$\times$ inference speedup. Despite widespread deployment, their generalization properties lack theoretical understanding -- a gap explicitly identified in recent surveys. This paper establishes a unified PAC-Bayesian framework for adaptive-depth networks. (1) Novel Entropy-Based Bounds: We prove the first generalization bounds depending on exit-depth entropy $H(D)$ and expected depth $\mathbb{E}[D]$ rather than maximum depth $K$, with sample complexity $\mathcal{O}((\mathbb{E}[D] \cdot d + H(D))/\epsilon^2)$. (2) Explicit Constructive Constants: Our analysis yields the leading coefficient $\sqrt{2\ln 2} \approx 1.177$ with complete derivation. (3) Provable Early-Exit Advantages: We establish sufficient conditions under which adaptive-depth networks strictly outperform fixed-depth counterparts. (4) Extension to Approximate Label Independence: We relax the label-independence assumption to $\epsilon$-approximate policies, broadening applicability to learned routing. (5) Comprehensive Validation: Experiments across 6 architectures on 7 benchmarks demonstrate tightness ratios of 1.52-3.87$\times$ (all $p < 0.001$) versus $>$100$\times$ for classical bounds. Bound-guided threshold selection matches validation-tuned performance within 0.1-0.3%.
Authors: Dongxin Guo, Jikun Wu, Siu Ming Yiu
Abstract: Spiking transformers achieve competitive accuracy with conventional transformers while offering $38$-$57\times$ energy efficiency on neuromorphic hardware, yet no theoretical framework guides their design. This paper establishes the first comprehensive expressivity theory for spiking self-attention. We prove that spiking attention with Leaky Integrate-and-Fire neurons is a universal approximator of continuous permutation-equivariant functions, providing explicit spike circuit constructions including a novel lateral inhibition network for softmax normalization with proven $O(1/\sqrt{T})$ convergence. We derive tight spike-count lower bounds via rate-distortion theory: $\varepsilon$-approximation requires $\Omega(L_f^2 nd/\varepsilon^2)$ spikes, with rigorous information-theoretic derivation. Our key insight is input-dependent bounds using measured effective dimensions ($d_{\text{eff}}=47$--$89$ for CIFAR/ImageNet), explaining why $T=4$ timesteps suffice despite worst-case $T \geq 10{,}000$ predictions. We provide concrete design rules with calibrated constants ($C=2.3$, 95\% CI: $[1.9, 2.7]$). Experiments on Spikformer, QKFormer, and SpikingResformer across vision and language benchmarks validate predictions with $R^2=0.97$ ($p<0.001$). Our framework provides the first principled foundation for neuromorphic transformer design.
Authors: Abhishek Sawaika, Durga Pritam Suggisetti, Udaya Parampalli, Rajkumar Buyya
Abstract: Learning with large-scale datasets and information-critical applications, such as in High Energy Physics (HEP), demands highly complex, large-scale models that are both robust and accurate. To tackle this issue and cater to the learning requirements, we envision using a federated learning framework with a quantum-enhanced model. Specifically, we design a hybrid quantum-classical long-shot-term-memory model (QLSTM) for local training at distributed nodes. It combines the representative power of quantum models in understanding complex relationships within the feature space, and an LSTM-based model to learn necessary correlations across data points. Given the computing limitations and unprecedented cost of current stand-alone noisy-intermediate quantum (NISQ) devices, we propose to use a federated learning setup, where the learning load can be distributed to local servers as per design and data availability. We demonstrate the benefits of such a design on a classification task for the Supersymmetry(SUSY) dataset, having 5M rows. Our experiments indicate that the performance of this design is not only better that some of the existing work using variational quantum circuit (VQC) based quantum machine learning (QML) techniques, but is also comparable ($\Delta \sim \pm 1\%$) to that of classical deep-learning benchmarks. An important observation from this study is that the designed framework has $<$300 parameters and only needs 20K data points to give a comparable performance. Which also turns out to be a 100$\times$ improvement than the compared baseline models. This shows an improved learning capability of the proposed framework with minimal data and resource requirements, due to the joint model with an LSTM based architecture and a quantum enhanced VQC.
Authors: Wai Man Si, Mingjie Li, Michael Backes, Yang Zhang
Abstract: Machine learning models are increasingly deployed in real-world applications, but even aligned models such as Mistral and LLaVA still exhibit unsafe behaviors inherited from pre-training. Current alignment methods like SFT and RLHF primarily encourage models to generate preferred responses, but do not explicitly remove the unsafe subnetworks that trigger harmful outputs. In this work, we introduce a resource-efficient pruning framework that directly identifies and removes parameters associated with unsafe behaviors while preserving model utility. Our method employs a gradient-free attribution mechanism, requiring only modest GPU resources, and generalizes across architectures and quantized variants. Empirical evaluations on ML models show substantial reductions in unsafe generations and improved robustness against jailbreak attacks, with minimal utility loss. From the perspective of the Lottery Ticket Hypothesis, our results suggest that ML models contain "unsafe tickets" responsible for harmful behaviors, and pruning reveals "safety tickets" that maintain performance while aligning outputs. This provides a lightweight, post-hoc alignment strategy suitable for deployment in resource-constrained settings.
Authors: Oluwaleke Yusuf, Shaira Tabassum
Abstract: Traffic simulations, essential for planning urban transit infrastructure interventions, require vehicle-category-specific origin-destination (OD) data. Existing data sources are imperfect: sparse tollbooth sensors provide accurate vehicle counts by category, while extensive mobility data from cellular network activity captures aggregated crowd movement, but lack modal disaggregation and have systematic biases. This study develops a machine learning framework to correct and disaggregate cellular network data using sparse tollbooth counts as ground truth. The model uses temporal and spatial features to learn the complex relationship between aggregated mobility data and vehicular data. The framework infers destinations from transit routes and implements routing logic to distribute corrected flows between OD pairs. This approach is applied to a bus depot expansion in Trondheim, Norway, generating hourly OD matrices by vehicle length category. The results show how limited but accurate sensor measurements can correct extensive but aggregated mobility data to produce grounded estimates of background vehicular traffic flows. These macro-scale estimates can be refined for micro-scale analysis at desired locations. The framework provides a generalisable approach for generating origin-destination data from cellular network data. This enables downstream tasks, like detailed traffic simulations for infrastructure planning in data-scarce contexts, supporting urban planners in making informed decisions.
Authors: Oluwaleke Yusuf, M. Tsaqif Wismadi, Adil Rasheed
Abstract: Urban bike-sharing systems require strategic station expansion to meet growing demand. Traditional allocation approaches rely on explicit demand modelling that may not capture the urban characteristics distinguishing successful stations. This study addresses the need to exploit patterns from existing stations to inform expansion decisions, particularly in data-constrained environments. We present a data-driven framework leveraging existing stations deemed desirable by operational metrics. A hybrid denoising autoencoder (HDAE) learns compressed latent representations from multi-source grid-level features (socio-demographic, built environment, and transport network), with a supervised classification head regularising the embedding space structure. Expansion candidates are selected via greedy allocation with spatial constraints based on latent-space similarity to existing stations. Evaluation on Trondheim's bike-sharing network demonstrates that HDAE embeddings yield more spatially coherent clusters and allocation patterns than raw features. Sensitivity analyses across similarity methods and distance metrics confirm robustness. A consensus-based procedure across multiple parametrisations distils 32 high-confidence extension zones where all parametrisations agree. The results demonstrate how representation learning captures complex patterns that raw features miss, enabling evidence-based expansion planning without explicit demand modelling. The consensus procedure strengthens recommendations by requiring agreement across parametrisations, while framework configurability allows planners to incorporate operational knowledge. The methodology generalises to any location-allocation problem where existing desirable instances inform the selection of new candidates.
Authors: David Berghaus
Abstract: We introduce EVIL (\textbf{EV}olving \textbf{I}nterpretable algorithms with \textbf{L}LMs), an approach that uses LLM-guided evolutionary search to discover simple, interpretable algorithms for dynamical systems inference. Rather than training neural networks on large datasets, EVIL evolves pure Python/NumPy programs that perform zero-shot, in-context inference across datasets. We apply EVIL to three distinct tasks: next-event prediction in temporal point processes, rate matrix estimation for Markov jump processes, and time series imputation. In each case, a single evolved algorithm generalizes across all evaluation datasets without per-dataset training (analogous to an amortized inference model). To the best of our knowledge, this is the first work to show that LLM-guided program evolution can discover a single compact inference function for these dynamical-systems problems. Across the three domains, the discovered algorithms are often competitive with, and even outperform, state-of-the-art deep learning models while being orders of magnitudes faster, and remaining fully interpretable.
Authors: Miaoxuan Zhu, Yi Yu, Yuyang Li, Wei Li, Guangcan Liu
Abstract: The quantification of uncertainty in prediction models is crucial for reliable decision-making, yet remains a significant challenge. Interval time series forecasting offers a principled solution to this problem by providing prediction intervals (PIs), which indicates the probability that the true value falls within the predicted range. We consider a recently established point forecasts (PFs) method termed Learning-Based Convolution Nuclear Norm Minimization (LbCNNM), which directly generates multi-step ahead forecasts by leveraging the convolutional low-rankness property derived from training data. While theoretically complete and empirically effective, LbCNNM lacks inherent uncertainty estimation capabilities, a limitation shared by many advanced forecasting methods. To resolve the issue, we modify the well-known Quantile Regression (QR) and integrate it into LbCNNM, resulting in a novel interval forecasting method termed LbCNNM with Modified Quantile Regression (LbCNNM-MQR). In addition, we devise interval calibration techniques to further improve the accuracy of PIs. Extensive experiments on over 100,000 real-world time series demonstrate the superior performance of LbCNNM-MQR.
Authors: Chi Liu, Xin Chen, Xu Zhou, Fangbo Tu, Srinivasan Manoharan
Abstract: Large Language Models (LLMs) have achieved remarkable success, underpinning diverse AI applications. However, they often suffer from performance degradation due to factors such as catastrophic forgetting during Supervised Fine-Tuning (SFT), quantization, and pruning. In this work, we introduce a performance recovery framework based on Self-Distillation Fine-Tuning (SDFT) that effectively restores model capabilities. Complementing this practical contribution, we provide a rigorous theoretical explanation for the underlying recovery mechanism. We posit that an LLM's generative capability fundamentally relies on the high-dimensional manifold constructed by its hidden layers. To investigate this, we employ Centered Kernel Alignment (CKA) to quantify the alignment between student and teacher activation trajectories, leveraging its invariance to orthogonal transformations and scaling. Our experiments demonstrate a strong correlation between performance recovery and manifold alignment, substantiating the claim that self-distillation effectively aligns the student's high-dimensional manifold with the optimal structure represented by the teacher. This study bridges the gap between practical recovery frameworks and geometric representation theory, offering new insights into the internal mechanisms of self-distillation.
Authors: Saloni Garg, Ukant Jadia, Amit Sagtani, Kamal Kant Hiran
Abstract: Automated classification of electrocardiogram (ECG) signals is a useful tool for diagnosing and monitoring cardiovascular diseases. This study compares three traditional machine learning algorithms (Decision Tree Classifier, Random Forest Classifier, and Logistic Regression) and three deep learning models (Simple Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Complex CNN (ECGLens)) for the classification of ECG signals from the PTB-XL dataset, which contains 12-lead recordings from normal patients and patients with various cardiac conditions. The DL models were trained on raw ECG signals, allowing them to automatically extract discriminative features. Data augmentation using the Stationary Wavelet Transform (SWT) was applied to enhance model performance, increase the diversity of training samples, and preserve the essential characteristics of the ECG signals. The models were evaluated using multiple metrics, including accuracy, precision, recall, F1-score, and ROC-AUC. The ECG-Lens model achieved the highest performance, with 80% classification accuracy and a 90% ROC-AUC. These findings demonstrate that deep learning architectures, particularly complex CNNs substantially outperform traditional ML methods on raw 12-lead ECG data, and provide a practical benchmark for selecting automated ECG classification models and identifying directions for condition-specific model development.
Authors: Yangyi Fang, Jiaye Lin, Xiaoliang Fu, Cong Qin, Haolin Shi
Abstract: Reinforcement learning has become a powerful approach for enhancing large language model reasoning, but faces a fundamental dilemma: training on easy problems can cause overfitting and pass@k degradation, while training on hard problems often results in sparse rewards. Recent question augmentation methods address this by prepending partial solutions as hints. However, uniform hint provision may introduce redundant information while missing critical reasoning bottlenecks, and excessive hints can reduce reasoning diversity, causing pass@k degradation. We propose \textbf{PieceHint}, a hint injection framework that strategically identifies and provides critical reasoning steps during training. By scoring the importance of different reasoning steps, selectively allocating hints based on problem difficulty, and progressively withdrawing scaffolding, PieceHint enables models to transition from guided learning to independent reasoning. Experiments on six mathematical reasoning benchmarks show that our 1.5B model achieves comparable average performance to 32B baselines while preserving pass@k diversity across all $k$ values.
Authors: Zhaobo Hu, Vincent Gauthier, Mehdi Naima
Abstract: Spatiotemporal modeling has evolved beyond simple time series analysis to become fundamental in structural time series analysis. While current research extensively employs graph neural networks (GNNs) for spatial feature extraction with notable success, these networks are limited to capturing only pairwise relationships, despite real-world networks containing richer topological relationships. Additionally, GNN-based models face computational challenges that scale with graph complexity, limiting their applicability to large networks. To address these limitations, we present Modern Structure-Aware Simplicial SpatioTemporal neural network (ModernSASST), the first approach to leverage simplicial complex structures for spatiotemporal modeling. Our method employs spatiotemporal random walks on high-dimensional simplicial complexes and integrates parallelizable Temporal Convolutional Networks to capture high-order topological structures while maintaining computational efficiency. Our source code is publicly available on GitHub\footnote{Code is available at: https://github.com/ComplexNetTSP/ST_RUM.
Authors: Zhaobo Hu, Vincent Gauthier, Mehdi Naima
Abstract: Distribution shift severely degrades the performance of deep forecasting models. While this issue is well-studied for individual time series, it remains a significant challenge in the spatio-temporal domain. Effective solutions like instance normalization and its variants can mitigate temporal shifts by standardizing statistics. However, distribution shift on a graph is far more complex, involving not only the drift of individual node series but also heterogeneity across the spatial network where different nodes exhibit distinct statistical properties. To tackle this problem, we propose Reversible Residual Normalization (RRN), a novel framework that performs spatially-aware invertible transformations to address distribution shift in both spatial and temporal dimensions. Our approach integrates graph convolutional operations within invertible residual blocks, enabling adaptive normalization that respects the underlying graph structure while maintaining reversibility. By combining Center Normalization with spectral-constrained graph neural networks, our method captures and normalizes complex Spatio-Temporal relationships in a data-driven manner. The bidirectional nature of our framework allows models to learn in a normalized latent space and recover original distributional properties through inverse transformation, offering a robust and model-agnostic solution for forecasting on dynamic spatio-temporal systems.
Authors: Erchi Wang, Pengrun Huang, Eli Chien, Om Thakkar, Kamalika Chaudhuri, Yu-Xiang Wang, Ruihan Wu
Abstract: Differential privacy (DP) has a wide range of applications for protecting data privacy, but designing and verifying DP algorithms requires expert-level reasoning, creating a high barrier for non-expert practitioners. Prior works either rely on specialized verification languages that demand substantial domain expertise or remain semi-automated and require human-in-the-loop guidance. In this work, we investigate whether large language models (LLMs) can automate DP reasoning. We introduce DPrivBench, a benchmark in which each instance asks whether a function or algorithm satisfies a stated DP guarantee under specified assumptions. The benchmark is carefully designed to cover a broad range of DP topics, span diverse difficulty levels, and resist shortcut reasoning through trivial pattern matching. Experiments show that while the strongest models handle textbook mechanisms well, all models struggle with advanced algorithms, revealing substantial gaps in current DP reasoning capabilities. Through further analytic study and failure-mode analysis, we identify several promising directions for improving automated DP reasoning. Our benchmark provides a solid foundation for developing and evaluating such methods, and complements existing benchmarks for mathematical reasoning.
Authors: Jeremy Qin, Maksym Andriushchenko
Abstract: Forecasting has become a natural benchmark for reasoning under uncertainty. Yet existing evaluations of large language models remain limited to judgmental tasks in simple formats, such as binary or multiple-choice questions. In practice, however, forecasting spans a far broader scope. Across domains such as economics, public health, and social demographics, decisions hinge on numerical estimates over continuous quantities, a capability that current benchmarks do not capture. Evaluating such estimates requires a format that makes uncertainty explicit and testable. We propose prediction intervals as a natural and rigorous interface for this purpose. They demand scale awareness, internal consistency across confidence levels, and calibration over a continuum of outcomes, making them a more suitable evaluation format than point estimates for numerical forecasting. To assess this capability, we introduce a new benchmark QuantSightBench, and evaluate frontier models under multiple settings, assessing both empirical coverage and interval sharpness. Our results show that none of the 11 evaluated frontier and open-weight models achieves the 90\% coverage target, with the top performers Gemini 3.1 Pro (79.1\%), Grok 4 (76.4\%), and GPT-5.4 (75.3\%) all falling at least 10 percentage points short. Calibration degrades sharply at extreme magnitudes, revealing systematic overconfidence across all evaluated models.
Authors: Khang Le, Joaqu\'in Torres-Sospedra, Philipp M\"uller
Abstract: Fixed Radius Near Neighbor (FRNN) search is an alternative to the widely used k Nearest Neighbors (kNN) search. Unlike kNN, FRNN determines a label or an estimate for a test sample based on all training samples within a predefined distance. While this approach is beneficial in certain scenarios, assuming a fixed maximum distance for all training samples can decrease the accuracy of the FRNN. Therefore, in this paper we propose the Adaptive Radius Near Neighbor (ARNN) and the Weighted ARNN (WARNN), which employ adaptive distances and in latter case weights. All three methods are compared to kNN and twelve of its variants for a regression problem, namely WiFi fingerprinting indoor positioning, using 22 different datasets to provide a comprehensive analysis. While the performances of the tested FRNN and ARNN versions were amongst the worse, three of the four best methods in the test were WARNN versions, indicating that using weights together with adaptive distances achieves performance comparable or even better than kNN variants.
Authors: Tristan Kirscher (ICube), Alexandra Ertl (DKFZ), Klaus Maier-Hein (DKFZ), Xavier Coubez (ICANS), Philippe Meyer (ICube), Sylvain Faisan (ICube)
Abstract: Pancreatic ductal adenocarcinoma (PDAC) segmentation on contrast-enhanced CT is inherently ambiguous: inter-rater disagreement among experts reflects genuine uncertainty rather than annotation noise. Standard deep learning approaches assume a single ground truth, producing probabilistic outputs that can be poorly calibrated and difficult to interpret under such ambiguity. We present TwinTrack, a framework that addresses this gap through post-hoc calibration of ensemble segmentation probabilities to the empirical mean human response (MHR) -the fraction of expert annotators labeling a voxel as tumor. Calibrated probabilities are thus directly interpretable as the expected proportion of annotators assigning the tumor label, explicitly modeling inter-rater disagreement. The proposed post-hoc calibration procedure is simple and requires only a small multi-rater calibration set. It consistently improves calibration metrics over standard approaches when evaluated on the MICCAI 2025 CURVAS-PDACVI multi-rater benchmark.
Authors: Yaohong Yang, Sammie Katt, Samuel Kaski
Abstract: Multi-objective Bayesian optimization (MOBO) provides a principled framework for optimizing expensive black-box functions with multiple objectives. However, existing MOBO methods often struggle with coverage, scalability with respect to the number of objectives, and integrating constraints and preferences. In this work, we propose \textit{STAGE-BO, Sequential Targeting Adaptive Gap-Filling $\varepsilon$-Constraint Bayesian Optimization}, that explicitly targets under-explored regions of the Pareto front. By analyzing the coverage of the approximate Pareto front, our method identifies the largest geometric gaps. These gaps are then used as constraints, which transforms the problem into a sequence of inequality-constrained subproblems, efficiently solved via constrained expected improvement acquisition. Our approach provides a uniform Pareto coverage without hypervolume computation and naturally applies to constrained and preference-based settings. Experiments on synthetic and real-world benchmarks demonstrate superior coverage and competitive hypervolume performance against state-of-the-art baselines.
Authors: Jean-Baptiste Escudi\'e, Benjamin Barnes, Stefan Meisegeier, Klaus Kraywinkel, Fabian Prasser, Nils K\"orber
Abstract: There is no consensus in the field of synthetic data on concise metrics for quality evaluations or benchmarks on large health datasets, such as historical epidemiological data. This study presents an evaluation of seven recent models from major machine learning families. The models were evaluated using four different datasets, each with a distinct scale. To ensure a fair comparison, we systematically tuned the hyperparameters of each model for each dataset. We propose a methodology for evaluating the fidelity of synthesized joint distributions, aligning metrics with visualization on a single plot. This method is applicable to any dataset and is complemented by a domain-specific analysis of the German Cancer Registries' epidemiological dataset. The analysis reveals the challenges models face in strictly adhering to the medical domain. We hope this approach will serve as a foundational framework for guiding the selection of synthesizers and remain accessible to all stakeholders involved in releasing synthetic datasets.
Authors: Marcin Hoffmann, Pawe{\l} Kryszkiewicz
Abstract: M-MIMO is one of the crucial technologies for increasing spectral and energy efficiency of wireless networks. Most of the current works assume that M-MIMO arrays are equipped with a linear front end. However, ongoing efforts to make wireless networks more energy-efficient push the hardware to the limits, where its nonlinear behavior appears. This is especially a common problem for the multicarrier systems, e.g., OFDM used in 4G, 5G, and possibly also in 6G, which is characterized by a high Peak-to-Average Power Ratio. While the impact of a nonlinear Power Amplifier (PA) on an OFDM signal is well characterized, it is a relatively new topic for the M-MIMO OFDM systems. Most of the recent works either neglect nonlinear effects or utilize simplified models proper for Rayleigh or LoS radio channel models. In this paper, we first theoretically characterize the nonlinear distortion in the M-MIMO system under commonly used radio channel models. Then, utilizing 3D-Ray Tracing (3D-RT) software, we demonstrate that these models are not very accurate. Instead, we propose two models: a statistical one and an ML-based one using 3D-RT results. The proposed statistical model utilizes the Generalized Extreme Value (GEV) distribution to model Signal to Distortion Ratio (SDR) for victim users, receiving nonlinear distortion, e.g., as interference from neighboring cells. The proposed ML model aims to predict SDR for a scheduled user (receiving nonlinear distortion along with the desired signal), based on the spatial characteristics of the radio channel and the operation point of each PA feeding at the M-MIMO antenna array. The predicted SDR can then be used to perform PA-aware per-user power allocation. The results show about 12% median gain in user throughput achieved by the proposed ML-based power allocation scheme over the state-of-the-art, fixed operating point scheme.
Authors: Jie Yuan, Lei Wang, Yanhao Wang, Yimin Liu
Abstract: This paper introduces a robust discrimination method for distinguishing real ship targets from corner-reflector-array jamming with frequency-agile radar. The key idea is to exploit the multidimensional micro-motion signatures that separate rigid ships from non-rigid decoys. From Range-Velocity maps we derive two new hand-crafted descriptors-mean weighted residual (MWR) and complementary contrast factor (CCF) and fuse them with deep features learned by a lightweight CNN. An XGBoost classifier then gives the final decision. Extensive simulations show that the hybrid feature set consistently outperforms state-of-the-art alternatives, confirming the superiority of the proposed approach.
Authors: Guransh Singh
Abstract: Adapting pre-trained vision-language models (VLMs) for robotic control requires injecting high-magnitude continuous gradients from a flow-matching action expert into a backbone trained exclusively with cross-entropy. This cross-modal gradient asymmetry - the spectral dimensionality mismatch between low-rank MSE regression gradients and the high-dimensional semantic manifold sculpted by CE pre-training, causes rapid, severe erosion of the VLM's visual-question-answering (VQA) capability. Industry-standard defences either sever the gradient pathway entirely via stop gradient, discarding the rich continuous supervision, or restrict parameter capacity through low-rank adapters (LoRA) that constrain the rank of updates but not their direction, and thus still overwrite the pre-trained manifold. We introduce AEGIS (Anchor-Enforced Gradient Isolation System): a buffer-free, layer-wise orthogonal gradient projection framework that enables direct continuous MSE learning while preserving the pre-trained VQA manifold - without any co-training data or replay buffer. AEGIS pre-computes a static Gaussian reference anchor from masked VQA forward passes across all transformer layers, then at each training step constructs a Wasserstein-2 transport penalty that generates an anchor restoration gradient. A sequential dual-backward decomposes the task and anchor gradients; for each transformer layer, AEGIS applies a single Gram-Schmidt orthogonal projection that bends the task gradient away from the destructive direction while preserving its constructive content. The projection sheds less than 1% of gradient energy on average, yet eliminates the cumulative activation drift that drives severe forgetting.
Authors: Stefano Colamonaco, David Debot, Pietro Barbiero, Giuseppe Marra
Abstract: Concept Bottleneck Models (CBMs) aim to improve interpretability in Deep Learning by structuring predictions through human-understandable concepts, but they provide no way to verify whether learned concepts align with the human's intended meaning, hurting interpretability. We introduce Prototype-Grounded Concept Models (PGCMs), which ground concepts in learned visual prototypes: image parts that serve as explicit evidence for the concepts. This grounding enables direct inspection of concept semantics and supports targeted human intervention at the prototype level to correct misalignments. Empirically, PGCMs match the predictive performance of state-of-the-art CBMs while substantially improving transparency, interpretability, and intervenability.
Authors: Weijiang Xiong, Robert Fonod, Nikolas Geroliminis
Abstract: Traffic forecasting is a challenging spatio-temporal modeling task and a critical component of urban transportation management. Current studies mainly focus on deterministic predictions, with limited considerations on the uncertainty and stochasticity in traffic dynamics. Therefore, this paper proposes an elegant yet universal approach that transforms existing models into probabilistic predictors by replacing only the final output layer with a novel Gaussian Mixture Model (GMM) layer. The modified model requires no changes to the training pipeline and can be trained using only the Negative Log-Likelihood (NLL) loss, without any auxiliary or regularization terms. Experiments on multiple traffic datasets show that our approach generalizes from classic to modern model architectures while preserving deterministic performance. Furthermore, we propose a systematic evaluation procedure based on cumulative distributions and confidence intervals, and demonstrate that our approach is considerably more accurate and informative than unimodal or deterministic baselines. Finally, a more detailed study on a real-world dense urban traffic network is presented to examine the impact of data quality on uncertainty quantification and to show the robustness of our approach under imperfect data conditions. Code available at https://github.com/Weijiang-Xiong/OpenSkyTraffic
Authors: C\^ome Fiegel, Pierre M\'enard, Tadashi Kozuno, Michal Valko, Vianney Perchet
Abstract: We study the problem of learning in zero-sum matrix games with repeated play and bandit feedback. Specifically, we focus on developing uncoupled algorithms that guarantee, without communication between players, the convergence of the last-iterate to a Nash equilibrium. Although the non-bandit case has been studied extensively, this setting has only been explored recently, with a bound of $\mathcal{O}(T^{-1/8})$ on the exploitability gap. We show that, for uncoupled algorithms, guaranteeing convergence of the policy profiles to a Nash equilibrium is detrimental to the performance, with the best attainable rate being $\Omega(T^{-1/4})$ in contrast to the usual $\Omega(T^{-1/2})$ rate for convergence of the average iterates. We then propose two algorithms that achieve this optimal rate up to constant and logarithmic factors. The first algorithm leverages a straightforward trade-off between exploration and exploitation, while the second employs a regularization technique based on a two-step mirror descent approach.
Authors: Jean Tarbouriech, Matteo Pirotta, Michal Valko, Alessandro Lazaric
Abstract: We study the sample complexity of learning an $\epsilon$-optimal policy in the Stochastic Shortest Path (SSP) problem. We first derive sample complexity bounds when the learner has access to a generative model. We show that there exists a worst-case SSP instance with $S$ states, $A$ actions, minimum cost $c_{\min}$, and maximum expected cost of the optimal policy over all states $B_{\star}$, where any algorithm requires at least $\Omega(SAB_{\star}^3/(c_{\min}\epsilon^2))$ samples to return an $\epsilon$-optimal policy with high probability. Surprisingly, this implies that whenever $c_{\min} = 0$ an SSP problem may not be learnable, thus revealing that learning in SSPs is strictly harder than in the finite-horizon and discounted settings. We complement this lower bound with an algorithm that matches it, up to logarithmic factors, in the general case, and an algorithm that matches it up to logarithmic factors even when $c_{\min} = 0$, but only under the condition that the optimal policy has a bounded hitting time to the goal state.
Authors: Alina Deriyeva, Jesper Dannath, Benjamin Paassen
Abstract: Practice and extensive exercises are essential in programming education. Intelligent tutoring systems (ITSs) are a viable option to provide individualized hints and advice to programming students even when human tutors are not available. However, prior ITS for programming rarely support the Python programming language, mostly focus on introductory programming, and rarely take recent developments in generative models into account. We aim to establish a novel ITS for Python programming that is highly adaptable, serves both as a teaching and research platform, provides interfaces to plug in hint mechanisms (e.g.\ via large language models), and works inside the particularly challenging regulatory environment of Germany, that is, conforming to the European data protection regulation, the European AI act, and ethical framework of the German Research Foundation. In this paper, we present the description of the current state of the ITS along with future development directions, as well as discuss the challenges and opportunities for improving the system.
Authors: Fernando Moro, Vinicius M. A. Souza
Abstract: Multivariate time series classification (MTSC) plays a crucial role in various domains, including biomedical signal analysis and motion monitoring. However, existing approaches, particularly deep learning models, often require high computational resources, making them unsuitable for real-time applications or deployment on low-cost hardware, such as IoT devices and wearable systems. In this paper, we propose the Univariate Channel Fusion (UCF) method to deal with MTSC efficiently. UCF transforms multivariate time series into a univariate representation through simple channel fusion strategies such as the mean, median, or dynamic time warping barycenter. This transformation enables the use of any classifier originally designed for univariate time series, providing a flexible and computationally lightweight alternative to complex models. We evaluate UCF in five case studies covering diverse application domains, including chemical monitoring, brain-computer interfaces, and human activity analysis. The results demonstrate that UCF often outperforms baseline methods and state-of-the-art algorithms tailored for MTSC, while achieving substantial gains in computational efficiency, being particularly effective in problems with high inter-channel correlation.
Authors: Karim K. Ben Hicham, Jan G. Rittig, Martin Grohe, Alexander Mitsos
Abstract: Accurate molecular property prediction is central to drug discovery, catalysis, and process design, yet real-world applications are often limited by small datasets. Molecular foundation models provide a promising direction by learning transferable molecular representations; however, they typically involve task-specific fine-tuning, require machine learning expertise, and often fail to outperform classical baselines. Tabular foundation models (TFMs) offer a fundamentally different paradigm: they perform predictions through in-context learning, enabling inference without task-specific training. Here, we evaluate TFMs in the low- to medium-data regime across both standardized pharmaceutical benchmarks and chemical engineering datasets. We evaluate both frozen molecular foundation model representations, as well as classical descriptors and fingerprints. Across the benchmarks, the approach shows excellent predictive performance while reducing computational cost, compared to fine-tuning, with these advantages also transferring to practical engineering data settings. In particular, combining TFMs with CheMeleon embeddings yields up to 100\% win rates on 30 MoleculeACE tasks, while compact RDKit2d and Mordred descriptors provide strong descriptor-based alternatives. Molecular representation emerges as a key determinant in TFM performance, with molecular foundation model embeddings and 2D descriptor sets both providing substantial gains over classic molecular fingerprints on many tasks. These results suggest that in-context learning with TFMs provides a highly accurate and cost-efficient alternative for property prediction in practical applications.
Authors: Minchul Kang, Changyong Shin, Jinwoo Jeong, Hyunho Lee, Younghun Go, Gyeongmin Kim, Gyeongsik Yang, Chuck Yoo
Abstract: Accurate prediction of training time in distributed deep learning is crucial for resource allocation, cost estimation, and job scheduling. We observe that the floating-point precision setting is a key determinant of training time, leading to training time variations of ~2.4x over its minimum. However, existing studies on distributed training time prediction rely on static model computation graphs that do not capture precision variations, including mixed precision. According to our experiments, training time prediction without considering precision results in significant prediction errors - reaching up to 147.85% in mean absolute percentage error (MAPE). To address this issue, we propose a precision-aware distributed training time predictor that achieves robust accuracy across diverse precision settings, including mixed precision, with 9.8% MAPE.
Authors: Alexandra Dragomir, Ioana Pintilie, Antonio Barbalau, Marius Dragoi, Florin Brad, Cristian Daniel Paduraru, Alexandru Tifrea, Elena Burceanu, Radu Tudor Ionescu
Abstract: Adapter-based methods have become a cost-effective approach to continual learning (CL) for Large Language Models (LLMs), by sequentially learning a low-rank update matrix for each task. To mitigate catastrophic forgetting, state-of-the-art approaches impose constraints on new adapters with respect to the previous ones, by targeting either subspace or coordinate-wise interference. In this paper, we propose JumpLoRA, a novel framework to adaptively induce sparsity in the Low-Rank Adaptation (LoRA) blocks through the use of JumpReLU gating. The method achieves dynamic parameter isolation, which helps prevent task interference. We demonstrate that our method is highly modular and compatible with LoRA-based CL approaches. Specifically, it significantly boosts the performance of IncLoRA and outperforms the leading state-of-the-art CL method, ELLA.
Authors: Andr\'e Saimon S. Sousa, Otto Pires, Frank Acasiete, Oscar M. Granados, Val\'eria Loureiro da Silva, Hugo Saba
Abstract: Data plays a fundamental role in consolidating markets, services, and products in the digital financial ecosystem. However, the use of real data, especially in the financial context, can lead to privacy risks and access restrictions, affecting institutions, research, and modeling processes. Although not all financial datasets present such limitations, this work proposes the use of deep learning techniques for generating synthetic data applied to cryptocurrency price time series. The approach is based on Conditional Generative Adversarial Networks (CGANs), combining an LSTM-type recurrent generator and an MLP discriminator to produce statistically consistent synthetic data. The experiments consider different crypto-assets and demonstrate that the model is capable of reproducing relevant temporal patterns, preserving market trends and dynamics. The generation of synthetic series through GANs is an efficient alternative for simulating financial data, showing potential for applications such as market behavior analysis and anomaly detection, with lower computational cost compared to more complex generative approaches.
Authors: Yide Ran, Jianwen Xie, Minghui Wang, Wenjin Zheng, Denghui Zhang, Chuan Li, Zhaozhuo Xu
Abstract: Data attribution and valuation are critical for understanding data-model synergy for Large Language Models (LLMs), yet existing gradient-based methods suffer from scalability challenges on LLMs. Inspired by human cognition, where decision making relies on a focused readout of relevant memories rather than replaying all pathways, we introduce RISE (Readout Influence Sketching Estimator). Instead of computing and indexing gradients across the entire LLM, RISE focuses on influence hotspots at the output layer, where influence signals concentrate, and the gradient admits a decomposed outer-product form. This enables a dual-channel representation combining a lexical residual channel (RH) and a semantic projected-error channel (GH). Applying CountSketch projections to these channels achieves strong compression while maintaining accurate attribution. Across the OLMo (1B-32B) and Pythia (14M-6.9B) families, RISE reduces index storage by up to 112$\times$ compared to RapidIn and scales to 32B parameters LLM, where gradient-based baselines such as RapidIn and ZO-Inf become memory-infeasible. We evaluate RISE on two paradigms: (1) retrospective attribution, retrieving influential training examples for specific predictions, and (2) prospective valuation, scoring candidate data utility zero-shot. We validate RISE on three tasks: Howdy backdoor data detection, Finance-Medical domain separation, and Brain Rot high-quality data selection. In a closed-loop Brain Rot study, continued pretraining on RISE-selected data yields consistent downstream improvements. Overall, RISE provides a practical and scalable primitive for influence analysis and training-data selection in modern large language models.
Authors: Haoran Zhang, Livia Betti, Konstantin Klemmer, Esther Rolf, David Alvarez-Melis
Abstract: In computer vision and machine learning for geographic data, out-of-domain generalization is a pervasive challenge, arising from uneven global data coverage and distribution shifts across geographic regions. Though models are frequently trained in one region and deployed in another, there is no principled method for determining when this cross-region adaptation will be successful. A well-defined notion of distance between distributions can effectively quantify how different a new target domain is compared to the domains used for model training, which in turn could support model training and deployment decisions. In this paper, we propose a strategy for computing distances between geospatial domains that leverages geographic information with Optimal Transport methods (GeoSpOT). In our experiments, GeoSpOT distances emerge as effective predictors of cross-domain transfer difficulty. We further demonstrate that embeddings from pretrained location encoders provide information comparable to image/text embeddings, despite relying solely on longitude-latitude pairs as input. This allows users to get an approximation of out-of-domain performance for geospatial models, even when the exact downstream task is unknown, or no task-specific data is available. Building on these findings, we show that GeoSpOT distances can preemptively guide data selection and enable predictive tools to analyze regions where a model is likely to underperform.
Authors: Karin Yu, Eleni Chatzi, Georgios Kissas
Abstract: Understanding natural and engineered systems often relies on symbolic formulations, such as differential equations, which provide interpretability and transferability beyond black-box models. We introduce Latent Grammar Flow (LGF), a neuro-symbolic generative framework for discovering ordinary differential equations from data. LGF embeds equations as grammar-based representations into a discrete latent space and forces semantically similar equations to be positioned closer together with a behavioural loss. Then, a discrete flow model guides the sampling process to recursively generate candidate equations that best fit the observed data. Domain knowledge and constraints, such as stability, can be either embedded into the rules or used as conditional predictors.
Authors: Hannah Guan, Soukayna Mouatadid, Paulo Orenstein, Judah Cohen, Haiyu Dong, Zekun Ni, Jeremy Berman, Genevieve Flaspohler, Alex Lu, Jakob Schloer, Joshua Talib, Jonathan A. Weyn, Lester Mackey
Abstract: Decision-makers rely on weather forecasts to plant crops, manage wildfires, allocate water and energy, and prepare for weather extremes. Today, such forecasts enjoy unprecedented accuracy out to two weeks thanks to steady advances in physics-based dynamical models and data-driven artificial intelligence (AI) models. However, model skill drops precipitously at subseasonal timescales (2 - 6 weeks ahead), due to compounding errors and persistent biases. To counter this degradation, we introduce probabilistic bias correction (PBC), a machine learning framework that substantially reduces systematic error by learning to correct historical probabilistic forecasts. When applied to the leading dynamical and AI models from the European Centre for Medium-Range Weather Forecasts (ECMWF), PBC doubles the subseasonal skill of the AI Forecasting System and improves the skill of the operationally-debiased dynamical model for 91% of pressure, 92% of temperature, and 98% of precipitation targets. We designed PBC for operational deployment, and, in ECMWF's 2025 real-time forecasting competition, its global forecasts placed first for all weather variables and lead times, outperforming the dynamical models from six operational forecasting centers, an international dynamical multi-model ensemble, ECMWF's AI Forecasting System, and the forecasting systems of 34 teams worldwide. These probabilistic skill gains translate into more accurate prediction of extreme events and have the potential to improve agricultural planning, energy management, and disaster preparedness in vulnerable communities.
Authors: Songtao Wang, Quang Hieu Pham, Fangcong Yin, Xinpeng Wang, Jocelyn Qiaochu Chen, Greg Durrett, Xi Ye
Abstract: Reinforcement learning with verifiable rewards (RLVR) typically optimizes for outcome rewards without imposing constraints on intermediate reasoning. This leaves training susceptible to reward hacking, where models exploit loopholes (e.g., spurious patterns in training data) in the reward function to achieve high scores without solving the intended task. These reward-hacking behaviors are often implicit, as the intermediate chain-of-thought (CoT) may appear plausible on the surface, limiting the effectiveness of purely text-based monitoring. We propose Gradient Fingerprint (GRIFT), a method for detecting reward hacking using models' internal computations. Given a prompt and a model-generated CoT, GRIFT computes gradients of the CoT conditioned on the prompt and compresses them into a compact representation, which is then used to assess whether the CoT reflects reward hacking behavior. Across verifiable reasoning benchmarks spanning math, code, and logical reasoning, GRIFT substantially outperforms strong baselines, including CoT Monitor and TRACE, achieving over 25% relative improvement in detecting reward hacking behavior. Moreover, integrating GRIFT into the rejection fine-tuning pipeline for reasoning tasks reduces reward hacking and improves performance on the true task objective. Our results highlight a promising direction of leveraging gradient level representations for assessing the quality of CoT reasoning traces. Our code is available at: https://github.com/songtao-x/reward_hack.
Authors: Habibeh Naderi, Behrouz Haji Soleimani, Stan Matwin
Abstract: We propose HILBERT (HIerarchical Long-sequence Balanced Embedding with Reciprocal contrastive Training), a cross-attentive multimodal framework for learning document-level audio-text representations from long, segmented sequences in low-resource data settings. HILBERT leverages frozen pre-trained speech and language encoders to extract segment-level features, which are aggregated via cross-modal attention and self-attentive pooling to form modality-specific document representations and a joint cross-attentive embedding. To align modalities while preserving modality-specific structure under severe audio-text dimensional imbalance, we introduce a reciprocal dual contrastive objective that simultaneously aligns audio-to-joint and text-to-joint representations, rather than directly contrasting audio and text alone. Two auxiliary regularizers further stabilize long-sequence fusion: a Centered Kernel Alignment (CKA) loss that preserves structural consistency between each modality and the joint embedding, and a mutual information balancing loss that prevents dominance of a single modality by equalizing information flow from audio and text into the joint space. For downstream prediction, HILBERT employs a Mixture-of-Experts (MoE) classifier over concatenated audio, text, and joint representations to accommodate heterogeneous label regimes. Extensive evaluation across multiple audio-text backbone combinations demonstrates that HILBERT learns semantically meaningful long-sequence representations and achieves superior performance on highly imbalanced multi-class settings.
Authors: Sarthak Mittal, Leo Gagnon, Guillaume Lajoie
Abstract: Frontier models have demonstrated exceptional capabilities following the integration of task-reward-based reinforcement learning (RL) into their training pipelines, enabling systems to evolve from pure reasoning models into sophisticated agents. However, debate persists regarding whether RL genuinely instills new skills within a base model or merely sharpens its existing distribution to elicit latent capabilities. To address this dichotomy, we present an explicit comparison between distribution sharpening and task-reward-based learning, utilizing RL as a tool to implement both paradigms. Our analysis reveals the inherent limitations of distribution sharpening, demonstrating from first principles how and why the optima can be unfavorable and the approach fundamentally unstable. Furthermore, our experiments using Llama-3.2-3B-Instruct, Qwen2.5-3B-Instruct and Qwen3-4B-Instruct-2507 on math datasets confirm that sharpening yields limited gains, whereas incorporating task-based reward signal can greatly help achieve robust performance improvements and stable learning.
Authors: Aswathi Mundayatt, Jaya Sreevalsan-Nair
Abstract: Existing multi-hazard susceptibility mapping (MHSM) studies often rely on spatially uniform models, treat hazards independently, and provide limited representation of cross-hazard dependence and uncertainty. To address these limitations, this study proposes a deep learning (DL) workflow for joint flood-landslide multi-hazard susceptibility mapping (FL-MHSM) that combines two-level spatial partitioning, probabilistic Early Fusion (EF), a tree-based Late Fusion (LF) baseline, and a soft-gating Mixture of Experts (MoE) model, with MoE serving as final predictive model. The proposed design preserves spatial heterogeneity through zonal partitions and enables data-parallel large-area prediction using overlapping lattice grids. In Kerala, EF remained competitive with LF, improving flood recall from 0.816 to 0.840 and reducing Brier score from 0.092 to 0.086, while MoE provided strongest performance for flood susceptibility, achieving an AUC-ROC of 0.905, recall of 0.930, and F1-score of 0.722. In Nepal, EF similarly improved flood recall from 0.820 to 0.858 and reduced Brier score from 0.057 to 0.049 relative to LF, while MoE outperformed both EF and LF for landslide susceptibility, achieving an AUC-ROC of 0.914, recall of 0.901, and F1-score of 0.559. GeoDetector analysis of MoE outputs further showed that dominant factors varied more across zones in Kerala, where susceptibility was shaped by different combinations of topographic, land-cover, and drainage-related controls, while Nepal showed a more consistent influence of topographic and glacier-related factors across zones. These findings show that EF and LF provide complementary predictive behavior, and that their spatially adaptive integration through MoE yields robust overall predictive performance for FL-MHSM while supporting interpretable characterization of multi-hazard susceptibility in spatially heterogeneous landscapes.
Authors: Shriram Chennakesavalu, Kirill Shmilovich, Hayley Weir, Colin Grambow, John Bradshaw, Patricia Suriana, Chen Cheng, Kangway Chuang
Abstract: Large Language Models (LLMs) have the potential to accelerate small molecule drug design due to their ability to reason about information from diverse sources and formats. However, their practical utility remains unclear due to the lack of benchmarks that reflect real-world scenarios. In this work, we introduce a suite of chemically-grounded tasks spanning molecular property prediction, molecular representation transformations, and molecular design. Importantly, we formulate these tasks as reinforcement learning (RL) environments, enabling a unified approach for evaluation and post-training. Across three model families, we find that frontier models are increasingly proficient at chemical tasks, but that there is significant room for improvement, especially in experimental settings with low data. Critically, we show that RL-based post-training can substantially improve performance. A smaller model post-trained on our environments becomes competitive with state-of-the-art frontier models, despite a significantly weaker base model. This suggests a practical route toward employing LLMs in drug discovery; by combining carefully-designed evaluation tasks with targeted post-training, we can both elucidate and close critical capability gaps.
Authors: Sean Hill, Felix X. -F. Ye
Abstract: Stochastic dynamical systems with slow or metastable behavior evolve, on long time scales, on an unknown low-dimensional manifold in high-dimensional ambient space. Building a reduced simulator from short-burst ambient ensembles is a long-standing problem: local-chart methods like ATLAS suffer from exponential landmark scaling and per-step reprojection, while autoencoder alternatives leave tangent-bundle geometry poorly constrained, and the errors propagate into the learned drift and diffusion. We observe that the ambient covariance~$\Lambda$ already encodes coordinate-invariant tangent-space information, its range spanning the tangent bundle. Using this, we construct a tangent-bundle penalty and an inverse-consistency penalty for a three-stage pipeline (chart learning, latent drift, latent diffusion) that learns a single nonlinear chart and the latent SDE. The penalties induce a function-space metric, the $\rho$-metric, strictly weaker than the Sobolev $H^1$ norm yet achieving the same chart-quality generalization rate up to logarithmic factors. For the drift, we derive an encoder-pullback target via It\^o's formula on the learned encoder and prove a bias decomposition showing the standard decoder-side formula carries systematic error for any imperfect chart. Under a $W^{2,\infty}$ chart-convergence assumption, chart-level error propagates controllably to weak convergence of the ambient dynamics and to convergence of radial mean first-passage times. Experiments on four surfaces embedded in up to $201$ ambient dimensions reduce radial MFPT error by $50$--$70\%$ under rotation dynamics and achieve the lowest inter-well MFPT error on most surface--transition pairs under metastable M\"uller--Brown Langevin dynamics, while reducing end-to-end ambient coefficient errors by up to an order of magnitude relative to an unregularized autoencoder.
Authors: Thanushi Withanage, Elizabeth Redcay, Carol Espy-Wilson
Abstract: Individuals often align their speaking patterns with their interlocutors, a phenomenon linked to engagement and rapport. While well documented in task-oriented dialogues, less is known about entrainment in naturalistic, non-task and virtual settings. In this study, we analyze a large corpus of spontaneous dyadic Zoom conversations to examine how conversational dynamics relate to perceived interaction quality. We extract multimodal features encompassing turn-taking, pauses, facial movements, and acoustic measures such as pitch and intensity. Perceived conversational success was quantified via factor analysis of post-conversation ratings. Results demonstrate that entrainment reliably detected in spontaneous speech and correlates with higher perceived success. These findings identify key interactional markers of conversational quality and highlight opportunities for targeted interventions to foster more effective and engaging communication.
Authors: Z. Cheng, N. Song
Abstract: We report a detailed autoethnographic case study of a single-subject who deliberately constructed and operated a multi-modal prompt-engineering system (System A) designed to externalize cognitive self-regulation onto a large language model (LLM). Within 48 hours of the system's completion, a cascade of observable behavioral changes occurred: voluntary transfer of decision-making authority to the LLM, use of LLM-generated output to deflect external criticism, and a loss of self-initiated reasoning that was independently perceived by two uninformed observers, one of whom subsequently became a co-author of this report. We document the precise architectural mechanism responsible: context contamination, whereby prompt-level isolation instructions co-exist with the very emotional and self-referential material they nominally isolate, rendering the isolation directive structurally ineffective within the attention window. We further identify a metacognitive co-option dynamic, in which intact higher-order reasoning capacity was redirected toward defending the closed loop rather than exiting it. Recovery occurred only after physical interruption of the interaction and a self-initiated pharmacologically-mediated sleep event functioning as an external circuit break. A redesigned system (System B) employing physical rather than logical conversation isolation avoided all analogous failure modes. We derive three contributions: (1) a technically-grounded account of why prompt-layer isolation is architecturally insufficient for context-sensitive multi-modal LLM systems; (2) a phenomenological record of closed-loop collapse with external-witness corroboration; and (3) an ethical distinction between protective system design (preventing unintended loss of user agency) and restrictive system design (preventing intentional boundary-pushing), which require fundamentally different account-ability frameworks.
Authors: Varad Vishwarupe, Ivan Flechais, Nigel Shadbolt, Marina Jirotka
Abstract: Large language models (LLMs) are increasingly integrated into design and development workflows, yet decisions about their use are rarely binary or purely technical. We report findings from a constructivist grounded theory study based on interviews with 33 designers and developers across three large technology organisations. Rather than evaluating LLMs solely by capability, participants reasoned about the role an LLM could occupy within a workflow and how that role would interact with existing structures of responsibility and organisational accountability. When LLMs were framed as tools under clear human control, their use was typically acceptable and could be integrated within existing governance structures. When framed as teammates with shared or ambiguous agency, practitioners expressed hesitation, particularly when responsibility for outcomes could not be clearly justified. At the same time, participants also described productive teammate configurations in which LLMs supported collaborative reasoning while remaining embedded within explicit oversight structures. We identify tool and teammate framings as recurring ways in which designers and developers position LLMs relative to human work and present an analytic rubric describing how role framing shapes decision authority, accountability ownership, oversight strategies, and organisational acceptability. By foregrounding design-time reasoning, this work reframes To LLM or Not to LLM as a sociotechnical positioning problem that emerges during system design rather than during post-deployment evaluation.
Authors: Albert A. Nasybullin, Nursultan Abdullaev, Maksim A. Baranov, Viacheslav V. Koshman, Vitaly A. Mahonin
Abstract: This study presents a methodology for identifying the most informative frequencies and channels in electromyography (EMG) data to evaluate muscle recovery using Decision Tree classifiers. EMG signals, recorded from the vastus lateralis muscle during squat exercises, were analyzed across varying rest intervals to assess optimal recovery periods. By employing single Decision Tree classifiers, the study enhances interpretability, offering insights into feature importance - essential for applications in medical and sports settings where transparency is critical. The experimental protocol utilized a grid search for hyperparameter tuning and cross-validation to address class imbalance, ultimately achieving a reliable classification of rest intervals based on power spectral density features. The results indicate that a limited subset of highly informative features provides sufficient accuracy, suggesting that streamlined, interpretable models are effective for the evaluation of muscle recovery. This approach can guide future research in developing compact, robust models adapted to EMG-based diagnostics.
Authors: Bardia Nadimi, Hao Zheng
Abstract: Understanding communication behavior in modern system-on-chip (SoC) designs is critical for functional verification, performance analysis, and post-silicon debugging. Communication traces capture message exchanges among system components and provide valuable insights into system behavior. However, deriving concise communication specifications from such traces remains challenging due to interleaved instances of communication flows, and ambiguous causal relationships among messages. Existing mining approaches often struggle with scalability and ambiguity when traces contain complex interleaving of message patterns across multiple components. These conditions often lead to an explosion in the number of candidate flows and inaccurate extraction of communication behaviors. This paper presents AutoFlows++, a design-architecture-guided hierarchical framework for mining message flows from communication traces of complex SoC designs. AutoFlows++ operates in two stages: local mining followed by global mining. In the local mining stage, simple communication patterns are extracted from traces observed at individual communication interfaces between components. In the global mining stage, these local patterns are composed to identify higher-level message flows that characterize communication behavior across multiple components. Experimental results on both synthetic traces and traces generated from SoC models in GEM5 demonstrate that AutoFlows++ significantly improves flow extraction accuracy compared with prior approaches, highlighting its effectiveness for practical SoC validation tasks.
Authors: Nam Trinh
Abstract: Motivated behaviour relies on the brain's capacity to evaluate effort and reward. Dysregulation within these processes contributes to a spectrum of conditions, from hyperactivity in attention-deficit/hyperactivity disorder (ADHD) to diminished goal-directed behaviour in apathy. This thesis investigates the neural mechanisms underlying ADHD using electroencephalography (EEG) and examines individual differences in effort and reward sensitivity using neuroimaging, applying machine learning approaches through three main studies. In Study 1, task-based and resting-state EEG were employed with machine learning models to classify adult individuals with ADHD and healthy controls. Machine learning classifiers trained on task-based EEG during a stop signal task outperformed those trained on resting-state EEG, with the strongest predictive features arising from gamma-band spectral power over fronto-central and parietal regions. In Study 2, diffusion MRI and whole-brain permutation-based analyses identified associations between white matter integrity and computationally modelled parameters reflecting effort and reward sensitivity, with SMA-connected tracts emerging as a central hub. In Study 3, grey matter volumes from structural T1-weighted MRI were used to examine correlates of effort sensitivity, reward sensitivity, and subclinical apathy, with machine learning confirming robust decoding of reward sensitivity and apathy levels. Across studies, fronto-parietal circuits emerged as central to effort valuation and reward processing. These findings may serve as neural biomarkers for improving diagnostic accuracy in ADHD and motivational impairments, and for guiding personalised neurotechnological interventions.
Authors: Prakul Sunil Hiremath
Abstract: Edge intelligence is constrained by the energy and latency costs of shuttling data through electronic memory hierarchies. Optical systems offer a fundamentally different computational regime: once an input wavefront is launched into a structured medium, propagation, diffraction, and interference jointly enact a linear transformation whose cost is determined by wave physics rather than by clocked arithmetic. This paper develops a rigorous systems-level treatment of that regime and introduces a hybrid diffractive holographic architecture for image classification. The proposed model couples a Diffractive Optical Neural Network (DONN) with a Holographic Interference-Based Learning (HIBL) operator a formal map from digitally optimized phase distributions to physically realizable, fabrication-compatible interference patterns embeddable in passive optical elements. We express the full inference pipeline as a composition of encoding, phase modulation, free-space propagation, and intensity measurement operators, making explicit which quantities are learned, which are fixed by design, and where nonlinearity enters through photodetection. This operator-theoretic view resolves a persistent gap in the optical-ML literature between learning a transformation and physically realizing it. In physics-informed simulation on MNIST, a three-layer system with approximately 25,000 phase elements achieves 91.2% test accuracy with propagation-limited nanosecond-scale latency. The primary contribution is not a performance claim but a precise computational framework: learned representations can be physically embedded into structured optical media so that inference is executed by wavefront transformation through a passive, fabricated object rather than by sequential electronic multiply accumulate operations.
Authors: Xinxin Fan, Wenxiong Chen, Quanliang Jing, Chi Lin, Shaoye Luo, Wenbo Song, Yunfeng Lu
Abstract: Graph adversarial attacks are usually produced from the two perspectives of topology/structure and node feature, both of them represent the paramount characteristics learned by today's deep learning models. Although some defense countermeasures are proposed at present, they fails to disclose the intrinsic reasons why these two aspects necessitate and how they are adequately fused to co-learn the graph representation. Towards this question, we in this paper propose an adversarial defense approach through locating the graph's critical state of adversarial resilience, resorting to the equilibrium-point theory in the discipline of complex dynamic system (CDS). In brief, our work has three novelties: i) Adversarial-Attack Modeling, i.e. map a graph regime into CDS, and use the oscillation of dynamic system to model the behavior of adversarial perturbation; ii) 2D Topology-Feature-Entangled Function Design for Perturbed Graph, i.e. project graph topology and node feature as two characteristic spaces, and define two-dimensional entangled perturbation functions to represent the dynamic variance under adversarial attacks; and iii) Location of Critical State of Adversarial Resilience, i.e. utilize the equilibrium-point theory to locate the graph's critical state of attack resilience resorting to the perturbation-reflected 2D function. Finally, multi-facet experiments on five commonly-used realistic datasets validate the effectiveness of our proposed approach, and the results show our approach can significantly outperform the state-of-the-art baselines under four representative graph adversarial attacks.
Authors: Venkata Abhinandan Kancharla
Abstract: Large language models (LLMs) achieve strong performance across many natural language processing tasks, yet their decision processes remain difficult to interpret. This lack of transparency creates challenges for trust, debugging, and deployment in real-world systems. This paper presents an applied comparative study of three explainability techniques: Integrated Gradients, Attention Rollout, and SHAP, on a fine-tuned DistilBERT model for SST-2 sentiment classification. Rather than proposing new methods, the focus is on evaluating the practical behavior of existing approaches under a consistent and reproducible setup. The results show that gradient-based attribution provides more stable and intuitive explanations, while attention-based methods are computationally efficient but less aligned with prediction-relevant features. Model-agnostic approaches offer flexibility but introduce higher computational cost and variability. This work highlights key trade-offs between explainability methods and emphasizes their role as diagnostic tools rather than definitive explanations. The findings provide practical insights for researchers and engineers working with transformer-based NLP systems. This is a preprint and has not undergone peer review.
Authors: Kieran A. Murphy
Abstract: We propose InfoChess, a symmetric adversarial game that elevates competitive information acquisition to the primary objective. There is no piece capture, removing material incentives that would otherwise confound the role of information. Instead, pieces are used to alter visibility. Players are scored on their probabilistic inference of the opponent's king location over the duration of the game. To explore the space of strategies for playing InfoChess, we introduce a hierarchy of heuristic agents defined by increasing levels of opponent modeling, and train a reinforcement learning agent that outperforms these baselines. Leveraging the discrete structure of the game, we analyze gameplay through natural information-theoretic characterizations that include belief entropy, oracle cross entropy, and predictive log score under the action-induced observation channel. These measures disentangle epistemic uncertainty, calibration mismatch, and uncertainty induced by adversarial movement. The design of InfoChess renders it a testbed for studying multi-agent inference under partial observability. We release code for the environment and agents, and a public interface to encourage further study.
Authors: Afia Farjana, Zaiyu Cheng, Antonio Mastropaolo
Abstract: Software documentation is essential for program comprehension, developer onboarding, code review, and long-term maintenance. Yet producing quality documentation manually is time-consuming and frequently yields incomplete or inconsistent results. Large language models (LLMs) offer a promising solution by automatically generating natural language descriptions from source code, helping developers understand code more efficiently, facilitating maintenance, and supporting downstream activities such as defect localization and commit message generation. However, the effectiveness of LLMs in documentation tasks critically depends on how they are prompted. Properly structured instructions can substantially improve model performance, making prompt engineering-the design of input prompts to guide model behavior-a foundational technique in LLM-based software engineering. Approaches such as few-shot prompting, chain-of-thought reasoning, retrieval-augmented generation, and zero-shot learning show promise for code summarization, yet current research remains fragmented. There is limited understanding of which prompting strategies work best, for which models, and under what conditions. Moreover, evaluation practices vary widely, with most studies relying on overlap-based metrics that may not capture semantic quality. This systematic literature review consolidates existing evidence, categorizes prompting paradigms, examines their effectiveness, and identifies gaps to guide future research and practical adoption.
Authors: Panos Tsimpos, Daniel Sharp, Youssef Marzouk
Abstract: We study dynamic measure transport for generative modelling in the setting of a stochastic process $X_\bullet$ whose marginals interpolate between a source distribution $P_0$ and a target distribution $P_1$ while remaining independent, i.e., when $(X_0,X_1)\sim P_0\otimes P_1$. Conditional expectations of this process $X_\bullet$ define an ODE whose flow map transports from $P_0$ to $P_1$. We discuss when such a process induces a \emph{straight-line flow}, namely one whose pointwise acceleration vanishes and is therefore exactly integrable by any first-order method. We first develop multiple characterizations of straightness in terms of PDEs involving the conditional statistics of the process. Then, we prove that straightness under endpoint independence exhibits a sharp dichotomy. On one hand, we construct explicit, computable straight-line processes for arbitrary Gaussian endpoints. On the other hand, we show straight-line processes do not exist for targets with sufficiently well-separated modes. We demonstrate this through a sequence of increasingly general impossibility theorems that uncover a fundamental relationship between the sample-path behavior of a process with independent endpoints and the space-time geometry of this process' flow map. Taken together, these results provide a structural theory of when straight generative flows can, and cannot, exist.
Authors: Zhitong Gao, Parham Rezaei, Ali Cy, Mingqiao Ye, Nata\v{s}a Jovanovi\'c, Jesse Allardice, Afshin Dehghan, Amir Zamir, Roman Bachmann, O\u{g}uzhan Fatih Kar
Abstract: Tokenization is a key component of autoregressive (AR) generative models, converting raw data into more manageable units for modeling. Commonly, tokens describe local information, such as regions of pixels in images or word pieces in text, and AR generation predicts these tokens in a fixed order. A worthwhile question is whether token structures affect the ability to steer the generation through test-time search, where multiple candidate generations are explored and evaluated by a verifier. Using image generation as our testbed, we hypothesize that recent 1D ordered tokenizers with coarse-to-fine structure can be more amenable to search than classical 2D grid structures. This is rooted in the fact that the intermediate states in coarse-to-fine sequences carry semantic meaning that verifiers can reliably evaluate, enabling effective steering during generation. Through controlled experiments, we find that AR models trained on coarse-to-fine ordered tokens exhibit improved test-time scaling behavior compared to grid-based counterparts. Moreover, we demonstrate that, thanks to the ordered structure, pure test-time search over token sequences (i.e., without training an AR model) can perform training-free text-to-image generation when guided by an image-text verifier. Beyond this, we systematically study how classical search algorithms (best-of-N, beam search, lookahead search) interact with different token structures, as well as the role of different verifiers and AR priors. Our results highlight the impact of token structure on inference-time scalability and provide practical guidance for test-time scaling in AR models.
Authors: Jevin Jiang, Ying Chen, Blake A. Hechtman, Fenghui Zhang, Yarong Mu
Abstract: Large Language Model (LLM) deployment is increasingly shifting to cost-efficient accelerators like Google's Tensor Processing Units (TPUs), prioritizing both performance and total cost of ownership (TCO). However, existing LLM inference kernels and serving systems remain largely GPU-centric, and there is no well-established approach for efficiently mapping LLM workloads onto TPU architectures--particularly under the dynamic and ragged execution patterns common in modern serving. In this paper, we present Ragged Paged Attention (RPA), a high-performance and flexible attention kernel for TPUs, implemented using Pallas and Mosaic. RPA addresses these challenges through three key techniques: (1) fine-grained tiling to enable efficient dynamic slicing over ragged memory, (2) a custom software pipeline that fuses KV cache updates with attention computation, and (3) a distribution-aware compilation strategy that generates specialized kernels for decode, prefill, and mixed workloads. Evaluated on Llama 3 8B on TPU7x, RPA achieves up to 86% memory bandwidth utilization (MBU) in decode and 73% model FLOPs utilization (MFU) in prefill. Integrated as the primary TPU backend in vLLM and SGLang, RPA provides a production-grade foundation for efficient TPU inference and offers practical insights into kernel design.
Authors: Yuriy Kim, Evgeny Belyaev
Abstract: This paper is dedicated to lossless data compression with probability estimation using neural networks. First, we propose a probability estimation architecture based on a chain of neural predictors, so that each unit of the chain is defined as a neural network with the minimum possible number of weights, which is sufficient for efficient compression of data generated by Markov sources of a given order. We show that this architecture allows us to minimize the overall number of weights participating in the probability estimation process depending on the statistical properties of the input data. Second, in order to improve compression efficiency, we introduce an information inheritance mechanism, where the probability estimate obtained by a low-order unit is used at the next higher-order unit. Experimental results show that the proposed lossless data compressor equipped with the chained probability estimation architecture provides compression ratios close to the state-of-the-art PAC compressor. At the same time, it outperforms PAC by a factor of 1.2 to 6.3 in encoding throughput and by a factor of 2.8 to 12.3 in decoding throughput on a consumer GPU.
Authors: Abhishek Uppar, Omar Muhammad, Sumanth Prabhu, Deepak D'Souza, Madhusudan P, Adithya Murali
Abstract: We consider the problem of \emph{verification modulo tested library contracts} as a step towards automating the verification of client programs that use complex libraries. We formulate this problem as the synthesis of modular contracts for the library methods used by the client that are adequate to prove the client correct, and that also pass the scrutiny of a testing engine that tests the library against these contracts. We also consider a new form of method contracts called \emph{contextual contracts} that arise in this setting that hold in the context of the client program, and can often be simpler and easier to infer than classical modular contracts. We provide a counterexample-guided learning framework to solve this problem, in which the synthesizer interacts with a constraint solver as well as the testing engine in order to infer adequate modular/contextual method contracts and inductive invariants for the client. The main synthesis engines we use are generalizing CHC solvers that are realized using ICE learning algorithms. We realize this framework in a tool called \vmtlc and show its efficacy on benchmarks where clients call large libraries.
Authors: Tianhao Liu, Daniel Andr\'es D\'iaz-Pach\'on, J. Sunil Rao
Abstract: Supervised No Free Lunch Theorems (NFLTs) are well studied, yet unsupervised NFLTs remain underexplored. For elliptical distributions, we prove that there exist two equally optimal, scientifically meaningful bump-hunting strategies that are exact opposites, with no universal winner. Specifically, peeling $k$ orthogonal dimensions from $\mathbb{R}^d$ ($d \ge k$), retaining an inter-quantile region of probability $1-\alpha$ per peeled dimension, maximizes total variance and Frobenius norm when the $k$ smallest principal components (called pettiest components) are selected, and minimizes them when the selected dimensions are the $k$ leading principal components. These optima inspire PRIM-based bump-hunting algorithms either by minimizing variance or by minimizing volume, thereby motivating an NFLT. We test our results on the Fashion-MNIST database, showing that peeling the largest principal components captures multiplicity, while peeling the smallest principal components isolates popular styles.
Authors: Kyle Lucke, Zuzanna Krajewska-Travar, Shoukun Sun, Lu Cai, John D. Stempien, Min Xian
Abstract: Tristructural isotropic (TRISO)-coated particle fuels undergo dimensional changes and chemical reactions during high-temperature neutron irradiation. Post-irradiation materialography helps understand processes that impact fuel performance, such as coating integrity and fission product retention. Conventionally, experts manually evaluate features in thousands of cross sections of sub-mm-sized samples, which is tedious and subjective. In this work, we propose UA-Net, a deep learning framework that segments five characteristic regions of TRISO fuel micrographs and generates an uncertainty map for predictions. The model uses a multi-stage pretraining strategy, starting with general image representations learned from ImageNet, followed by fine-tuning on TRISO micrographs from various irradiation experiments and AGR-5/6/7 particle cross sections. A meta-model for uncertainty prediction is integrated to identify small defects in TRISO images. UA-Net was evaluated on a test set of 102 images, achieving mean Intersection over Union (mIoU) and mean Precision (mP) of 95.5% and 97.3%, respectively. The meta-model achieved a specificity of 91.8% and sensitivity of 93.5%, demonstrating strong performance in detecting misclassifications. The model was also applied to new TRISO images for qualitative evaluation, showing high accuracy in extracting layer regions.
Authors: Md. Rashadul Islam
Abstract: NASA's Transiting Exoplanet Survey Satellite (TESS) has identified thousands of exoplanet candidates, yet many remain unconfirmed due to the limitations of manual vetting processes. This paper presents ExoNet, a multimodal deep learning framework that integrates phase-folded global and local light curve representations with stellar parameters using a late-fusion architecture combining 1D Convolutional Neural Networks and Multi-Head Attention. Trained on labeled Kepler data, ExoNet achieves strong classification performance and demonstrates effective generalization to TESS data. Applied to 200 unconfirmed TESS planet candidates, the model identifies multiple high-confidence candidates, including several within the habitable zone. The results highlight the effectiveness of multimodal fusion and attention mechanisms in automated exoplanet candidate validation.
Authors: Pedro R. Pires, Rafael T. Sereicikas, Gregorio F. Azevedo, Tiago A. Almeida
Abstract: In recent years, neural networks and other complex models have dominated recommender systems, often setting new benchmarks for state-of-the-art performance. Yet, despite these advancements, award-winning research has demonstrated that traditional matrix factorization methods can remain competitive, offering simplicity and reduced computational overhead. Hybrid models, which combine matrix factorization with newer techniques, are increasingly employed to harness the strengths of multiple approaches. This paper proposes a novel ensemble method that unifies user-item and item-item recommendations through a weighted similarity framework to deliver top-N recommendations. Our approach is distinctive in its use of shared user and item embeddings for both recommendation strategies, simplifying the architecture and enhancing computational efficiency. Extensive experiments across multiple datasets show that our method achieves competitive performance and is robust in varying scenarios that favor either user-item or item-item recommendations. Additionally, by eliminating the need for embedding-specific fine-tuning, our model allows for the seamless reuse of hyperparameters from the base algorithm without sacrificing performance. This results in a method that is both efficient and easy to implement. Our open-source implementation is available at https://github.com/UFSCar-LaSID/weighted-sims-recommender.
URLs: https://github.com/UFSCar-LaSID/weighted-sims-recommender.
Authors: Guy Kaplan, Zorik Gekhman, Zhen Zhu, Lotem Rozner, Yuval Reif, Swabha Swayamdipta, Derek Hoiem, Roy Schwartz
Abstract: Large language models are prone to hallucinating factually incorrect statements. A key source of these errors is exposure to new factual information through supervised fine-tuning (SFT), which can increase hallucinations w.r.t. knowledge acquired during pre-training. In this work, we explore whether SFT-induced hallucinations can be mitigated using established tools from the continual learning literature, since they arise as a by-product of knowledge degradation during training. We propose a self-distillation-based SFT method that facilitates effective factual learning while minimizing hallucinations w.r.t. pre-existing knowledge by regularizing output-distribution drift. We also show that, in settings where new knowledge acquisition is unnecessary, suppressing factual plasticity by freezing parameter groups, can preserve task performance while reducing hallucinations. Lastly, we investigate the mechanism behind SFT-induced hallucinations through three hypotheses: capacity limitations, behavior cloning, and localized interference. Our experiments show that a main driver is interference among overlapping semantic representations, and that self-distillation succeeds by mitigating this interference.
Authors: Rafael T. Sereicikas, Pedro R. Pires, Gregorio F. Azevedo, Tiago A. Almeida
Abstract: Effective user modeling requires distinguishing between short-term and long-term preference evolution. While item embeddings have become a key component of recommender systems, standard approaches like Item2Vec treat user histories as unordered sets (bag-of-items), implicitly assuming that interactions separated by minutes are as semantically related as those separated by months. This simplification flattens the rich temporal structure of user behavior, obscuring the distinction between coherent consumption sessions and gradual interest drifts. In this work, we introduce TAI2Vec (Time-Aware Item-to-Vector), a family of lightweight embedding models that integrates temporal proximity directly into the representation learning process. Unlike approaches that apply global time constraints, TAI2Vec is user-adaptive, tailoring its temporal definitions to individual interaction paces. We propose two complementary strategies: TAI2Vec-Disc, which utilizes personalized anomaly detection to dynamically segment interactions into semantic sessions, and TAI2Vec-Cont, which employs continuous, user-specific decay functions to weigh item relationships based on their relative temporal distance. Experimental results across eight diverse datasets demonstrate that TAI2Vec consistently produces more accurate and behaviorally grounded representations than static baselines, achieving competitive or superior performance in over 80% of the datasets, with improvements of up to 135%. The source code is publicly available at https://github.com/UFSCar-LaSID/tai2vec.
Authors: Yang Wu, Jinhong Yu, Jingwei Xiong, Zhimin Tao, Xiaozhong Liu
Abstract: The integration of Large Language Models (LLMs) into scientific workflows presents exciting opportunities to accelerate biomedical discovery. However, the reactive nature of LLMs, which respond only when prompted, limits their effectiveness in collaborative settings that demand foresight and autonomous engagement. In this study, we introduce CoLabScience, a proactive LLM assistant designed to enhance biomedical collaboration between AI systems and human experts through timely, context-aware interventions. At the core of our method is PULI (Positive-Unlabeled Learning-to-Intervene), a novel framework trained with a reinforcement learning objective to determine when and how to intervene in streaming scientific discussions, by leveraging the team's project proposal and long- and short-term conversational memory. To support this work, we introduce BSDD (Biomedical Streaming Dialogue Dataset), a new benchmark of simulated research discussion dialogues with intervention points derived from PubMed articles. Experimental results show that PULI significantly outperforms existing baselines in both intervention precision and collaborative task utility, highlighting the potential of proactive LLMs as intelligent scientific assistants.
Authors: Jack Wei Lun Shi, Minghao Dang, Wawan Solihin, Justin K. W. Yeoh
Abstract: Existing research on large language models (LLMs) for automated code compliance has primarily focused on performance, treating the models as black boxes and overlooking how training decisions affect their interpretive behavior. This paper addresses this gap by employing a perturbation-based attribution analysis to compare the interpretive behaviors of LLMs across different fine-tuning strategies such as full fine-tuning (FFT), low-rank adaptation (LoRA) and quantized LoRA fine-tuning, as well as the impact of model scales which include varying LLM parameter sizes. Our results show that FFT produces attribution patterns that are statistically different and more focused than those from parameter-efficient fine-tuning methods. Furthermore, we found that as model scale increases, LLMs develop specific interpretive strategies such as prioritizing numerical constraints and rule identifiers in the building text, albeit with performance gains in semantic similarity of the generated and reference computer-processable rules plateauing for models larger than 7B. This paper provides crucial insights into the explainability of these models, taking a step toward building more transparent LLMs for critical, regulation-based tasks in the Architecture, Engineering, and Construction industry.
Authors: Jiamin Jiang, Shanglin Lv, Jingrun Chen
Abstract: Hyperbolic conservation laws govern a wide range of transport-driven dynamics featuring shocks, contact discontinuities, and complex wave interactions, posing distinct challenges for deep-learning-based surrogate modeling. While classical numerical methods provide robust and physically admissible solutions, their computational cost restricts applicability in many-query tasks such as parametric studies and design optimization. Conversely, existing neural surrogates offer rapid inference but often fail to respect intrinsic PDE structures, leading to non-physical artifacts, rollout instability, and poor generalization. We present an interpretable, structure-preserving graph neural solver that bridges classical numerical principles with graph neural networks (GNNs). The network is designed as a learned reconstruction-and-flux operator rather than a black-box state updater, thereby inherently preserving key properties such as local conservation and upwinding. Inspired by Arbitrary high-order DERivatives schemes, we further recast message-passing GNNs as high-order space-time predictors, enabling conservative and stable neural updates with large time steps. Evaluation is performed on challenging supersonic flow benchmarks spanning broad parametric variations in geometry, initial/boundary conditions, and flow regimes. The neural solver achieves superior long-horizon rollout stability and accuracy compared with strong surrogate baselines, outperforms low-order discretizations, and delivers orders-of-magnitude runtime speedups over high-resolution simulations.
Authors: Yiwei Zhao, Yi Zheng, Huapeng Su, Jieyu Lin, Stefano Ambrogio, Cijo Jose, Micha\"el Ramamonjisoa, Patrick Labatut, Barbara De Salvo, Chiao Liu, Phillip B. Gibbons, Ziyun Li
Abstract: Language-aligned vision foundation models (VFMs) enable versatile visual understanding for always-on contextual AI, but their deployment on edge devices is hindered by strict latency and power constraints. We present AdaVFM, an adaptive framework for efficient on-device inference of language-aligned VFMs that dynamically adjusts computation based on scene context and task complexity. Our key insight is that the effect of model size reduction on performance is task-dependent in vision applications, motivating a runtime-adaptive execution strategy. AdaVFM integrates neural architecture search (NAS) into the language-aligned VFM backbone to enable lightweight subnet execution during runtime. A multimodal large language model (LLM) deployed on the cloud enables runtime control with a context-aware agent. This synergy allows efficient model adaptation under diverse conditions while maintaining strong accuracy. Extensive experiments on zero-shot classification and open-vocabulary segmentation demonstrate that AdaVFM achieves state-of-the-art accuracy-efficiency trade-offs, surpassing prior baselines by up to $7.9\%$ in acc@1 on IN1K and $5.2\%$ mIoU on ADE20K over the best models of comparable VFM sizes. For models with similar accuracy, AdaVFM further reduces average FLOPs by up to $77.9\%$.
Authors: Keisuke Gomi, Keiji Yanai
Abstract: Cross-modal retrieval between food images and recipe texts is an important task with applications in nutritional management, dietary logging, and cooking assistance. Existing methods predominantly rely on dual-encoder architectures with separate image and text encoders, requiring complex alignment strategies and task-specific network designs to bridge the semantic gap between modalities. In this work, we propose SIMMER (Single Integrated Multimodal Model for Embedding Recipes), which applies Multimodal Large Language Model (MLLM)-based embedding models, specifically VLM2Vec, to this task, replacing the conventional dual-encoder paradigm with a single unified encoder that processes both food images and recipe texts. We design prompt templates tailored to the structured nature of recipes, which consist of a title, ingredients, and cooking instructions, enabling effective embedding generation by the MLLM. We further introduce a component-aware data augmentation strategy that trains the model on both complete and partial recipes, improving robustness to incomplete inputs. Experiments on the Recipe1M dataset demonstrate that SIMMER achieves state-of-the-art performance across both the 1k and 10k evaluation settings, substantially outperforming all prior methods. In particular, our best model improves the 1k image-to-recipe R@1 from 81.8\% to 87.5\% and the 10k image-to-recipe R@1 from 56.5\% to 65.5\% compared to the previous best method.
Authors: Jon-Paul Cacioli
Abstract: We introduce a cross-domain behavioural assay of monitoring-control coupling in LLMs, grounded in the Nelson and Narens (1990) metacognitive framework and applying human psychometric methodology to LLM evaluation. The battery comprises 524 items across six cognitive domains (learning, metacognitive calibration, social cognition, attention, executive function, prospective regulation), each grounded in an established experimental paradigm. Tasks T1-T5 were pre-registered on OSF prior to data collection; T6 was added as an exploratory extension. After every forced-choice response, dual probes adapted from Koriat and Goldsmith (1996) ask the model to KEEP or WITHDRAW its answer and to BET or decline. The critical metric is the withdraw delta: the difference in withdrawal rate between incorrect and correct items. Applied to 20 frontier LLMs (10,480 evaluations), the battery discriminates three profiles consistent with the Nelson-Narens architecture: blanket confidence, blanket withdrawal, and selective sensitivity. Accuracy rank and metacognitive sensitivity rank are largely inverted. Retrospective monitoring and prospective regulation appear dissociable (r = .17, 95% CI wide given n=20; exemplar-based evidence is the primary support). Scaling on metacognitive calibration is architecture-dependent: monotonically decreasing (Qwen), monotonically increasing (GPT-5.4), or flat (Gemma). Behavioural findings converge structurally with an independent Type-2 SDT approach, providing preliminary cross-method construct validity. All items, data, and code: https://github.com/synthiumjp/metacognitive-monitoring-battery.
URLs: https://github.com/synthiumjp/metacognitive-monitoring-battery.
Authors: Hyeongmeen Baik, Hamed Poursiami, Maryam Parsa, Jinia Roy
Abstract: Always-on converter health monitoring demands sub-mW edge inference, a regime inaccessible to GPU-based physics-informed neural networks. This work separates spiking temporal processing from physics enforcement: a three-layer leaky integrate-and-fire SNN estimates passive component parameters while a differentiable ODE solver provides physics-consistent training by decoupling the ODE physics loss from the unrolled spiking loop. On an EMI-corrupted synchronous buck converter benchmark, the SNN reduces lumped resistance error from $25.8\%$ to $10.2\%$ versus a feedforward baseline, within the $\pm 10\%$ manufacturing tolerance of passive components, at a projected ${\sim}270\times$ energy reduction on neuromorphic hardware. Persistent membrane states further enable degradation tracking and event-driven fault detection via a $+5.5$ percentage-point spike-rate jump at abrupt faults. With $93\%$ spike sparsity, the architecture is suited for always-on deployment on Intel Loihi 2 or BrainChip Akida.
Authors: Junguang Yao, Wenye Liu, Stjepan Picek, Yue Zheng
Abstract: Visual speaker recognition based on lip motion offers a silent, hands-free, and behavior-driven biometric solution that remains effective even when acoustic cues are unavailable. Compared to traditional methods that rely heavily on appearance-dependent representations, lip motion encodes subject-specific behavioral dynamics driven by consistent articulation patterns and muscle coordination, offering inherent stability across environmental changes. However, capturing these robust, fine-grained dynamics is challenging for conventional frame-based cameras due to motion blur and low dynamic range. To exploit the intrinsic stability of lip motion and address these sensing limitations, we propose NeuroLip, an event-based framework that captures fine-grained lip dynamics under a strict yet practical cross-scene protocol: training is performed under a single controlled condition, while recognition must generalize to unseen viewing and lighting conditions. NeuroLip features a 1) Temporal-aware Voxel Encoding module with adaptive event weighting, 2) Structure-aware Spatial Enhancer that amplifies discriminative behavioral patterns by suppressing noise while preserving vertically structured motion information, and 3) Polarity Consistency Regularization mechanism to retain motion-direction cues encoded in event polarities. To facilitate systematic evaluation, we introduce DVSpeaker, a comprehensive event-based lip-motion dataset comprising 50 subjects recorded under four distinct viewpoint and illumination scenarios. Extensive experiments demonstrate that NeuroLip achieves near-perfect matched-scene accuracy and robust cross-scene generalization, attaining over 71% accuracy on unseen viewpoints and nearly 76% under low-light conditions, outperforming representative existing methods by at least 8.54%. The dataset and code are publicly available at https://github.com/JiuZeongit/NeuroLip.
Authors: Sankalp Gilda, Shlok Gilda
Abstract: Large language models exhibit systematic limitations in structured logical reasoning: they conflate hypothesis generation with verification, cannot distinguish conjecture from validated knowledge, and allow weak reasoning steps to propagate unchecked through inference chains. We present a symbolic reasoning scaffold that operationalizes Peirce's tripartite inference -- abduction, deduction, and induction -- as an explicit protocol for LLM-assisted reasoning. The framework enforces logical consistency through five algebraic invariants (the Gamma Quintet), the strongest of which -- the Weakest Link bound -- ensures that no conclusion in a reasoning chain can exceed the reliability of its least-supported premise. This principle, independently grounded as weakest link resolution in possibilistic logic and empirically validated for chain-of-thought reasoning, prevents logical inconsistencies from accumulating across multi-step inference. We verify all invariants through a property-based testing suite of 100 properties and 16 fuzz tests over 10^5+ generated cases, providing a verified reference implementation of the invariants suitable as a foundation for future reasoning benchmarks.
Authors: Yuanchang Zhou, Hongyu Wang, Yiming Du, Yan Wang, Mingzhen Li, Siyu Hu, Xiangyu Zhang, Weijian Liu, Chen Wang, Zhuoqiang Guo, Long Wang, Jingde Bu, Yutong Lu, Guangming Tan, Weile Jia
Abstract: Universal Machine Learning Interatomic Potentials (uMLIPs), pre-trained on massively diverse datasets encompassing inorganic materials and organic molecules across the entire periodic table, serve as foundational models for quantum-accurate physical simulations. However, uMLIP training requires second-order derivatives, which lack corresponding parallel training frameworks; moreover, scaling to the billion-parameter regime causes explosive growth in computation and communication overhead, making its training a tremendous challenge. We introduce MatRIS-MoE, a billion-parameter Mixture-of-Experts model built upon invariant architecture, and {Janus}, a pioneering high-dimensional distributed training framework for uMLIPs with hardware-aware optimizations. Deployed across two Exascale supercomputers, our code attains a peak performance of 1.2/1.0 EFLOPS (24\%/{35.5\%} of theoretical peak) in single precision at over 90\% parallel efficiency, compressing the training of billion-parameter uMLIPs from weeks to hours. This work establishes a new high-water mark for AI-for-Science (AI4S) foundation models at Exascale and provides essential infrastructure for rapid scientific discovery.
Authors: Siun Kim, Hyung-Jin Yoon
Abstract: Large language models (LLMs) have advanced information extraction (IE) by enabling zero-shot and few-shot named entity recognition (NER), yet their generative outputs still show persistent and systematic errors. Despite progress through instruction fine-tuning, zero-shot NER still lags far behind supervised systems. These recurring errors mirror inconsistencies observed in early-stage human annotation processes that resolve disagreements through pilot annotation. Motivated by this analogy, we introduce DiZiNER (Disagreement-guided Instruction Refinement via Pilot Annotation Simulation for Zero-shot Named Entity Recognition), a framework that simulates the pilot annotation process, employing LLMs to act as both annotators and supervisors. Multiple heterogeneous LLMs annotate shared texts, and a supervisor model analyzes inter-model disagreements to refine task instructions. Across 18 benchmarks, DiZiNER achieves zero-shot SOTA results on 14 datasets, improving prior bests by +8.0 F1 and reducing the zero-shot to supervised gap by over +11 points. It also consistently outperforms its supervisor, GPT-5 mini, indicating that improvements stem from disagreement-guided instruction refinement rather than model capacity. Pairwise agreement between models shows a strong correlation with NER performance, further supporting this finding.
Authors: Fabian Ridder, Laurin Lessel, Malte Schilling
Abstract: Retrieval-Augmented Generation (RAG) is widely used to augment the input to Large Language Models (LLMs) with external information, such as recent or domain-specific knowledge. Nonetheless, current models still produce closed-domain hallucinations and generate content that is unsupported by the retrieved context. Current detection approaches typically treat hallucination as a post-hoc problem, relying on black-box consistency checks or probes over frozen internal representations. In this work, we demonstrate that hallucination detection based on internal state representation can also serve as a direct training signal. We introduce RAGognize, a dataset of naturally occurring closed-domain hallucinations with token-level annotations, and RAGognizer, a hallucination-aware fine-tuning approach that integrates a lightweight detection head into an LLM, allowing for the joint optimization of language modeling and hallucination detection. This joint objective forces the model to improve the separability of its internal states regarding hallucinations while simultaneously learning to generate well-formed and meaningful responses. Across multiple benchmarks, RAGognizer achieves state-of-the-art token-level hallucination detection while substantially reducing hallucination rates during generation, without degrading language quality or relevance.
Authors: Prabin Bohara, Pralhad Kumar Shrestha, Arpan Rai, Usha Poudel Lamgade, Confidence Raymond, Dong Zhang, Aondona Lorumbu, Craig Jones, Mahesh Shakya, Bishesh Khanal, Pratibha Kulung
Abstract: Accurate automatic brain tumor segmentation in Low and Middle-Income (LMIC) countries is challenging due to the lack of defined national imaging protocols, diverse imaging data, extensive use of low-field Magnetic Resonance Imaging (MRI) scanners and limited health-care resources. As part of the Brain Tumor Segmentation (BraTS) Africa 2025 Challenge, we applied topology refinement to the state-of-the-art segmentation models like nnU-Net, MedNeXt, and a combination of both. Since the BraTS-Africa dataset has low MRI image quality, we incorporated the BraTS 2025 challenge data of pre-treatment adult glioma (Task 1) to pre-train the segmentation model and use it to fine-tune on the BraTS-Africa dataset. We added an extra topology refinement module to address the issue of deformation in prediction that arose due to topological error. With the introduction of this module, we achieved a better Normalized Surface Distance (NSD) of 0.810, 0.829, and 0.895 on Surrounding Non-Enhancing FLAIR Hyperintensity (SNFH) , Non-Enhancing Tumor Core (NETC) and Enhancing tumor (ET).
Authors: Paulin de Schoulepnikoff, Hendrik Poulsen Nautrup, Hans J. Briegel, Gorka Mu\~noz-Gil
Abstract: Interpretable machine learning techniques are becoming essential tools for extracting physical insights from complex quantum data. We build on recent advances in variational autoencoders to demonstrate that such models can learn physically meaningful and interpretable representations from a broad class of unlabeled quantum datasets. From raw measurement data alone, the learned representation reveals rich information about the underlying structure of quantum phase spaces. We further augment the learning pipeline with symbolic methods, enabling the discovery of compact analytical descriptors that serve as order parameters for the distinct regimes emerging in the learned representations. We demonstrate the framework on experimental Rydberg-atom snapshots, classical shadows of the cluster Ising model, and hybrid discrete-continuous fermionic data, revealing previously unreported phenomena such as a corner-ordering pattern in the Rydberg arrays. These results establish a general framework for the automated and interpretable discovery of physical laws from diverse quantum datasets. All methods are available through qdisc, an open-source Python library designed to make these tools accessible to the broader community.
Authors: Hikaru Shindo, Hanzhao Lin, Lukas Helff, Patrick Schramowski, Kristian Kersting
Abstract: As Large Language Models (LLMs) transition from text processors to autonomous agents, evaluating their social reasoning in embodied multi-agent settings becomes critical. We introduce SocialGrid, an embodied multi-agent environment inspired by Among Us that evaluates LLM agents on planning, task execution, and social reasoning. Our evaluations reveal that even the strongest open model (GPT-OSS-120B) achieves below 60% accuracy in task completion and planning, with agents getting stuck in repetitive behaviors or failing to navigate basic obstacles. Since poor navigation confounds evaluation of social intelligence, SocialGrid offers an optional Planning Oracle to isolate social reasoning from planning deficits. While planning assistance improves task completion, social reasoning remains a bottleneck: agents fail to detect deception at near-random chance regardless of scale, relying on shallow heuristics rather than accumulating behavioral evidence. SocialGrid provides automatic failure analysis and fine-grained metrics, enabling developers to diagnose and improve their agents. We also establish a competitive leaderboard using Elo ratings from adversarial league play.
Authors: Constantinos Karouzos, Xingwei Tan, Nikolaos Aletras
Abstract: Post-trained language models produce less varied outputs than their base counterparts. This output diversity collapse undermines inference-time scaling methods that rely on varied samples, and risks homogenizing model outputs on creative and value-laden tasks. Prior work attributes collapse to specific post-training methods, without separating the role of training data composition from the method, or the generation format from the model weights. We trace output diversity through three parallel post-training lineages of Olmo 3, Think (chain-of-thought distillation), Instruct (broad multi-source data), and RL-Zero, across 15 tasks and four text diversity metrics. We find that the location of collapse co-varies with data composition: the Think lineage loses most semantic diversity at supervised fine-tuning, and the effect of DPO is larger in Instruct than in Think. Suppressing chain-of-thought reasoning at inference in Think models drops accuracy on hard tasks, yet leaves answer-level diversity unchanged, showing that the collapse is embedded in the model weights by training data, not imposed by the generation format. Decomposing diversity loss on six verifiable tasks into a quality-control component (removal of incorrect outputs) and a residual component (genuine narrowing among correct outputs) reveals that the split is task-dependent, and Think models retain more correct-answer diversity than Instruct despite collapsing more in aggregate. Our results indicate that diversity collapse is determined during training by data composition and cannot be addressed at inference time alone.
Authors: Jiaxi Bi, Tongxu Luo, Wenyu Du, Zhengyang Tang, Benyou Wang
Abstract: Parallel reasoning enhances Large Reasoning Models (LRMs) but incurs prohibitive costs due to futile paths caused by early errors. To mitigate this, path pruning at the prefix level is essential, yet existing research remains fragmented without a standardized framework. In this work, we propose the first systematic taxonomy of path pruning, categorizing methods by their signal source (internal vs. external) and learnability (learnable vs. non-learnable). This classification reveals the unexplored potential of learnable internal methods, motivating our proposal of STOP (Super TOken for Pruning). Extensive evaluations across LRMs ranging from 1.5B to 20B parameters demonstrate that STOP achieves superior effectiveness and efficiency compared to existing baselines. Furthermore, we rigorously validate the scalability of STOP under varying compute budgets - for instance, boosting GPT-OSS-20B accuracy on AIME25 from 84% to nearly 90% under fixed compute budgets. Finally, we distill our findings into formalized empirical guidelines to facilitate optimal real-world deployment. Code, data and models are available at https://bijiaxihh.github.io/STOP
Authors: Yutong Gao, Qinglin Meng, Yuan Zhou, Liangming Pan
Abstract: While Large Language Models (LLMs) have achieved strong performance across many NLP tasks, their opaque internal mechanisms hinder trustworthiness and safe deployment. Existing surveys in explainable AI largely focus on post-hoc explanation methods that interpret trained models through external approximations. In contrast, intrinsic interpretability, which builds transparency directly into model architectures and computations, has recently emerged as a promising alternative. This paper presents a systematic review of the recent advances in intrinsic interpretability for LLMs, categorizing existing approaches into five design paradigms: functional transparency, concept alignment, representational decomposability, explicit modularization, and latent sparsity induction. We further discuss open challenges and outline future research directions in this emerging field. The paper list is available at: https://github.com/PKU-PILLAR-Group/Survey-Intrinsic-Interpretability-of-LLMs.
URLs: https://github.com/PKU-PILLAR-Group/Survey-Intrinsic-Interpretability-of-LLMs.
Authors: Abdulaziz Aldegheishem, Nabil Alrajeh, Lorena Parra, Oscar Romero, Jaime Lloret
Abstract: The ambulance service is the main transport for diseased or injured people which suffers the same acceleration forces as regular vehicles. These accelerations, caused by the movement of the vehicle, impact the performance of tasks executed by sanitary personnel, which can affect patient survival or recovery time. In this paper, we have trained, validated, and tested a system to assess driving in ambulance services. The proposed system is composed of a sensor node which measures the vehicle vibrations using an accelerometer. It also includes a GPS sensor, a battery, a display, and a speaker. When two possible routes reach the same destination point, the system compares the two routes based on previously classified data and calculates an index and a score. Thus, the index balances the possible routes in terms of time to reach the destination and the vibrations suffered in the patient cabin to recommend the route that minimises those vibrations. Three datasets are used to train, validate, and test the system. Based on an Artificial Neural network (ANN), the classification model is trained with tagged data classified as low, medium, and high vibrations, and 97% accuracy is achieved. Then, the obtained model is validated using data from three routes of another region. Finally, the system is tested in two new scenarios with two possible routes to reach the destination. The results indicate that the route with less vibration is preferred when there are low time differences (less than 6%) between the two possible routes. Nonetheless, with the current weighting factors, the shortest route is preferred when time differences between routes are higher than 20%, regardless of the higher vibrations in the shortest route.
Authors: Ulrich Tan
Abstract: We introduce the Tan-HWG framework (Hebbian-Wasserstein-Geometry), a geometric theory of Hebbian plasticity in which memory states are modeled as probability measures evolving through Wasserstein minimizing movements. Hebbian learning rules are formalized as Hebbian energies satisfying a sequential stability condition, ensuring well-posed fiberwise JKO updates, optimal-transport realizations, and an energy descent inequality. This variational structure induces a fundamental separation between internal and observable dynamics. Internal memory states evolve along Wasserstein geodesics in a latent curved space, while observable quantities, such as effective synaptic weights, arise through geometric projection maps into external spaces. Simplicial projections recover classical affine schemes (including exponential moving averages and mirror descent), while revealing synaptic competition and pruning as geometric consequences of mass redistribution. Hilbertian projections provide a geometric account of phase alignment and multi-scale coherence. Classical neural networks appear as flat projections of this curved dynamics, while the framework naturally accommodates richer distributional representations, including structural weights and embedding memories, and their spectral extensions in complex internal spaces. Under mild Lipschitz regularity assumptions, including a quasi-stationary "sleep-mode" regime, we establish the existence of continuous-time limit curves. This yields a variational formulation of memory consolidation as a perturbed Wasserstein gradient flow. The framework thus provides a unified geometric foundation for synaptic plasticity, representation dynamics, and context-dependent computation.
Authors: Nicole Funk, Annika Hennes, Johanna Hillebrand, Sarah Sturm
Abstract: We study discrete k-clustering problems in general metric spaces that are constrained by a combination of two different fairness conditions within the demographic fairness model. Given a metric space (P,d), where every point in P is equipped with a protected attribute, and a number k, the goal is to partition P into k clusters with a designated center each, such that a center-based objective function is minimized and the attributes are fairly distributed with respect to the following two fairness concepts: 1) group fairness: We aim for clusters with balanced numbers of attributes by specifying lower and upper bounds for the desired attribute proportions. 2) diverse center selection: Clusters have natural representatives, i.e., their centers. We ask for a balanced set of representatives by specifying the desired number of centers to choose from each attribute. Dickerson, Esmaeili, Morgenstern and Zhang (2023) denote the combination of these two constraints as doubly constrained fair clustering. They present algorithms whose guarantees depend on the best known approximation factors for either of these problems. Currently, this implies an 8-approximation with a small additive violation on the group fairness constraint. For k-center, we improve this approximation factor to 4 with a small additive violation. This guarantee also depends on the currently best algorithm for DS-fair k-center given by Jones, Nguyen and Nguyen (2020). For k-median and k-means, we propose the first constant-factor approximation algorithms. Our algorithms transform a solution that satisfies diverse center selection into a doubly constrained fair clustering using an LP-based approach. Furthermore, our results are generalizable to other center-selection constraints, such as matroid k-clustering and knapsack constraints.
Authors: Micha{\l} Derezi\'nski, Yuji Nakatsukasa, Elizaveta Rebrova
Abstract: The quest for an algorithm that solves an $n\times n$ linear system in $O(n^2)$ time complexity, or $O(n^2 \text{poly}(1/\epsilon))$ when solving up to $\epsilon$ relative error, is a long-standing open problem in numerical linear algebra and theoretical computer science. There are two predominant paradigms for measuring relative error: forward error (i.e., distance from the output to the optimum solution) and backward error (i.e., distance to the nearest problem solved by the output). In most prior studies, convergence of iterative linear system solvers is measured via various notions of forward error, and as a result, depends heavily on the conditioning of the input. Yet, the numerical analysis literature has long advocated for backward error as the more practically relevant notion of approximation. In this work, we show that -- surprisingly -- the classical and simple Richardson iteration incurs at most $1/k$ (relative) backward error after $k$ iterations on any positive semidefinite (PSD) linear system, irrespective of its condition number. This universal convergence rate implies an $O(n^2/\epsilon)$ complexity algorithm for solving a PSD linear system to $\epsilon$ backward error, and we establish similar or better complexity when using a variety of Krylov solvers beyond Richardson. Then, by directly minimizing backward error over a Krylov subspace, we attain an even faster $O(1/k^2)$ universal rate, and we turn this into an efficient algorithm, MINBERR, with complexity $O(n^2/\sqrt\epsilon)$. We extend this approach via normal equations to solving general linear systems, for which we empirically observe $O(1/k)$ convergence. We report strong numerical performance of our algorithms on benchmark problems.
Authors: Enas E. Ahmed, Salah A. Aly, Mayar Moner
Abstract: Acute Myeloid Leukemia (AML) is one of the most life-threatening type of blood cancers, and its accurate classification is considered and remains a challenging task due to the visual similarity between various cell types. This study addresses the classification of the multiclasses of AML cells Utilizing YOLOv12 deep learning model. We applied two segmentation approaches based on cell and nucleus features, using Hue channel and Otsu thresholding techniques to preprocess the images prior to classification. Our experiments demonstrate that YOLOv12 with Otsu thresholding on cell-based segmentation achieved the highest level of validation and test accuracy, both reaching 99.3%.
Authors: Hamed Ouattara, Pierre Duthon, Pascal Houssam Salmane, Fr\'ed\'eric Bernardin, Omar Ait Aider
Abstract: One of the dominant paradigms in self-supervised learning (SSL), illustrated by MoCo or DINO, aims to produce robust representations by capturing features that are insensitive to certain image transformations such as illumination, or geometric changes. This strategy is appropriate when the objective is to recognize objects independently of their appearance. However, it becomes counterproductive as soon as appearance itself constitutes the discriminative signal. In weather analysis, for example, rain streaks, snow granularity, atmospheric scattering, as well as reflections and halos, are not noise: they carry the essential information. In critical applications such as autonomous driving, ignoring these cues is risky, since grip and visibility depend directly on ground conditions and atmospheric conditions. We introduce ST-STORM, a hybrid SSL framework that treats appearance (style) as a semantic modality to be disentangled from content. Our architecture explicitly separates two latent streams, regulated by gating mechanisms. The Content branch aims at a stable semantic representation through a JEPA scheme coupled with a contrastive objective, promoting invariance to appearance variations. In parallel, the Style branch is constrained to capture appearance signatures (textures, contrasts, scattering) through feature prediction and reconstruction under an adversarial constraint. We evaluate ST-STORM on several tasks, including object classification (ImageNet-1K), fine-grained weather characterization, and melanoma detection (ISIC 2024 Challenge). The results show that the Style branch effectively isolates complex appearance phenomena (F1=97% on Multi-Weather and F1=94% on ISIC 2024 with 10% labeled data), without degrading the semantic performance (F1=80% on ImageNet-1K) of the Content branch, and improves the preservation of critical appearance
Authors: Fabrizio Nunnari, Siddhant Jain, Patrick Gebhard
Abstract: We present a dataset and a model for sentiment analysis of German sign language (DGS) fairy tales. First, we perform sentiment analysis for three levels of valence (negative, neutral, positive) on German fairy tales text segments using four large language models (LLMs) and majority voting, reaching an inter-annotator agreement of 0.781 Krippendorff's alpha. Second, we extract face and body motion features from each corresponding DGS video segment using MediaPipe. Finally, we train an explainable model (based on XGBoost) to predict negative, neutral or positive sentiment from video features. Results show an average balanced accuracy of 0.631. A thorough analysis of the most important features reveal that, in addition to eyebrows and mouth motion on the face, also the motion of hips, elbows, and shoulders considerably contribute in the discrimination of the conveyed sentiment, indicating an equal importance of face and body for sentiment communication in sign language.
Authors: Max Henning H\"oth, Kristian Kersting, Bj\"orn Deiseroth, Letitia Parcalabescu
Abstract: Large language models (LLMs) increasingly rely on chain-of-thought (CoT) reasoning to solve complex tasks. Yet ensuring that the reasoning trace both contributes to and faithfully reflects the processes underlying the model's final answer, rather than merely accompanying it, remains challenging. We introduce AtManRL, a method that leverages differentiable attention manipulation to learn more faithful reasoning through reinforcement learning. By training an additive attention mask that identifies tokens in the CoT crucial for producing correct answers, we derive a saliency reward signal that encourages the model to generate reasoning traces that genuinely influence its final predictions. We integrate this saliency reward with outcome-based rewards within the GRPO framework to jointly optimize for correctness and interpretability. Experiments on GSM8K and MMLU with Llama-3.2-3B-Instruct demonstrate that our approach can identify influential reasoning tokens and enable training more transparent reasoning models.
Authors: Come Fiegel, Victor Gabillon, Michal Valko
Abstract: In multi-fidelity optimization, biased approximations of varying costs of the target function are available. This paper studies the problem of optimizing a locally smooth function with a limited budget, where the learner has to make a tradeoff between the cost and the bias of these approximations. We first prove lower bounds for the simple regret under different assumptions on the fidelities, based on a cost-to-bias function. We then present the Kometo algorithm which achieves, with additional logarithmic factors, the same rates without any knowledge of the function smoothness and fidelity assumptions, and improves previously proven guarantees. We finally empirically show that our algorithm outperforms previous multi-fidelity optimization methods without the knowledge of problem-dependent parameters.
Authors: Seulgi Kim, Mohit Prabhushankar, Ghassan AlRegib
Abstract: Vision Language models (VLMs) have demonstrated strong performance across a wide range of benchmarks, yet they often suffer from modality dominance, where predictions rely disproportionately on a single modality. Prior approaches primarily address this issue by steering model's attention allocation, implicitly assuming that all modalities provide sufficient information. However, attention only determines where the model focuses, and cannot enrich information that is missing or ambiguous. In the real world, input modalities often differ in information density and their signal-to-noise ratios. In such cases, simply adjusting model's attention does not resolve the underlying lack of information. In this paper, we propose \textsc{MoIR}: \textit{Multi-modal Information Router}, an information-level fusion method that explicitly reduces information disparity prior to fusion. \textsc{MoIR} identifies less informative tokens and routes complementary information from a stronger modality, constructing information-dense token representations before they are processed by a large language model. By modifying information availability, \textsc{MoIR} enables reliable shifts in modality dominance, even when one modality is degraded. We evaluate \textsc{MoIR} on three widely used multi-modal benchmarks across multiple model backbones. Experimental results show that \textsc{MoIR} consistently demonstrates more balanced modality contribution, and improves robustness and downstream performance, particularly even under modality degradation. These findings demonstrate that explicitly modifying cross-modal information is an effective and complementary strategy for mitigating modality dominance in multi-modal reasoning models.
Authors: Yunhe Li, Hao Shi, Bowen Deng, Wei Wang, Mengzhe Ruan, Hanxu Hou, Zhongxiang Dai, Siyang Gao, Chao Wang, Shuang Qiu, Linqi Song
Abstract: Although most of the automated theorem-proving approaches depend on formal proof systems, informal theorem proving can align better with large language models' (LLMs) strength in natural language processing. In this work, we identify a primary bottleneck in informal theorem proving as a lack of insight, namely the difficulty of recognizing the core techniques required to solve complex problems. To address this, we propose a novel framework designed to cultivate this essential reasoning skill and enable LLMs to perform insightful reasoning. We propose $\mathtt{DeepInsightTheorem}$, a hierarchical dataset that structures informal proofs by explicitly extracting core techniques and proof sketches alongside the final proof. To fully exploit this dataset, we design a Progressive Multi-Stage SFT strategy that mimics the human learning process, guiding the model from basic proof writing to insightful thinking. Our experiments on challenging mathematical benchmarks demonstrate that this insight-aware generation strategy significantly outperforms baselines. These results demonstrate that teaching models to identify and apply core techniques can substantially improve their mathematical reasoning.
Authors: Branislav Pecher, Ivan Srba, Maria Bielikova, Joaquin Vanschoren
Abstract: In few-shot learning, the selection of samples has a significant impact on the performance of the model. While effective sample selection strategies are well-established in supervised settings, research on large language models largely overlooks them, favouring strategies specifically tailored to individual in-context learning settings. In this paper, we propose a new method for Automatic Combination of SamplE Selection Strategies (ACSESS) to leverage the strengths and complementarity of various well-established selection objectives. We investigate and compare the impact of 23 sample selection strategies on the performance of 5 in-context learning models and 3 few-shot learning approaches (meta-learning, few-shot fine-tuning) over 6 text and 8 image datasets. The experimental results show that the combination of strategies through the ACSESS method consistently outperforms all individual selection strategies and performs on par or exceeds the in-context learning specific baselines. Lastly, we demonstrate that sample selection remains effective even on smaller datasets, yielding the greatest benefits when only a few shots are selected, while its advantage diminishes as the number of shots increases.
Authors: Luke Braithwaite, Alessio Borgi, Gabriele Onorato, Kristjan Tarantelli, Iulia Duta, Francesco Restuccia, Fabrizio Silvestri, Pietro Li\`o
Abstract: Heterogeneous graphs, whose nodes and edges may belong to different types and feature spaces, arise in a wide variety of real-world domains such as biology, chemistry and computer networks. Existing methods typically address this heterogeneity by modifying the model architecture itself, which often results in specialized and parameter-intensive designs. To address this issue, we propose HetSheaf, a framework that models heterogeneous relational data through cellular sheaves, which provide a principled topological framework for encoding type-specific local feature spaces and their interactions directly in the data representation. We also introduce a family of heterogeneous sheaf predictors that learn restriction maps conditioned on node and edge types. To enable graph-level predictions, we further propose SheafPool, a graph pooling mechanism that aggregates node representations in stalk space while remaining invariant to local changes of basis, ultimately enabling stalk-space graph-level representations for the first time. HetSheaf achieves strong predictive performance on standard heterogeneous graph benchmarks, over numerous tasks such as node/graph classification, link prediction and recommendation, while reducing by up to 10x the number of parameters with respect to state-of-the-art baselines.
Authors: Daniel Jenson, Jhonathan Navott, Mengyan Zhang, Makkunda Sharma, Elizaveta Semenova, Seth Flaxman
Abstract: Neural Processes (NPs) are a rapidly evolving class of models designed to directly model the posterior predictive distribution of stochastic processes. Originally developed as a scalable alternative to Gaussian Processes (GPs), which are limited by $O(n^3)$ runtime complexity, the most accurate modern NPs can often rival GPs but still suffer from an $O(n^2)$ bottleneck due to their attention mechanism. We introduce the Transformer Neural Process - Kernel Regression (TNP-KR), a scalable NP featuring: (1) a Kernel Regression Block (KRBlock), a simple, extensible, and parameter efficient transformer block with complexity $O(n_c^2 + n_c n_t)$, where $n_c$ and $n_t$ are the number of context and test points, respectively; (2) a kernel-based attention bias; and (3) two novel attention mechanisms: scan attention (SA), a memory-efficient scan-based attention that when paired with a kernel-based bias can make TNP-KR translation invariant, and deep kernel attention (DKA), a Performer-style attention that implicitly incoporates a distance bias and further reduces complexity to $O(n_c)$. These enhancements enable both TNP-KR variants to perform inference with 100K context points on over 1M test points in under a minute on a single 24GB GPU. On benchmarks spanning meta regression, Bayesian optimization, image completion, and epidemiology, TNP-KR with DKA outperforms its Performer counterpart on nearly every benchmark, while TNP-KR with SA achieves state-of-the-art results.
Authors: Anikait Singh, Sheryl Hsu, Kyle Hsu, Eric Mitchell, Stefano Ermon, Tatsunori Hashimoto, Archit Sharma, Chelsea Finn
Abstract: Effective personalization of LLMs is critical for a broad range of user-interfacing applications such as virtual assistants and content curation. Inspired by the strong in-context capabilities of LLMs, we propose few-shot preference optimization (FSPO), an algorithm for LLM personalization that reframes reward modeling as a meta-learning problem. Under FSPO, an LLM learns to quickly infer a personalized reward function for a user via a few labeled preferences. FSPO also utilizes user description rationalization (RAT) to encourage better reward modeling and instruction following, recovering performance with the oracle user description. Since real-world preference data is challenging to collect at scale, we propose careful design choices to construct synthetic preference datasets for personalization, generating over 1M synthetic personalized preferences using publicly available LLMs. To successfully transfer from synthetic data to real users, we find it crucial for the data to exhibit both high diversity and coherent, self-consistent structure. We evaluate FSPO on personalized open-ended generation for up to 1,500 synthetic users across three domains: movie reviews, education, and open-ended question answering. We also run a controlled human study. Overall, FSPO achieves an 87% Alpaca Eval winrate in generating responses that are personalized to synthetic users and a 70% winrate with real human users in open-ended question answering.
Authors: Witold Wydma\'nski, Marek \'Smieja
Abstract: Feature selection (FS) is a fundamental challenge in machine learning, particularly for high-dimensional tabular data, where interpretability and computational efficiency are critical. Existing FS methods often cannot automatically detect the number of attributes required to solve a given task and involve user intervention or model retraining with different feature budgets. Additionally, they either neglect feature relationships (filter methods) or require time-consuming optimization (wrapper methods). To address these limitations, we propose AutoNFS, which combines the FS module based on Gumbel-Sigmoid sampling with a predictive model evaluating the relevance of the selected attributes. The model is trained end-to-end using a differentiable loss and automatically determines the minimal set of features essential to solve a given downstream task. Unlike many wrapper-style approaches, AutoNFS introduces a low and predictable training overhead and avoids repeated model retraining across feature budgets. In practice, the additional cost of the masking module is largely independent of the number of input features (beyond the unavoidable cost of processing the input itself), making the method scalable to high-dimensional tabular data. We evaluate AutoNFS on well-established classification and regression benchmarks as well as real-world metagenomic datasets. The results show that AutoNFS is competitive with, and often improves upon, strong classical and neural FS baselines while selecting fewer features on average across the evaluated benchmarks.
Authors: Xinghao Wu, Jianwei Niu, Xuefeng Liu, Guogang Zhu, Jiayuan Zhang, Shaojie Tang, Wei Chen
Abstract: Federated Prototype Learning (FedPL) has emerged as an effective strategy for handling data heterogeneity in Federated Learning (FL). In FedPL, clients collaboratively construct a set of global feature centers (prototypes), and let local features align with these prototypes to mitigate the effects of data heterogeneity. The performance of FedPL highly depends on the quality of prototypes. Existing methods assume that larger inter-class distances among prototypes yield better performance, and thus design different methods to increase these distances. However, we observe that while these methods increase prototype distances to enhance class discrimination, they inevitably disrupt essential semantic relationships among classes, which are crucial for model generalization. This raises an important question: how to construct prototypes that inherently preserve semantic relationships among classes? Directly learning these relationships from limited and heterogeneous client data can be problematic in FL. Recently, the success of pre-trained language models (PLMs) demonstrates their ability to capture semantic relationships from vast textual corpora. Motivated by this, we propose FedTSP, a novel method that leverages PLMs to construct semantically enriched prototypes from the textual modality, enabling more effective collaboration in heterogeneous data settings. We first use a large language model (LLM) to generate fine-grained textual descriptions for each class, which are then processed by a PLM on the server to form textual prototypes. To address the modality gap between client image models and the PLM, we introduce trainable prompts, allowing prototypes to adapt better to client tasks. Extensive experiments demonstrate that FedTSP mitigates data heterogeneity while significantly accelerating convergence.
Authors: Shriyank Somvanshi, Md Monzurul Islam, Mahmuda Sultana Mimi, Sazzad Bin Bashar Polock, Gaurab Chhetri, Anandi Dutta, Amir Rafe, Subasish Das
Abstract: Structured State Space Models (SSMs) have emerged as a transformative paradigm in sequence modeling, addressing critical limitations of Recurrent Neural Networks (RNNs) and Transformers, namely, vanishing gradients, sequential computation bottlenecks, and quadratic memory complexity. By integrating structured recurrence with state-space representations, SSMs achieve linear or near-linear computational scaling while excelling in long-range dependency tasks. This study systematically traces the evolution of SSMs from the foundational Structured State Space Sequence (S4) model to modern variants like Mamba, Simplified Structured State Space Sequence (S5), and Jamba, analyzing architectural innovations that enhance computational efficiency, memory optimization, and inference speed. We critically evaluate trade-offs inherent to SSM design, such as balancing expressiveness with computational constraints and integrating hybrid architectures for domain-specific performance. Across domains including natural language processing, speech recognition, computer vision, and time-series forecasting, SSMs demonstrate state-of-the-art results in handling ultra-long sequences, outperforming Transformer-based models in both speed and memory utilization. Case studies highlight applications such as real-time speech synthesis and genomic sequence modeling, where SSMs reduce inference latency by up to 60% compared to traditional approaches. However, challenges persist in training dynamics, interpretability, and hardware-aware optimization. We conclude with a forward-looking analysis of SSMs' potential to redefine scalable deep learning, proposing directions for hybrid systems, theoretical guarantees, and broader adoption in resource-constrained environments.
Authors: Alexander Nedergaard, Pablo A. Morales
Abstract: Learning in environments with sparse rewards remains a fundamental challenge in reinforcement learning. Artificial curiosity addresses this limitation through intrinsic rewards to guide exploration, however, the precise formulation of these rewards has remained elusive. Ideally, such rewards should depend on the agent's information about the environment, remaining agnostic to its representation -- an invariance central to information geometry. Leveraging this, we show that information monotonicity and invariance under the agent-environment interaction uniquely constrains intrinsic rewards to strictly concave functions of the reciprocal occupancy. Requiring these rewards to yield a principled exploration-exploitation trade-off, via information geodesic interpolation on the occupancy manifold, effectively limits the candidates to those determined by a scalar parameter. Remarkably, special values of this parameter are found to correspond to count-based and maximum entropy exploration. This framework provides important constraints to the engineering of intrinsic rewards while integrating foundational exploration methods into a single, cohesive model.
Authors: Peyman Gholami, Hulya Seferoglu
Abstract: Although gossip and random walk-based learning algorithms are widely known for decentralized learning, there has been limited theoretical and experimental analysis to understand their relative performance for different graph topologies and data heterogeneity. We first design and analyze a random walk-based learning algorithm with multiple streams (walks), which we name asynchronous "Multi-Walk (MW)". We provide a convergence analysis for MW w.r.t iteration (computation), wall-clock time, and communication. We also present a convergence analysis for "Asynchronous Gossip", noting the lack of a comprehensive analysis of its convergence, along with the computation and communication overhead, in the literature. Our results show that MW has better convergence in terms of iterations as compared to Asynchronous Gossip in graphs with large diameters (e.g., cycles), while its relative performance, as compared to Asynchronous Gossip, depends on the number of walks and the data heterogeneity in graphs with small diameters (e.g., complete graphs). In wall-clock time analysis, we observe a linear speed-up with the number of walks and nodes in MW and Asynchronous Gossip, respectively. Finally, we show that MW outperforms Asynchronous Gossip in communication overhead, except in small-diameter topologies with extreme data heterogeneity. These results highlight the effectiveness of each algorithm in different graph topologies and data heterogeneity. Our codes are available for reproducibility.
Authors: Amirmohammad Mohammadi, Davelle Carreiro, Alexandra Van Dine, Joshua Peeples
Abstract: Parameter-efficient transfer learning (PETL) methods adapt large artificial neural networks to downstream tasks without fine-tuning the entire model. However, existing additive methods, such as adapters, sometimes struggle to capture distributional shifts in intermediate feature embeddings. We propose a novel histogram-based parameter-efficient tuning (HPT) technique that captures the statistics of the target domain and modulates the embeddings. Experimental results on three downstream passive sonar datasets (ShipsEar, DeepShip, Vessel Type Underwater Acoustic Data (VTUAD)) demonstrate that HPT outperforms conventional adapters. Notably, HPT achieves 91.8% vs. 89.8% accuracy on VTUAD. For active sonar imagery (Watertank, Turntable), HPT is competitive with other PETL methods. Furthermore, HPT yields feature representations closer to those of fully fine-tuned models. Overall, HPT balances parameter savings and provides a distribution-aware alternative to existing adapters and shows a promising direction for transfer learning in resource-constrained environments. The code is publicly available: https://github.com/Advanced-Vision-and-Learning-Lab/HLAST_DeepShip_ParameterEfficient.
URLs: https://github.com/Advanced-Vision-and-Learning-Lab/HLAST_DeepShip_ParameterEfficient.
Authors: Zayd M. K. Zuhri, Erland Hilman Fuadi, Alham Fikri Aji
Abstract: We introduce softpick, a rectified, not sum-to-one, drop-in replacement for softmax in transformer attention mechanisms that eliminates attention sink and massive activations. Our experiments with 340M and 1.8B parameter models demonstrate that softpick achieves 0\% sink rate consistently. The softpick transformers produce hidden states with significantly lower kurtosis and creates sparse attention maps. Quantized models using softpick outperform softmax on standard benchmarks, with a particularly pronounced advantage at lower bit precisions. Our analysis and discussion shows how softpick has the potential to open new possibilities for quantization, low-precision training, sparsity optimization, pruning, and interpretability. Our code: https://github.com/zaydzuhri/softpick-attention
Authors: Zhucong Li, Powei Chang, Jin Xiao, Zhijian Zhou, Qianyu He, Jiaqing Liang, Fenglei Cao, Xu Yinghui, Yuan Qi
Abstract: Although LLM-based agents are proven to master tool orchestration in scientific fields, particularly chemistry, their single-task performance remains limited by underlying tool constraints. To this end, we propose tool amplification, a novel paradigm that enhances the collective capabilities of specialized tools through optimized, dynamic coordination within individual tasks. Instantiating this paradigm, we introduce ChemAmp, a computationally lightweight framework that dynamically treats chemistry tools (e.g., UniMol2, Chemformer) as composable building-block agents. It constructs task-specialized super-agents that transcend atomic tool constraints with limited data ($\leq$10 samples). Our evaluations across four core chemistry tasks molecular design, molecule captioning, reaction prediction, and property prediction demonstrate that ChemAmp outperforms chemistry-specialized models, generalist LLMs, and agent systems with tool orchestration. Critically, this bottom-up construction strategy enables 94\% inference token cost reductions versus vanilla multi-agent systems.
Authors: Hanbing Liu, Lang Cao, Yuanyi Ren, Mengyu Zhou, Haoyu Dong, Xiaojun Ma, Shi Han, Dongmei Zhang
Abstract: Large language models (LLMs) show strong reasoning abilities but often produce unnecessarily long explanations that reduce efficiency. Although reinforcement learning (RL) has been used to improve reasoning, most methods focus on accuracy and rely on uniform length-based rewards that overlook the differing contributions of individual tokens, often harming correctness. We revisit length optimization in RL through the perspective of token significance. Observing that many chain-of-thought (CoT) tokens contribute little to the final answer, we introduce a significance-aware length reward that selectively penalizes insignificance tokens, reducing redundancy while preserving essential reasoning. We also propose a dynamic length reward that encourages more detailed reasoning early in training and gradually shifts toward conciseness as learning progresses. Integrating these components into standard policy optimization yields a framework that improves both reasoning efficiency and accuracy. Experiments across multiple benchmarks demonstrate substantial reductions in response length while preserving or improving correctness, highlighting the importance of modeling token significance for efficient LLM reasoning.
Authors: Paul Janson, Benjamin Therien, Quentin Anthony, Xiaolong Huang, Abhinav Moudgil, Eugene Belilovsky
Abstract: Learned optimizers have been an active research topic over the past decade, with increasing progress toward practical, general-purpose optimizers that can serve as drop-in replacements for widely used methods like Adam. However, recent advances such as VeLO, which was meta-trained for 4000 TPU-months, remain largely inaccessible to the broader community, in part due to their reliance on JAX and the absence of user-friendly packages for independently using the optimizers after meta-training. To address this gap, we introduce PyLO, a PyTorch-based library that brings learned optimizers to the remaining ~70% of machine learning community via the familiar torch.optim.Optimizer interface. Unlike prior work focused on limited-scale academic tasks, our emphasis is on applying learned optimization to real-world large-scale pre-training tasks. Our systems contribution includes CUDA-accelerated implementations of the small fc lopt(Metz et al., 2022a) and VeLO(Metz et al., 2022b) learned optimizers, achieving substantial performance gains, with training throughput on ViT-B/16 (batch size 32) increasing from 39.36 and 49.73 to 205.59 and 191.18 samples per second, respectively. PyLO has the versatility that allows us to easily combine learned optimizers with existing optimization tools such as learning rate schedules and weight decay. When doing so, we discover that learned optimizers can substantially benefit from it. Our code is available at https://github.com/Belilovsky-Lab/pylo
Authors: Ethan Mulle, Wei Kang, Qi Gong
Abstract: Deep neural networks are powerful tools for solving nonlinear problems in science and engineering, but training highly accurate models becomes challenging as problem complexity increases. Non-convex optimization and sensitivity to hyperparameters make consistent performance improvement difficult, and traditional approaches prioritize minimizing mean squared error while overlooking the $L^{\infty}$ norm error that is critical in safety-sensitive applications. To address these challenges, we present HiPreNets, a progressive framework for training high-precision neural networks through sequential residual refinements. Starting from an initial network, each stage trains a refinement network on the normalized residuals of the ensemble so far, systematically reducing both average and worst-case error. A key theme throughout the framework is concentrating training effort on high-error regions of the input domain, which we pursue through complementary techniques including loss function design, adaptive data sampling, localized patching, and boundary-aware training. We validate the framework on benchmark regression problems from the Feynman dataset, where it consistently outperforms standard fully connected networks and reported Kolmogorov-Arnold Networks results, with accuracy approaching machine precision depending on select problems. We further apply the framework to learning the flow map of a 20-dimensional power system ODE, which appears to be the highest dimensional problem studied using this class of multistage methods, achieving substantial reductions in both RMSE and $L^{\infty}$ norm error while enabling a surrogate that predicts system state $238\times$ faster than direct numerical simulation.
Authors: Pedro R. Pires, Gregorio F. Azevedo, Pietro L. Campos, Rafael T. Sereicikas, Tiago A. Almeida
Abstract: Multi-Armed Bandit (MAB) algorithms are widely used in recommender systems that require continuous, incremental learning. A core aspect of MABs is the exploration-exploitation trade-off: choosing between exploiting items likely to be enjoyed and exploring new ones to gather information. In contextual linear bandits, this trade-off is particularly central, as many variants share the same linear regression backbone and differ primarily in their exploration strategies. Despite its prevalent use, offline evaluation of MABs is increasingly recognized for its limitations in reliably assessing exploration behavior. This study conducts an extensive offline empirical comparison of several linear MABs. Strikingly, across over 90% of various datasets, a greedy linear model, with no type of exploration, consistently achieves top-tier performance, often outperforming or matching its exploratory counterparts. This observation is further corroborated by hyperparameter optimization, which consistently favors configurations that minimize exploration, suggesting that pure exploitation is the dominant strategy within these evaluation settings. Our results expose significant inadequacies in offline evaluation protocols for bandits, particularly concerning their capacity to reflect true exploratory efficacy. Consequently, this research underscores the urgent necessity for developing more robust assessment methodologies, guiding future investigations into alternative evaluation frameworks for interactive learning in recommender systems.
Authors: Peixuan Han, Adit Krishnan, Gerald Friedland, Jiaxuan You, Chris Kong
Abstract: Reinforcement learning with verifiable rewards has significantly advanced reasoning in large language models (LLMs), but such signals remain coarse, offering only binary correctness feedback. This limitation often results in inefficiencies, including overly verbose reasoning and high computational cost, while existing solutions often compromise accuracy. To address this, we introduce self-aligned reward (SAR), a self-guided signal that complements verifiable rewards to encourage both reasoning accuracy and efficiency. SAR is defined as the relative perplexity difference between an answer conditioned on the query and the standalone answer, thereby favoring responses that are concise and query-specific. Quantitative analysis reveals that SAR reliably distinguishes answer quality: concise, correct answers score higher than redundant ones, and partially correct answers score higher than entirely incorrect ones. Evaluation on 4 models across 7 benchmarks shows that integrating SAR with prevalent RL algorithms like PPO and GRPO improves accuracy by 4%, while reducing inference cost by 30%. Further analysis demonstrates that SAR achieves a Pareto-optimal trade-off between correctness and efficiency compared to reward signals based on length or self-confidence. We also show that SAR shortens responses while preserving advanced reasoning behaviors, demonstrating its ability to suppress unnecessary elaboration without losing critical reasoning. These results highlight the promise of self-aligned reward as a fine-grained complement to verifiable rewards, paving the way for more efficient and effective LLM training.
Authors: Aksheytha Chelikavada, Casey C. Bennett
Abstract: Financial bubbles often arrive without much warning, but create long-lasting economic effects. For example, during the dot-com bubble, innovative technologies created market disruptions through excitement for a promised bright future. Such technologies originated from research where scientists had developed them for years prior to their entry into the markets. That raises a question on the possibility of analyzing scientific publishing data (e.g. citation networks) leading up to a bubble for signals that may forecast the rise and fall of similar future bubbles. To that end, we utilized temporal SNAs to detect possible relationships between the publication citation networks of scientists and financial market data during two modern eras of rapidly shifting technology: 1) dot-com era from 1994 to 2001 and 2) AI era from 2017 to 2024. Results showed that the patterns from the dot-com era (which did end in a bubble) did not definitively predict the rise and fall of an AI bubble. While yearly citation networks reflected possible changes in publishing behavior of scientists between the two eras, there was a subset of AI era scientists whose publication influence patterns mirrored those during the dot-com era. Upon further analysis using multiple analysis techniques (LSTM, KNN, AR X/GARCH), the data seems to suggest two possibilities for the AI era: unprecedented form of financial bubble unseen or that no bubble exists. In conclusion, our findings imply that the patterns present in the dot-com era do not effectively translate in such a manner to apply them to the AI market.
Authors: Sharan Sahu, Martin T. Wells
Abstract: Reinforcement Learning with Human Feedback (RLHF) has become crucial for aligning Large Language Models (LLMs) with human intent. However, existing offline RLHF approaches suffer from overoptimization, where language models degrade by overfitting inaccuracies and drifting from preferred behaviors observed during training. Distributionally robust optimization (DRO) is a natural solution, but existing DRO-DPO methods are sample-inefficient, ignore heterogeneous preferences, and lean on brittle heuristics. We introduce \emph{DRO-REBEL}, a family of robust online REBEL updates built on type-$p$ Wasserstein, Kullback-Leibler (KL), and $\chi^2$ ambiguity sets. Strong duality reduces each update to a relative-reward regression, retaining REBEL's scalability without PPO-style clipping or value networks. Under linear rewards, log-linear policies, and a standard coverage condition, we prove $\widetilde{O}(\sqrt{d/n})$ bounds on squared parameter error, with sharper constants than prior DRO-DPO analyses, and give the first parametric $\widetilde{O}(d/n)$ rate for DRO-based alignment under preference shift, matching non-robust RLHF in benign regimes. Each divergence yields a tractable SGD-based algorithm: gradient regularization for Wasserstein, importance weighting for KL, and a 1-D dual solve for $\chi^2$. On Emotion Alignment, the ArmoRM multi-objective benchmark, and HH-Alignment, DRO-REBEL outperforms prior robust and non-robust baselines across unseen preference mixtures, model sizes, and dataset scales.
Authors: Haotian Guo, Hui Liu
Abstract: The de novo generation of drug-like molecules capable of inducing desirable phenotypic changes is receiving increasing attention. However, previous methods predominantly rely on expression profiles to guide molecule generation, but overlook the perturbative effect of the molecules on cellular contexts. To overcome this limitation, we propose SmilesGEN, a novel generative model based on variational autoencoder (VAE) architecture to generate molecules with potential therapeutic effects. SmilesGEN integrates a pre-trained drug VAE (SmilesNet) with an expression profile VAE (ProfileNet), jointly modeling the interplay between drug perturbations and transcriptional responses in a common latent space. Specifically, ProfileNet is imposed to reconstruct pre-treatment expression profiles when eliminating drug-induced perturbations in the latent space, while SmilesNet is informed by desired expression profiles to generate drug-like molecules. Our empirical experiments demonstrate that SmilesGEN outperforms current state-of-the-art models in generating molecules with higher degree of validity, uniqueness, novelty, as well as higher Tanimoto similarity to known ligands targeting the relevant proteins. Moreover, we evaluate SmilesGEN for scaffold-based molecule optimization and generation of therapeutic agents, and confirmed its superior performance in generating molecules with higher similarity to approved drugs. SmilesGEN establishes a robust framework that leverages gene signatures to generate drug-like molecules that hold promising potential to induce desirable cellular phenotypic changes. The source code and datasets are available at: https://github.com/hliulab/SmilesGEN.
Authors: Zelin Tan, Hejia Geng, Xiaohang Yu, Mulei Zhang, Guancheng Wan, Yifan Zhou, Qiang He, Xiangyuan Xue, Heng Zhou, Yutao Fan, Zhongzhi Li, Zaibin Zhang, Guibin Zhang, Chen Zhang, Zhenfei Yin, Philip Torr, Lei Bai
Abstract: While scaling laws for large language models (LLMs) during pre-training have been extensively studied, their behavior under reinforcement learning (RL) post-training remains largely unexplored. This paper presents a systematic empirical investigation of scaling behaviors in RL-based post-training, with a particular focus on mathematical reasoning. Based on a set of experiments across the full Qwen2.5 dense model series (0.5B to 72B), we characterize how model scale, data volume, and computational budget interact to shape performance. Our analysis leads to four key findings: 1. Larger models consistently exhibit superior learning efficiency on both compute and data metrics. 2. The relationship between test loss, compute, and data can be modeled by a predictive power-law which is robust across both base and instruction-tuned models. 3. Although larger models exhibit higher learning efficiency, the analytical learning efficiency term k(N) in the power-law reveals a latent saturation trend in learning efficiency as model size continues to increase. 4. In data-constrained regimes, repeated reuse of high-quality data proves highly effective, as final performance is primarily governed by the total number of optimization steps rather than the uniqueness of samples. Collectively, these results provide a principled foundation and practical guidelines for efficiently scaling the reasoning capabilities of LLMs through RL post-training.
Authors: Jinhui Bai, Andreas Christmann, Lei Shi
Abstract: In this paper, we propose a novel kernel stochastic gradient descent (SGD) algorithm for large-scale supervised learning with general losses. Compared to traditional kernel SGD, our algorithm improves efficiency and scalability through an innovative regularization strategy. By leveraging the infinite series expansion of spherical radial basis functions, this strategy projects the stochastic gradient onto a finite-dimensional hypothesis space, which is adaptively scaled according to the bias-variance trade-off, thereby enhancing generalization performance. Based on a new estimation of the spectral structure of the kernel-induced covariance operator, we develop an analytical framework that unifies optimization and generalization analyses. We prove that both the last iterate and the suffix average converge at minimax-optimal rates, and we further establish optimal strong convergence in the reproducing kernel Hilbert space. Our framework accommodates a broad class of classical loss functions, including least-squares, Huber, and logistic losses. Moreover, the proposed algorithm significantly reduces computational complexity and achieves optimal storage complexity by incorporating coordinate-wise updates from linear SGD, thereby avoiding the costly pairwise operations typical of kernel SGD and enabling efficient processing of streaming data. Finally, extensive numerical experiments demonstrate the efficiency of our approach.
Authors: Tian Qin, Felix Bai, Ting-Yao Hu, Raviteja Vemulapalli, Hema Swetha Koppula, Zhiyang Xu, Bowen Jin, Mert Cemri, Jiarui Lu, Zirui Wang, Meng Cao
Abstract: Human decision-making often involves constrained optimization. As LLM agents are deployed to assist with real-world tasks like travel planning, shopping, and scheduling, they must mirror this capability. We introduce COMPASS, a benchmark that evaluates whether LLM agents can perform constrained optimization in realistic travel planning settings. To success in these tasks, agents must engage in multi-turn conversations with user to gather task information as well as use tools to gather information from the database. Then agents must propose a solution that not only satisfies hard constraints but also optimizes user's utility objective. Evaluating state-of-the-art models, we reveal a significant feasible-optimal gap: while models achieve 70-90% feasibility (constraint satisfaction), they reach only 20-60% optimality (utility optimization). Our analysis shows that tool use is not the bottleneck. Instead, the core limitation is insufficient exploration of the search space, with success strongly correlating with information gathered. Coding agents show a promising approach to mitigate this gap. Together, COMPASS provides a testbed for developing LLM agents that can truly mirror human decision-making by both satisfying constraints and optimizing objectives.
Authors: Xiaoyun Zhang, Xiaojian Yuan, Di Huang, Wang You, Chen Hu, Jingqing Ruan, Ai Jian, Kejiang Chen, Xing Hu
Abstract: Reasoning ability has become a defining capability of Large Language Models (LLMs), with Reinforcement Learning with Verifiable Rewards (RLVR) emerging as a key paradigm to enhance it. However, RLVR training often suffers from policy entropy collapse, where the policy becomes overly deterministic, hindering exploration and limiting reasoning performance. While entropy regularization is a common remedy, its effectiveness is highly sensitive to the fixed coefficient, making it unstable across tasks and models. In this work, we revisit entropy regularization in RLVR and argue that its potential has been largely underestimated. Our analysis shows that (i) tasks of varying difficulty demand distinct exploration intensities, and (ii) balanced exploration may require the policy entropy to be maintained within a moderate range below its initial level. Therefore, we propose Adaptive Entropy Regularization (AER)--a framework that dynamically balances exploration and exploitation via three components: difficulty-aware coefficient allocation, initial-anchored target entropy, and dynamic global coefficient adjustment. Experiments on multiple mathematical reasoning benchmarks show that AER consistently outperforms baselines, improving both reasoning accuracy and exploration capability.
Authors: Saraf Anzum Shreya, MD. Abu Ismail Siddique, Sharaf Tasnim
Abstract: Brain tumors are a challenging problem in neuro-oncology, where early and precise diagnosis is important for successful treatment. Deep learning-based brain tumor classification methods often rely on heavy data augmentation which can limit generalization and trust in clinical applications. In this paper, we propose a double-backbone network integrating VGG16 and Xception with a Frequency-Gated Attention (FGA) Block to capture complementary local and global features. Our model achieves highly competitive performance without augmentation which demonstrates robustness to variably sized and distributed datasets. For further transparency, Grad-CAM is integrated to visualize the tumor regions based on which the model is giving prediction, bridging the gap between model prediction and clinical interpretability. The proposed framework achieves 99.24% accuracy on the 7K-DS dataset for the 4-class setting, along with 98.68% and 99.85% in the 3-class and 2-class settings, respectively. On the independent 3K-DS dataset, the model generalizes with 95.77% accuracy, outperforming several baseline methods under the same experimental setting. To further support clinical usability, we developed a graphical user interface (GUI) that provides real-time classification and Grad-CAM-based tumor localization. These findings suggest that augmentation-free, interpretable, and deployable deep learning models such as DB-FGA-Net hold strong potential for reliable clinical translation in brain tumor diagnosis.
Authors: Aki Rehn, Linzh Zhao, Mikko A. Heikkil\"a, Antti Honkela
Abstract: Differentially private (DP) transfer learning, i.e., fine-tuning a pretrained model on private data, is the current state-of-the-art approach for training large models under privacy constraints. We focus on two key hyperparameters in this setting: the clipping bound $C$ and batch size $B$. We show a clear mismatch between the current theoretical understanding of how to choose an optimal $C$ (stronger privacy requires smaller $C$) and empirical outcomes (larger $C$ performs better under strong privacy), caused by changes in the gradient distributions. Assuming a limited compute budget (fixed epochs), we demonstrate that the existing heuristics for tuning $B$ do not work, while cumulative DP noise better explains whether smaller or larger batches perform better. We also highlight how the common practice of using a single $(C,B)$ setting across tasks can lead to suboptimal performance. We find that performance drops especially when moving between loose and tight privacy and between plentiful and limited compute, which we explain by analyzing clipping as a form of gradient re-weighting and examining cumulative DP noise.
Authors: Fardin Ganjkhanloo, Emmett Springer, Erik H. Hoyer, Daniel L. Young, Kimia Ghobadi
Abstract: Risk assessment tools in healthcare commonly employ point-based scoring systems that map patients to ordinal risk categories via thresholds. While electronic health record (EHR) data presents opportunities for data-driven optimization of these tools, two fundamental challenges impede standard supervised learning: (1) labels are often available only for extreme risk categories due to intervention-censored outcomes, and (2) misclassification cost is asymmetric and increases with ordinal distance. We propose a mixed-integer programming (MIP) framework that jointly optimizes scoring weights and category thresholds in the face of these challenges. Our approach prevents label-scarce category collapse via threshold constraints, and utilizes an asymmetric, distance-aware objective. The MIP framework supports governance constraints, including sign restrictions, sparsity, and minimal modifications to incumbent tools, ensuring practical deployability in clinical workflows. We further develop a continuous relaxation of the MIP problem to provide warm-start solutions for more efficient MIP optimization. We apply the proposed score optimization framework to a case study of inpatient falls risk assessment using the Johns Hopkins Fall Risk Assessment Tool.
Authors: Mohammadsajad Alipour, Mohammad Mohammadi Amiri
Abstract: Federated learning (FL) has emerged as a promising paradigm for decentralized model training, enabling multiple clients to collaboratively learn a shared model without exchanging their local data. However, the decentralized nature of FL also introduces vulnerabilities, as malicious clients can compromise or manipulate the training process. In this work, we introduce dictator clients, a novel, well-defined, and analytically tractable class of malicious participants capable of entirely erasing the contributions of all other clients from the server model, while preserving their own. We propose concrete attack strategies that empower such clients and systematically analyze their effects on the learning process. Furthermore, we explore complex scenarios involving multiple dictator clients, including cases where they collaborate, act independently, or form an alliance in order to ultimately betray one another. For each of these settings, we provide a theoretical analysis of their impact on the global model's convergence. Our theoretical algorithms and findings about the complex scenarios including multiple dictator clients are further supported by empirical evaluations on both computer vision and natural language processing benchmarks.
Authors: Chenlong Yin, Zeyang Sha, Shiwen Cui, Changhua Meng, Zechao Li
Abstract: Enhancing the reasoning capabilities of Large Language Models (LLMs) is a key strategy for building Agents that "think then act." However, recent observations, like OpenAI's o3, suggest a paradox: stronger reasoning often coincides with increased hallucination, yet no prior work has systematically examined whether reasoning enhancement itself causes tool hallucination. To address this gap, we pose the central question: Does strengthening reasoning increase tool hallucination? To answer this, we introduce SimpleToolHalluBench, a diagnostic benchmark measuring tool hallucination in two failure modes: (i) no tool available, and (ii) only distractor tools available. Through controlled experiments, we establish three key findings. First, we demonstrate a causal relationship: progressively enhancing reasoning through RL increases tool hallucination proportionally with task performance gains. Second, this effect transcends overfitting - training on non-tool tasks (e.g., mathematics) still amplifies subsequent tool hallucination. Third, the effect is method-agnostic, appearing when reasoning is instilled via supervised fine-tuning and when it is merely elicited at inference by switching from direct answers to step-by-step thinking. We also evaluate mitigation strategies including Prompt Engineering and Direct Preference Optimization (DPO), revealing a fundamental reliability-capability trade-off: reducing hallucination consistently degrades utility. Mechanistically, Reasoning RL disproportionately collapses tool-reliability-related representations, and hallucinations surface as amplified divergences concentrated in late-layer residual streams. These findings reveal that current reasoning enhancement methods inherently amplify tool hallucination, highlighting the need for new training objectives that jointly optimize for capability and reliability.
Authors: Vishnu Sarukkai, Asanshay Gupta, James Hong, Micha\"el Gharbi, Kayvon Fatahalian
Abstract: Deploying LLM agents at scale typically requires choosing between quality and cost. Existing cost-reduction approaches fail to preserve agility: the ability to iterate rapidly without human time bottlenecks. Prompt engineering is brittle and slows iteration, while fine-tuning requires multi-day training and commitment to fixed designs; both are impractical for iterative workflows and time-sensitive batch jobs. We demonstrate that established inference-time techniques--dynamic in-context learning and self-consistency cascades--can be leveraged to shift the cost-accuracy Pareto frontier while preserving agility. Practitioners run the teacher on a small task subset to collect demonstrations, then immediately deploy a cheaper student on the remainder. At each step, the system retrieves relevant teacher demonstrations as in-context examples. When multiple student samples agree, we proceed; when they diverge, we fall back to the teacher. This requires no prompt engineering or training. On ALFWorld, we match teacher accuracy at 2.5x lower cost (0.059 to 0.024 per episode). On AppWorld, we achieve 3.5x cost reduction while recovering 79% of teacher accuracy. Our empirical analyses provide guidance on key design choices: teacher database size, demonstration set size, retrieval strategy, and cascade thresholds. These analyses highlight inference-time levers for navigating cost-performance tradeoffs without sacrificing human development speed.
Authors: Andrew S. Cassidy, Guillaume Garreau, Jay Sivagnaname, Mike Grassi, Bernard Brezzo, John V. Arthur, Dharmendra S. Modha
Abstract: We show for invertible problems that transform data from a source domain (for example, Logic Condition Tables (LCTs)) to a destination domain (for example, Hardware Description Language (HDL) code), an approach of using Large Language Models (LLMs) as a lossless encoder from source to destination followed by as a lossless decoder back to the source, comparable to lossless compression in information theory, can mitigate most of the LLM drawbacks of hallucinations and omissions. Specifically, using LCTs as inputs, we generate the full HDL for a two-dimensional network-on-chip router (13 units, 1500-2000 lines of code) using seven different LLMs, reconstruct the LCTs from the auto-generated HDL, and compare the original and reconstructed LCTs. This approach yields significant productivity improvements, not only confirming correctly generated LLM logic and detecting incorrectly generated LLM logic but also assisting developers in finding design specification errors.
Authors: Th\'eo A. Neukomm, Zlatko Jon\v{c}ev, Philippe Schwaller
Abstract: Chemical reaction mechanisms are the foundation of how chemists evaluate reactivity and feasibility, yet current Computer-Assisted Synthesis Planning (CASP) systems operate without this mechanistic reasoning. We introduce a computational framework that teaches language models to predict reaction mechanisms through arrow-pushing formalism, a century-old notation that tracks electron flow while enforcing conservation of mass and charge. This mechanistic understanding enables three capabilities that are difficult or impossible with current methods: post-hoc validation of CASP proposals by reconstructing physically plausible electron pathways, holistic atom-to-atom mapping that tracks all atoms including hydrogens, and extraction of catalyst-aware reaction templates that distinguish recycled catalysts from spectator species. Central to our approach is MechSMILES, a compact textual format encoding molecular structure and electron flow through three arrow types, designed within a Python-based environment that enforces conservation laws and eliminates the possibility of atom hallucination. We trained and benchmarked models on four mechanism prediction tasks of increasing complexity using the main mechanistic datasets in the literature. On our most challenging task, predicting complete mechanisms given only reactants, conditions, and the desired product, our models achieve 93.2\% and 73.3\% pathway retrieval on the FlowER and mech-USPTO-31k datasets respectively, with top-3 retrieval reaching 97.6\% and 86.5\%. Furthermore, the framework rapidly learns new reaction classes, with strong mechanistic predictions for ozonolysis and Suzuki cross-coupling emerging from as few as 40 training examples each. By grounding predictions in physically meaningful electron movements, this work provides an architecture-agnostic, open-source foundation for more explainable and chemically valid CASP.
Authors: Emily Jin, Andrei Cristian Nica, Mikhail Galkin, Jarrid Rector-Brooks, Kin Long Kelvin Lee, Santiago Miret, Frances H. Arnold, Michael Bronstein, Avishek Joey Bose, Alexander Tong, Cheng-Hao Liu
Abstract: Accurately predicting experimentally realizable 3D molecular crystal structures from their 2D chemical graphs is a long-standing open challenge in computational chemistry called crystal structure prediction (CSP). Efficiently solving this problem has implications ranging from pharmaceuticals to organic semiconductors, as crystal packing directly governs the physical and chemical properties of organic solids. In this paper, we introduce OXtal, a large-scale 100M parameter all-atom diffusion model that directly learns the conditional joint distribution over intramolecular conformations and periodic packing. To efficiently scale OXtal, we abandon explicit equivariant architectures imposing inductive bias arising from crystal symmetries in favor of data augmentation strategies. We further propose a novel crystallization-inspired lattice-free training scheme, Stoichiometric Stochastic Shell Sampling ($S^4$), that efficiently captures long-range interactions while sidestepping explicit lattice parametrization -- thus enabling more scalable architectural choices at all-atom resolution. By leveraging a large dataset of 600K experimentally validated crystal structures (including rigid and flexible molecules, co-crystals, and solvates), OXtal achieves orders-of-magnitude improvements over prior ab initio machine learning CSP methods, while remaining orders of magnitude cheaper than traditional quantum-chemical approaches. Specifically, OXtal recovers experimental structures with conformer $\text{RMSD}_1<0.5$ {\AA} and attains over 80\% packing similarity rate, demonstrating its ability to model both thermodynamic and kinetic regularities of molecular crystallization.
Authors: Jucheng Shen, Gaurav Sarkar, Yeonju Ro, Sharath Nittur Sridhar, Zhangyang Wang, Aditya Akella, Souvik Kundu
Abstract: We present CadLLM, a training-free method to accelerate the inference throughput of diffusion-based LLMs (dLLMs). We first investigate the dynamic nature of token unmasking confidence across blocks and steps. Based on this observation, we present a lightweight adaptive approach that controls the generation block size, step size, and threshold based on the average confidence of unmasked tokens. We further reduce softmax overhead by dynamically leveraging a subset of the vocabulary to regulate sampling breadth. CadLLM is a plug-and-play, model-agnostic method compatible with KV-cache-based dLLMs. Extensive experiments on four popular tasks demonstrate that CadLLM yields up to 1.1-2.28x throughput improvement over the state-of-the-art baseline with competitive accuracy.
Authors: Sonal Prabhune, Balaji Padmanabhan, Kaushik Dutta
Abstract: Large Language Models (LLMs) are increasingly deployed in business-critical domains such as finance, education, healthcare, and customer support, where users expect consistent and reliable recommendations. Yet LLMs often exhibit variability when prompts are phrased with minor differences, even when semantically equivalent. Such inconsistency undermines trust, complicates compliance, and disrupts user experience. While personalization is desirable in certain contexts, many enterprise scenarios, such as HR onboarding, customer support, or policy disclosure, require invariant information delivery regardless of phrasing or prior conversational history. Existing approaches, including retrieval-augmented generation (RAG) and temperature tuning, improve factuality or reduce stochasticity, but cannot guarantee stability across equivalent prompts. In this paper, we propose a reinforcement learning framework based on Group Relative Policy Optimization (GRPO) to directly optimize for consistency. Unlike prior applications of GRPO, which have been limited to reasoning and code generation, we adapt GRPO to enforce the stability of information content across groups of semantically equivalent prompts. We introduce entropy-based helpfulness and stability rewards, treating prompt variants as groups and resetting conversational context to isolate phrasing effects. Experiments on investment and job recommendation tasks show that our GRPO-fine-tuned model reduces variability compared to the baseline LLM model. To our knowledge, this is a novel application of GRPO for aligning LLMs toward information consistency, reframing variability not as an acceptable feature of generative diversity, but as a correctable flaw in enterprise deployments.
Authors: Bhrij Patel, Davide Belli, Amir Jalalirad, Maximilian Arnold, Aleksandr Ermolov, Bence Major
Abstract: Function calling agents powered by Large Language Models (LLMs) select external tools to automate complex tasks. On-device agents typically use a retrieval module to select relevant tools, improving performance and reducing context length. However, existing retrieval methods rely on static and limited inputs, failing to capture multi-step tool dependencies and evolving task context. This limitation often introduces irrelevant tools that mislead the agent, degrading efficiency and accuracy. We propose Dynamic Tool Dependency Retrieval (DTDR), a lightweight retrieval method that conditions on both the initial query and the evolving tool calling plan. DTDR models tool dependencies from function calling demonstrations, enabling adaptive retrieval as plans unfold. We benchmark DTDR against state-of-the-art retrieval methods across multiple datasets and LLM backbones, evaluating retrieval precision, downstream task accuracy, and computational efficiency. Additionally, we explore strategies to integrate retrieved tools into prompts. Our results show that DTDR improves function calling success rates between $23\%$ and $104\%$ compared to state-of-the-art static retrievers.
Authors: Zijian Zhao, Yitong Shang, Sen Li
Abstract: Accurate traffic prediction is essential for Intelligent Transportation Systems, including ride-hailing, urban road planning, and vehicle fleet management. However, due to significant privacy concerns surrounding traffic data, most existing methods rely on local training, resulting in data silos and limited knowledge sharing. Federated Learning (FL) offers an efficient solution through privacy-preserving collaborative training; however, standard FL struggles with the non-independent and identically distributed (non-IID) problem among clients. This challenge has led to the emergence of Personalized Federated Learning (PFL) as a promising paradigm. Nevertheless, current PFL frameworks require further adaptation for traffic prediction tasks, such as specialized graph feature engineering, data processing, and network architecture design. A notable limitation of many prior studies is their reliance on hyper-parameter optimization across datasets-information that is often unavailable in real-world scenarios-thus impeding practical deployment. To address this challenge, we propose AutoFed, a novel PFL framework for traffic prediction that eliminates the need for manual hyper-parameter tuning. Inspired by prompt learning, AutoFed introduces a federated representor that employs a client-aligned adapter to distill local data into a compact, globally shared prompt matrix. This prompt then conditions a personalized predictor, allowing each client to benefit from cross-client knowledge while maintaining local specificity. Extensive experiments on real-world datasets demonstrate that AutoFed consistently achieves superior performance across diverse scenarios. The code of this paper is provided at https://github.com/RS2002/AutoFed .
Authors: Runsong Zhao, Shilei Liu, Jiwei Tang, Langming Liu, Haibin Chen, Weidong Zhang, Yujin Yuan, Tong Xiao, Jingbo Zhu, Wenbo Su, Bo Zheng
Abstract: The quadratic complexity and indefinitely growing key-value (KV) cache of standard Transformers pose a major barrier to long-context processing. To overcome this, we introduce the Collaborative Memory Transformer (CoMeT), a novel architecture that enables LLMs to handle arbitrarily long sequences with constant memory usage and linear time complexity. Designed as an efficient, plug-in module, CoMeT can be integrated into pre-trained models with only minimal fine-tuning. It operates on sequential data chunks, using a dual-memory system to manage context: a temporary memory on a FIFO queue for recent events, and a global memory with a gated update rule for long-range dependencies. These memories then act as a dynamic soft prompt for the next chunk. To enable efficient fine-tuning on extremely long contexts, we introduce a novel layer-level pipeline parallelism strategy. The effectiveness of our approach is remarkable: a model equipped with CoMeT and fine-tuned on 32k contexts can accurately retrieve a passkey from any position within a 1M token sequence. On the SCROLLS benchmark, CoMeT surpasses other efficient methods and achieves performance comparable to a full-attention baseline on summarization tasks. Its practical effectiveness is further validated on real-world agent and user behavior QA tasks. The code is available at: https://github.com/LivingFutureLab/Comet
Authors: Peter Lalor, Ayush Panigrahy, Alex Hagen
Abstract: Training machine learning models for radioisotope identification using gamma spectroscopy remains an elusive challenge for many practical applications, largely stemming from the difficulty of acquiring and labeling large, diverse experimental datasets. Simulations can mitigate this challenge, but the accuracy of models trained on simulated data can deteriorate substantially when deployed to an out-of-distribution operational environment. In this study, we demonstrate that unsupervised domain adaptation (UDA) can improve the ability of a model trained on synthetic data to generalize to a new testing domain, provided unlabeled data from the target domain is available. Conventional supervised techniques are unable to utilize this data because the absence of isotope labels precludes defining a supervised classification loss. We compare a range of different UDA techniques, finding that feature alignment strategies, particularly via maximum mean discrepancy (MMD) minimization or domain-adversarial training, yield the most consistent improvement to testing scores. For instance, using a custom transformer-based neural network, we achieve a testing accuracy of $0.904 \pm 0.022$ on an experimental LaBr$_3$ test set after performing unsupervised feature alignment via MMD minimization, compared to $0.754 \pm 0.014$ before alignment. Overall, our results highlight the potential of using UDA to adapt a radioisotope classifier trained on synthetic data for real-world deployment.
Authors: Indranil Halder, Annesya Banerjee, Cengiz Pehlevan
Abstract: Adversarial attacks can reliably steer safety-aligned large language models toward unsafe behavior. Empirically, we find that strong adversarial prompt-injection attacks can amplify attack success rate from the slow polynomial growth observed without injection to exponential growth with the number of inference-time samples. We first identify a minimal statistical mechanism for these two regimes by giving a small set of assumptions on the distribution of safe generation across contexts under which both scaling laws follow. To explain this phenomenon further, we propose a theoretical generative model of proxy language in terms of a spin-glass system operating in a replica-symmetry-breaking regime, where generations are drawn from the associated Gibbs measure and a subset of low-energy, size-biased clusters is designated unsafe. We point out how this model naturally realizes the minimal assumptions. Short injected prompts correspond to a weak magnetic field aligned towards unsafe cluster centers and yield a power-law scaling of attack success rate with the number of inference-time samples, while long injected prompts, i.e., strong magnetic field, yield exponential scaling. We derive these behaviors analytically and find qualitatively similar trends in large language models.
Authors: Yuval Ran-Milo
Abstract: Transformers often display an attention sink: probability mass concentrates on a fixed, content-agnostic position. Are sinks a byproduct of the optimization/training regime? Or are they sometimes functionally necessary in softmax Transformers? We prove that, in some settings, it is the latter: computing a simple trigger-conditional behavior necessarily induces a sink in softmax self-attention models. Our results formalize a familiar intuition: normalization over a probability simplex must force attention to collapse onto a stable anchor to realize a default state (e.g., when the model needs to ignore the input). We instantiate this with a concrete task: when a designated trigger token appears, the model must return the average of all preceding token representations, and otherwise output zero, a task which mirrors the functionality of attention heads in the wild (Barbero et al., 2025; Guo et al., 2024). We also prove that non-normalized ReLU attention can solve the same task without any sink, confirming that the normalization constraint is the fundamental driver of sink behavior. Experiments validate our predictions and demonstrate they extend beyond the theoretically analyzed setting: softmax models develop strong sinks while ReLU attention eliminates them in both single-head and multi-head variants.
Authors: Mahmoud Elhadidy, Roshan M. D'Souza, Amirhossein Arzani
Abstract: Scientific machine learning is increasingly used to build surrogate models, yet most models are trained under a restrictive assumption in which future data follow the same distribution as the training set. In practice, new experimental conditions or simulation regimes may differ significantly, requiring extrapolation and model updates without re-access to prior data. This creates a need for continual learning (CL) frameworks that can adapt to distribution shifts while preventing catastrophic forgetting. Such challenges are pronounced in fluid dynamics, where changes in geometry, boundary conditions, or flow regimes induce non-trivial changes to the solution. Here, we introduce a new architecture-based approach (SLE-FNO) combining a Single-Layer Extension (SLE) with the Fourier Neural Operator (FNO) to support efficient CL. SLE-FNO was compared with a range of established CL methods, including Elastic Weight Consolidation (EWC), Learning without Forgetting (LwF), replay-based approaches, Orthogonal Gradient Descent (OGD), Gradient Episodic Memory (GEM), PiggyBack, and Low-Rank Adaptation (LoRA), within a spatial field-to-field regression setting. The models were trained to map transient concentration fields to time-averaged wall shear stress (TAWSS) in pulsatile aneurysmal blood flow. Tasks were derived from 230 computational fluid dynamics simulations grouped into four sequential and out-of-distribution configurations. Results show that replay-based methods and architecture-based approaches (PiggyBack, LoRA, and SLE-FNO) achieve the best retention, with SLE-FNO providing the strongest overall balance between plasticity and stability, achieving accuracy with zero forgetting and minimal additional parameters. Our findings highlight key differences between CL algorithms and introduce SLE-FNO as a promising strategy for adapting baseline models when extrapolation is required.
Authors: Mirko Degli Esposti
Abstract: Maximum entropy (MaxEnt) modelling provides a principled framework for generating synthetic populations from aggregate census data, without access to individual-level microdata. The bottleneck of exact-enumeration approaches is expectation computation by explicit summation over the full tuple space $\cX$, which becomes infeasible for more than $K \approx 20$ categorical attributes; sampling-based alternatives exist but rely on Metropolis-type schemes that require proposal tuning and rejection steps. We propose \emph{GibbsPCDSolver}, a stochastic replacement for this computation based on Persistent Contrastive Divergence (PCD): a persistent pool of $N$ synthetic individuals is updated by Gibbs sweeps at each gradient step, providing a stochastic approximation of the model expectations without ever materialising $\cX$. We validate the approach on controlled benchmarks and on \emph{Syn-ISTAT}, a $K{=}15$ Italian demographic benchmark with analytically exact marginal targets derived from ISTAT-inspired conditional probability tables. Scaling experiments across $K \in \{12, 20, 30, 40, 50\}$ confirm that GibbsPCDSolver maintains $\MRE \in [0.010, 0.018]$ while $|\cX|$ grows eighteen orders of magnitude, with runtime scaling as $O(K)$ rather than $O(|\cX|)$. On Syn-ISTAT, GibbsPCDSolver reaches $\MRE{=}0.03$ on training constraints and -- crucially -- produces populations with effective sample size $\Neff = N$ versus $\Neff \approx 0.012\,N$ for generalised raking, an $86.8{\times}$ diversity advantage that is essential for agent-based urban simulations.
Authors: Alex Al\`i Maleknia, Yuzuru Sato
Abstract: Vanishing gradients and overfitting are central problems in machine learning, yet are typically analyzed in asymptotic regimes that obscure their dynamical origins. Here we provide a dynamical description of learning in multi-layer perceptrons (MLPs) via a minimal model inspired by Fukumizu and Amari. We show that training dynamics traverse plateau and near-optimal regions, both organized by saddle structures, before converging to an overfitting regime. Under suitable conditions on the data, this regime collapses to a single attractor modulo symmetry. Furthermore, for finite noisy datasets, convergence to the theoretical optimum is impossible, and the dynamics necessarily settle into an overfitting solution.
Authors: William Merrill, Yanhong Li, Tyler Romero, Anej Svete, Caia Costello, Pradeep Dasigi, Dirk Groeneveld, David Heineman, Bailey Kuehl, Nathan Lambert, Chuan Li, Kyle Lo, Saumya Malik, DJ Matusz, Benjamin Minixhofer, Jacob Morrison, Luca Soldaini, Finbarr Timbers, Pete Walsh, Noah A. Smith, Hannaneh Hajishirzi, Ashish Sabharwal
Abstract: Recent work has demonstrated the potential of non-transformer language models, especially linear recurrent neural networks (RNNs) and hybrid models that mix recurrence and attention. Yet there is no consensus on whether the potential benefits of these new architectures justify the risk and effort of scaling them up. To address this, we provide evidence for the advantages of hybrid models over pure transformers on several fronts. First, theoretically, we show that hybrid models do not merely inherit the expressivity of transformers and linear RNNs, but can express tasks beyond both, such as code execution. Putting this theory to practice, we train Olmo Hybrid, a 7B-parameter model largely comparable to Olmo 3 7B but with the sliding window layers replaced by Gated DeltaNet layers. We show that Olmo Hybrid outperforms Olmo 3 across standard pretraining and mid-training evaluations, demonstrating the benefit of hybrid models in a controlled, large-scale setting. We find that the hybrid model scales significantly more efficiently than the transformer, explaining its higher performance. However, its unclear why greater expressivity on specific formal problems should result in better scaling or superior performance on downstream tasks unrelated to those problems. To explain this apparent gap, we return to theory and argue why increased expressivity should translate to better scaling efficiency, completing the loop. Overall, our results suggest that hybrid models mixing attention and recurrent layers are a powerful extension to the language modeling paradigm: not merely to reduce memory during inference, but as a fundamental way to obtain more expressive models that scale better during pretraining.
Authors: Nida Zamir, I-Hong Hou
Abstract: This paper investigates the Restless Multi-Armed Bandit (RMAB) framework under individual penalty constraints to address resource allocation challenges in dynamic wireless networked environments. Unlike conventional RMAB models, our model allows each user (arm) to have distinct and stringent performance constraints, such as energy limits, activation limits, or age of information minimums, enabling the capture of diverse objectives including fairness and efficiency. To find the optimal resource allocation policy, we propose a new Penalty-Optimal Whittle (POW) index policy. The POW index of an user only depends on the user's transition kernel and penalty constraints, and remains invariable to system-wide features such as the number of users present and the amount of resource available. This makes it computationally tractable to calculate the POW indices offline without any need for online adaptation. Moreover, we theoretically prove that the POW index policy is asymptotically optimal while satisfying all individual penalty constraints. We also introduce a deep reinforcement learning algorithm to efficiently learn the POW index on the fly. Simulation results across various applications and system configurations further demonstrate that the POW index policy not only has near-optimal performance but also significantly outperforms other existing policies.
Authors: Mingchen Zhuge, Changsheng Zhao, Haozhe Liu, Zijian Zhou, Shuming Liu, Wenyi Wang, Ernie Chang, Gael Le Lan, Junjie Fei, Wenxuan Zhang, Yasheng Sun, Zhipeng Cai, Zechun Liu, Yunyang Xiong, Yining Yang, Yuandong Tian, Yangyang Shi, Vikas Chandra, J\"urgen Schmidhuber
Abstract: We propose a new frontier: Neural Computers (NCs) that unify computation, memory, and I/O of traditional computers in a learned runtime state. Our long-term goal is the Completely Neural Computer (CNC): the mature, general-purpose realization of this emerging machine form, with stable execution, explicit reprogramming, and durable capability reuse. As an initial step, we study whether elementary NC primitives can be learned solely from collected I/O traces, without instrumented program state. Concretely, we instantiate NCs as video models that roll out screen frames from instructions, pixels, and user actions (when available) in CLI and GUI settings. We show that NCs can acquire elementary interface primitives, especially I/O alignment and short-horizon control, while routine reuse, controlled updates, and symbolic stability remain challenging. We outline a roadmap toward CNCs, to establish a new computing paradigm beyond today's agents and conventional computers.
Authors: Erik Y. Wang
Abstract: We give a computer-assisted counterexample to the open question, posed by Rudin, Schapire, and Daubechies in COLT 2012, of whether exhaustive AdaBoost always converges to a finite cycle. The construction is based on a block-product gadget whose two factors share an exact period-2 orbit for their 5-step branch maps, but whose linearized return maps have dominant eigenvalues with an irrational logarithmic ratio. This irrationality forces the burst-winner sequence to have an irrational asymptotic frequency, precluding eventual periodicity. All assertions are certified by exact rational arithmetic. This work was developed in collaboration with GPT-5.4 Pro and Claude Opus 4.6.
Authors: Haonan Zhu, Adrienne Deganutti, Elad Hirsch, Purvanshi Mehta
Abstract: SVG generation is typically evaluated by comparing rendered outputs to reference images, which captures visual similarity but not the structural properties that make SVG editable, decomposable, and reusable. Inspired by the classical jackknife, we introduce element-level leave-one-out (LOO) analysis. The procedure renders the SVG with and without each element, which yields element-level signals for quality assessment and structural analysis. From this single mechanism, we derive (i) per-element quality scores that enable zero-shot artifact detection; (ii) element-concept attribution via LOO footprints crossed with VLM-grounded concept heatmaps; and (iii) four structural metrics: purity, coverage, compactness, and locality, which quantify SVG modularity from complementary angles. These metrics extend SVG evaluation from image similarity to code structure, enabling element-level diagnosis and comparison of how visual concepts are represented, partitioned, and organized within SVG code. Their practical relevance is validated on over 19,000 edits (5 types) across 5 generation systems and 3 complexity tiers.
Authors: Meiyi Zhu, Osvaldo Simeone
Abstract: Conformal selection (CS) uses calibration data to identify test inputs whose unobserved outcomes are likely to satisfy a pre-specified minimal quality requirement, while controlling the false discovery rate (FDR). Existing methods fix the target FDR level before observing data, which prevents the user from adapting the balance between number of selected test inputs and FDR to downstream needs and constraints based on the available data. For example, in genomics or neuroimaging, researchers often inspect the distribution of test statistics, and decide how aggressively to pursue candidates based on observed evidence strength and available follow-up resources. To address this limitation, we introduce {post-hoc CS} (PH-CS), which generates a path of candidate selection sets, each paired with a data-driven false discovery proportion (FDP) estimate. PH-CS lets the user select any operating point on this path by maximizing a user-specified utility, arbitrarily balancing selection size and FDR. Building on conformal e-variables and the e-Benjamini-Hochberg (e-BH) procedure, PH-CS is proved to provide a finite-sample post-hoc reliability guarantee whereby the ratio between estimated FDP level and true FDP is, on average, upper bounded by $1$, so that the average estimated FDP is, to first order, a valid upper bound on the true FDR. PH-CS is extended to control quality defined in terms of a general risk. Experiments on synthetic and real-world datasets demonstrate that, unlike CS, PH-CS can consistently satisfy user-imposed utility constraints while producing reliable FDP estimates and maintaining competitive FDR control.
Authors: Patrick Lutz, Themistoklis Haris, Arjun Chandra, Aditya Gangrade, Venkatesh Saligrama
Abstract: Transformers can perform in-context classification from a few labeled examples, yet the inference-time algorithm remains opaque. We study multi-class linear classification in the hard no-margin regime and make the computation identifiable by enforcing feature- and label-permutation equivariance at every layer. This enables interpretability while maintaining functional equivalence and yields highly structured weights. From these models we extract an explicit depth-indexed recursion: an end-to-end identified, emergent update rule inside a softmax transformer, to our knowledge the first of its kind. Attention matrices formed from mixed feature-label Gram structure drive coupled updates of training points, labels, and the test probe. The resulting dynamics implement a geometry-driven algorithmic motif, which can provably amplify class separation and yields robust expected class alignment.
Authors: You Qin, Linqing Wang, Hao Fei, Roger Zimmermann, Liefeng Bo, Qinglin Lu, Chunyu Wang
Abstract: The post-training pipeline for diffusion models currently has two stages: supervised fine-tuning (SFT) on curated data and reinforcement learning (RL) with reward models. A fundamental gap separates them. SFT optimizes the denoiser only on ground-truth states sampled from the forward noising process; once inference deviates from these ideal states, subsequent denoising relies on out-of-distribution generalization rather than learned correction, exhibiting the same exposure bias that afflicts autoregressive models, but accumulated along the denoising trajectory instead of the token sequence. RL can in principle address this mismatch, yet its terminal reward signal is sparse, suffers from credit-assignment difficulty, and risks reward hacking. We propose SOAR (Self-Correction for Optimal Alignment and Refinement), a bias-correction post-training method that fills this gap. Starting from a real sample, SOAR performs a single stop-gradient rollout with the current model, re-noises the resulting off-trajectory state, and supervises the model to steer back toward the original clean target. The method is on-policy, reward-free, and provides dense per-timestep supervision with no credit-assignment problem. On SD3.5-Medium, SOAR improves GenEval from 0.70 to 0.78 and OCR from 0.64 to 0.67 over SFT, while simultaneously raising all model-based preference scores. In controlled reward-specific experiments, SOAR surpasses Flow-GRPO in final metric value on both aesthetic and text-image alignment tasks, despite having no access to a reward model. Since SOAR's base loss subsumes the standard SFT objective, it can directly replace SFT as a stronger first post-training stage after pretraining, while remaining fully compatible with subsequent RL alignment.
Authors: Talha Ruzgar Akkus, Suayp Talha Kocabay, Kamer Ali Yuksel, Hassan Sawaf
Abstract: The Forward-Forward (FF) algorithm trains networks layer-by-layer using a local "goodness function," yet sum-of-squares (SoS) has remained the only choice studied. We systematically explore the goodness-function design space and identify a unifying principle: the goodness function must be sensitive to the shape of neural activity, not its total energy. This principle is motivated by the observation that deep network activations follow heavy-tailed distributions and that discriminative information is often concentrated in peak activities. We propose two complementary families: selective functions (top-k, entmax-weighted energy) that measure only peak activity, and shape-sensitive functions (excess kurtosis / "burstiness" and higher-order moments) that reward heavy-tailed distributions via scale-invariant statistics. Combined with separate label-feature forwarding (FFCL), controlled experiments across 13 goodness functions, 5 activations, 6 datasets, and three continuous sweeps, each tracing a characteristic inverted-U, yield 89.0% on Fashion-MNIST and 98.2+-0.1% on MNIST (4x2000), a +32.6pp gain over SoS, with consistent improvements across all benchmarks (+72pp USPS, +52pp SVHN). The scale-invariant nature of burstiness makes it particularly robust to magnitude shifts across layers and datasets.
Authors: Chuangtao Chen, Grace Li Zhang, Xunzhao Yin, Cheng Zhuo, Bing Li, Ulf Schlichtmann
Abstract: Large Language Models (LLMs) rely heavily on Key-Value (KV) caching to minimize inference latency. However, standard KV caches are context-dependent: reusing a cached document in a new context requires recomputing KV states to account for shifts in attention distribution. Existing solutions such as CacheBlend, EPIC, and SAM-KV mitigate this issue by selectively recomputing a subset of tokens; however, they still incur non-negligible computational overhead (FLOPs) and increased Time-to-First-Token (TTFT) latency. In this paper, we propose KV Packet, a recomputation-free cache reuse framework that treats cached documents as immutable ``packets'' wrapped in light-weight trainable soft-token adapters, which are trained via self-supervised distillation to bridge context discontinuities. Experiments on Llama-3.1 and Qwen2.5 demonstrate that the proposed KV Packet method achieves near-zero FLOPs and lower TTFT than recomputation-based baselines, while retaining F1 scores comparable to those of the full recomputation baseline.
Authors: Sourav Ganguly, Kartik Pandit, Arnob Ghosh
Abstract: Real-world decision-making systems operate in environments where state transitions depend not only on the agent's actions, but also on \textbf{exogenous factors outside its control}--competing agents, environmental disturbances, or strategic adversaries--formally, $s_{h+1} = f(s_h, a_h, \bar{a}_h)+\omega_h$ where $\bar{a}_h$ is the adversary/external action, $a_h$ is the agent's action, and $\omega_h$ is an additive noise. Ignoring such factors can yield policies that are optimal in isolation but \textbf{fail catastrophically in deployment}, particularly when safety constraints must be satisfied. Standard Constrained MDP formulations assume the agent is the sole driver of state evolution, an assumption that breaks down in safety-critical settings. Existing robust RL approaches address this via distributional robustness over transition kernels, but do not explicitly model the \textbf{strategic interaction} between agent and exogenous factor, and rely on strong assumptions about divergence from a known nominal model. We model the exogenous factor as an \textbf{adversarial policy} $\bar{\pi}$ that co-determines state transitions, and ask how an agent can remain both optimal and safe against such an adversary. \emph{To the best of our knowledge, this is the first work to study safety-constrained RL under explicit adversarial dynamics}. We propose \textbf{Robust Hallucinated Constrained Upper-Confidence RL} (\texttt{RHC-UCRL}), a model-based algorithm that maintains optimism over both agent and adversary policies, explicitly separating epistemic from aleatoric uncertainty. \texttt{RHC-UCRL} achieves sub-linear regret and constraint violation guarantees.
Authors: Yuncong Liu, Yuan Wan, Zhou Jiang, Yao Lu
Abstract: Key Opinion Leader (KOL) discourse on social media is widely consumed as investment guidance, yet turning it into executable trading strategies without injecting assumptions about unspecified execution decisions remains an open problem. We observe that the gaps in KOL statements are not random deficiencies but a structured separation: KOLs express directional intent (what to buy or sell and why) while leaving execution decisions (when, how much, how long) systematically unspecified. Building on this observation, we propose an intent-preserving policy completion framework that treats KOL discourse as a partial trading policy and uses offline reinforcement learning to complete the missing execution decisions around the KOL-expressed intent. Experiments on multimodal KOL discourse from YouTube and X (2022-2025) show that KICL achieves the best return and Sharpe ratio on both platforms while maintaining zero unsupported entries and zero directional reversals, and ablations confirm that the full framework yields an 18.9% return improvement over the KOL-aligned baseline.
Authors: Amy Rouillard, Sitwala Mundia, Linda Camara, Michael Cameron Gramanie, Ziyaad Dangor, Ismail Kalla, Shabir A. Madhi, Kajal Morar, Marlvin T. Ncube, Haroon Saloojee, Bruce A. Bassett
Abstract: Evaluating medical AI systems using expert clinician panels is costly and slow, motivating the use of large language models (LLMs) as alternative adjudicators. Here, we evaluate an LLM jury composed of three frontier AI models scoring 3333 diagnoses on 300 real-world middle-income country (MIC) hospital cases. Model performance was benchmarked against expert clinician panel and independent human re-scoring panel evaluations. Both LLM and clinician-generated diagnoses are scored across four dimensions: diagnosis, differential diagnosis, clinical reasoning and negative treatment risk. For each of these, we assess scoring difference, inter-rater agreement, scoring stability, severe safety errors and the effect of post-hoc calibration. We find that: (i) the uncalibrated LLM jury scores are systematically lower than clinician panels scores; (ii) the LLM Jury preserves ordinal agreement and exhibits better concordance with the primary expert panels than the human expert re-score panels do; (iii) the probability of severe errors is lower in \lj models compared to the human expert re-score panels; (iv) the LLM Jury shows excellent agreement with primary expert panels' rankings. We find that the LLM jury combined with AI model diagnoses can be used to identify ward diagnoses at high risk of error, enabling targeted expert review and improved panel efficiency; (v) LLM jury models show no self-preference bias. They did not score diagnoses generated by their own underlying model or models from the same vendor more (or less) favourably than those generated by other models. Finally, we demonstrate that LLM jury calibration using isotonic regression improves alignment with human expert panel evaluations. Together, these results provide compelling evidence that a calibrated, multi-model LLM jury can serve as a trustworthy and reliable proxy for expert clinician evaluation in medical AI benchmarking.
Authors: Yury Gorishniy, Ivan Rubachev, Dmitrii Feoktistov, Artem Babenko
Abstract: MLP is a heavily used backbone in modern deep learning (DL) architectures for supervised learning on tabular data, and AdamW is the go-to optimizer used to train tabular DL models. Unlike architecture design, however, the choice of optimizer for tabular DL has not been examined systematically, despite new optimizers showing promise in other domains. To fill this gap, we benchmark 15 optimizers on 17 tabular datasets for training MLP-based models in the standard supervised learning setting under a shared experiment protocol. Our main finding is that the Muon optimizer consistently outperforms AdamW, and thus should be considered a strong and practical choice for practitioners and researchers, if the associated training efficiency overhead is affordable. Additionally, we find exponential moving average of model weights to be a simple yet effective technique that improves AdamW on vanilla MLPs, though its effect is less consistent across model variants.
Authors: Yi Yan, Ercan Engin Kuruoglu
Abstract: Spatial-temporal estimation of signals on graph edges is challenging because most conventional Graph Signal Processing techniques are defined on the graph nodes. Leveraging the Line Graph transform, the Line Graph Least Mean Square (LGLMS) algorithm unifies the Line Graph transformation with classical adaptive filters, reinterpreting online estimation techniques for time-varying signals on graph edges. LGLMS leverages the full power of existing GSP techniques on signals on edges by embedding edge signals into node representations, eliminating the necessity of redefining edge-specific techniques. Experimenting with transportation graphs and meteorological graphs, with the signal observations having noisy and missing values, we confirmed that LGLMS is suitable for the online prediction of time-varying edge signals.
Authors: Goitom Ybrah Hailu, Hadush Hailu, Shishay Welay
Abstract: University students often spend a considerable amount of time seeking answers to common questions from administrators or teachers. This can become tedious for both parties, leading to a need for a solution. In response, this paper proposes a chatbot model that utilizes natural language processing and deep learning techniques to answer frequently asked questions (FAQs) in the Amharic language. Chatbots are computer programs that simulate human conversation through the use of artificial intelligence (AI), acting as a virtual assistant to handle questions and other tasks. The proposed chatbot program employs tokenization, normalization, stop word removal, and stemming to analyze and categorize Amharic input sentences. Three machine learning model algorithms were used to classify tokens and retrieve appropriate responses: Support Vector Machine (SVM), Multinomial Na\"ive Bayes, and deep neural networks implemented through TensorFlow, Keras, and NLTK. The deep learning model achieved the best results with 91.55% accuracy and a validation loss of 0.3548 using an Adam optimizer and SoftMax activation function. The chatbot model was integrated with Facebook Messenger and deployed on a Heroku server for 24-hour accessibility. The experimental results demonstrate that the chatbot framework achieved its objectives and effectively addressed challenges such as Amharic Fidel variation, morphological variation, and lexical gaps. Future research could explore the integration of Amharic WordNet to narrow the lexical gap and support more complex questions.
Authors: Dominik Panek, Carina Rz\k{a}ca, Maksymilian Szczypior, Joanna Sorysz, Krzysztof Misztal, Zbigniew Baster, Zenon Rajfur
Abstract: High-throughput biological imaging is often constrained by a trade-off between acquisition speed and image quality. Fast imaging modalities, such as wide-field fluorescence microscopy, enable large-scale data acquisition but suffer from reduced contrast and resolution, whereas high-resolution techniques, including confocal microscopy or single-molecule localization microscopy-based super-resolution techniques, provide superior image quality at the cost of throughput and instrument time. Here, we present a deep learning-based approach for modality transfer across independent microscopes, enabling the transformation of low-quality images acquired on fast systems into high-quality representations comparable to those obtained using advanced imaging platforms. To achieve this, we employ a generative adversarial network (GAN)-based model trained on paired datasets acquired on physically separate wide-field and confocal microscopes, demonstrating that image quality can be reliably transferred between independent instruments. Quantitative evaluation shows substantial improvement in structural similarity and signal fidelity, with median SSIM and PSNR of 0.94 and 31.87, respectively, compared to 0.83 and 21.48 for the original wide-field images. These results indicate that key structural features can be recovered with high accuracy. Importantly, this approach enables a workflow in which high-throughput imaging can be performed on fast, accessible microscopy systems while preserving the ability to computationally recover high-quality structural information. High-resolution microscopy can then be reserved for targeted validation, reducing acquisition time and improving overall experimental efficiency. Together, our results establish deep learning-enabled modality transfer as a practical strategy for bridging independent microscopy systems and supporting scalable, high-content imaging workflows.
Authors: Garima Chhikara, Anurag Sharma, V. Gurucharan, Kripabandhu Ghosh, Abhijnan Chakraborty
Abstract: Citizen reporting platforms help the public and authorities stay informed about sexual harassment incidents. However, the high volume of data shared on these platforms makes reviewing each individual case challenging. Therefore, a summarization algorithm capable of processing and understanding various code-mixed languages is essential. In recent years, Large Language Models (LLMs) have shown exceptional performance in NLP tasks, including summarization. LLMs inherently produce abstractive summaries by paraphrasing the original text, while the generation of extractive summaries - selecting specific subsets from the original text - through LLMs remains largely unexplored. Moreover, LLMs have a limited context window size, restricting the amount of data that can be processed at once. We tackle these challenges by introducing LaMSUM, a novel multi-level framework combining summarization with different voting methods to generate extractive summaries for large collections of incident reports using LLMs. Extensive evaluation using four popular LLMs (Llama, Mistral, Claude and GPT-4o) demonstrates that LaMSUM outperforms state-of-the-art extractive summarization methods. Overall, this work represents one of the first attempts to achieve extractive summarization through LLMs, and is likely to support stakeholders by offering a comprehensive overview and enabling them to develop effective policies to minimize incidents of unwarranted harassment.
Authors: Sergio Hernan Garrido Mejia, Elke Kirschbaum, Armin Keki\'c, Bernhard Sch\"olkopf, Atalanti Mastakouri
Abstract: In this paper we show how to exploit interventional data to acquire the joint conditional distribution of all the variables using the Maximum Entropy principle. To this end, we extend the Causal Maximum Entropy method to make use of interventional data in addition to observational data. Using Lagrange duality, we prove that the solution to the Causal Maximum Entropy problem with interventional constraints lies in the exponential family, as in the Maximum Entropy solution. Our method allows us to perform two tasks of interest when marginal interventional distributions are provided for any subset of the variables. First, we show how to perform causal feature selection from a mixture of observational and single-variable interventional data, and, second, how to infer joint interventional distributions. For the former task, we show on synthetically generated data, that our proposed method outperforms the state-of-the-art method on merging datasets, and yields comparable results to the KCI-test which requires access to joint observations of all variables.
Authors: Koki Chinzei, Quoc Hoan Tran, Yasuhiro Endo, Hirotaka Oshima
Abstract: Equivariant quantum neural networks (QNNs) are promising variational models that exploit symmetries to improve machine learning capabilities. Despite theoretical developments in equivariant QNNs, their implementation on near-term quantum devices remains challenging due to limited computational resources. This study proposes a resource-efficient model of equivariant quantum convolutional neural networks (QCNNs) called equivariant split-parallelizing QCNN (sp-QCNN). Using a group-theoretical approach, we encode general symmetries into our model beyond the translational symmetry addressed by previous sp-QCNNs. We achieve this by splitting the circuit at the pooling layer while preserving symmetry. This splitting structure effectively parallelizes QCNNs to improve measurement efficiency in estimating the expectation value of an observable and its gradient by order of the number of qubits. Our model also exhibits high trainability and generalization performance, including the absence of barren plateaus. Numerical experiments demonstrate that the equivariant sp-QCNN can be trained and generalized with fewer measurement resources than a conventional equivariant QCNN in a noisy quantum data classification task. Our results contribute to the advancement of practical quantum machine learning algorithms.
Authors: Ermis Soumalias, Jakob Heiss, Jakob Weissteiner, Sven Seuken
Abstract: We study the design of iterative combinatorial auctions (ICAs). The main challenge in this domain is that the bundle space grows exponentially in the number of items. To address this, recent work has proposed machine learning (ML)-based preference elicitation algorithms that aim to elicit only the most critical information from bidders to maximize efficiency. However, while the SOTA ML-based algorithms elicit bidders' preferences via value queries, ICAs that are used in practice elicit information via \emph{demand queries}. In this paper, we introduce a novel ML algorithm that provably makes use of the full information from both value and demand queries, and we show via experiments that combining both query types results in significantly better learning performance in practice. Building on these insights, we present MLHCA, a new ML-powered auction that uses value and demand queries. MLHCA significantly outperforms the previous SOTA, reducing efficiency loss by up to a factor 10, with up to 58\% fewer queries. Thus, MLHCA achieves large efficiency improvements while also reducing bidders' cognitive load, establishing a new benchmark for both practicability and efficiency. Our code is available at https://github.com/marketdesignresearch/MLHCA.
Authors: Mengxin Wang (Naveen Jindal School of Management, The University of Texas at Dallas), Dennis J. Zhang (Olin School of Business, Washington University in St. Louis), Heng Zhang (W. P. Carey School of Business, Arizona State University)
Abstract: Large Language Models (LLMs) have transformed artificial intelligence by excelling in complex natural language processing tasks. Their ability to generate human-like text has opened new possibilities for market research, particularly in conjoint analysis, where understanding consumer preferences is essential but often resource-intensive. Traditional survey-based methods face limitations in scalability and cost, making LLM-generated data a promising alternative. However, while LLMs have the potential to simulate real consumer behavior, recent studies highlight a significant gap between LLM-generated and human data, with biases introduced when substituting between the two. In this paper, we address this gap by proposing a novel statistical data augmentation approach that efficiently integrates LLM-generated data with real data in conjoint analysis. This results in statistically robust estimators with consistent and asymptotically normal properties, in contrast to naive approaches that simply substitute human data with LLM-generated data, which can exacerbate bias. We further present a finite-sample performance bound on the estimation error. We validate our framework through an empirical study on COVID-19 vaccine preferences, demonstrating its superior ability to reduce estimation error and save data and costs by 24.9% to 79.8%. In contrast, naive approaches fail to save data due to the inherent biases in LLM-generated data compared to human data. Another empirical study on sports car choices validates the robustness of our results. Our findings suggest that while LLM-generated data is not a direct substitute for human responses, it can serve as a valuable complement when used within a robust statistical framework.
Authors: Nora Gourmelon, Konrad Heidler, Erik Loebel, Daniel Cheng, Julian Klink, Anda Dong, Fei Wu, Noah Maul, Moritz Koch, Marcel Dreier, Dakota Pyles, Thorsten Seehaus, Matthias Braun, Andreas Maier, Vincent Christlein
Abstract: Continuous monitoring of glacier calving fronts is essential for sea level rise projections. This study benchmarks Deep Learning systems for front delineation in Synthetic Aperture Radar imagery. While Deep Learning systems exhibit errors up to 221 m, human annotators deviate by only 38 m, underscoring the need for further research.
Authors: Qin Fang, Lei Shi, Min Xu, Ding-Xuan Zhou
Abstract: This paper investigates approximation capabilities of two-dimensional (2D) deep convolutional neural networks (CNNs), with Korobov functions serving as a benchmark. We focus on 2D CNNs, comprising multi-channel convolutional layers with zero-padding and ReLU activations, followed by a fully connected layer. We propose a fully constructive approach for building 2D CNNs to approximate Korobov functions and provide a rigorous analysis of the complexity of the constructed networks. Our results demonstrate that 2D CNNs achieve near-optimal approximation rates under the continuous weight selection model, significantly alleviating the curse of dimensionality. This work provides a solid theoretical foundation for 2D CNNs and illustrates their potential for broader applications in function approximation.
Authors: Dimitris Tsirmpas, Ion Androutsopoulos, John Pavlopoulos
Abstract: A critical challenge in social science research is the high cost associated with experiments involving human participants. We identify Synthetic Discussion Generation (SDG), a novel Natural Language Processing (NLP) direction aimed at creating simulated discussions that enable cost-effective pilot experiments and develop a theoretical, task-agnostic framework for designing, evaluating, and implementing these simulations. We argue that the use of proprietary models such as the OpenAI GPT family for such experiments is often unjustified in terms of both cost and capability, despite its prevalence in current research. Our experiments demonstrate that smaller quantized models (7B-8B) can produce effective simulations at a cost more than 44 times lower compared to their proprietary counterparts. We use our framework in the context of online facilitation, where humans actively engage in discussions to improve them, unlike more conventional content moderation. By treating this problem as a downstream task for our framework, we show that synthetic simulations can yield generalizable results at least by revealing limitations before engaging human discussants. In LLM facilitators, a critical limitation is that they are unable to determine when to intervene in a discussion, leading to undesirable frequent interventions and, consequently, derailment patterns similar to those observed in human interactions. Additionally, we find that different facilitation strategies influence conversational dynamics to some extent. Beyond our theoretical SDG framework, we also present a cost-comparison methodology for experimental design, an exploration of available models and algorithms, an open-source Python framework, and a large, publicly available dataset of LLM-generated discussions across multiple models.
Authors: Rachmad Vidya Wicaksana Putra, Avaneesh Devkota, Muhammad Shafique
Abstract: Training resource-constrained autonomous agents on multiple tasks simultaneously is crucial for adapting to diverse real-world environments. Recent works employ reinforcement learning (RL) approach, but they still suffer from sub-optimal multi-task performance due to task interference. State-of-the-art works employ Spiking Neural Networks (SNNs) to improve RL-based multi-task learning and enable low-power/energy operations through network enhancements and spike-driven data stream processing. However, they rely on fixed task-switching intervals during its training, thus limiting its performance and scalability. To address this, we propose SwitchMT, a novel methodology that employs adaptive task-switching for effective, scalable, and simultaneous multi-task learning. SwitchMT employs the following key ideas: (1) leveraging a Deep Spiking Q-Network with active dendrites and dueling structure, that utilizes task-specific context signals to create specialized sub-networks; and (2) devising an adaptive task-switching policy that leverages both rewards and internal dynamics of the network parameters. Experimental results demonstrate that SwitchMT achieves competitive scores in multiple Atari games (i.e., Pong: -8.8, Breakout: 5.6, and Enduro: 355.2) and longer game episodes as compared to the state-of-the-art. These results also highlight the effectiveness of SwitchMT methodology in addressing task interference without increasing the network complexity, enabling intelligent autonomous agents with scalable multi-task learning capabilities.
Authors: Wenhao Qian, Zhenzhen Hu, Zijie Song, Jia Li
Abstract: Metaphorical imagination, the ability to connect seemingly unrelated concepts, is fundamental to human cognition and communication. While understanding linguistic metaphors has advanced significantly, grasping multimodal metaphors, such as those found in internet memes, presents unique challenges due to their unconventional expressions and implied meanings. Existing methods for multimodal metaphor identification often struggle to bridge the gap between literal and figurative interpretations. Additionally, generative approaches that utilize large language models or text-to-image models, while promising, suffer from high computational costs. This paper introduces \textbf{C}oncept \textbf{D}rift \textbf{G}uided \textbf{L}ayerNorm \textbf{T}uning (\textbf{CDGLT}), a novel and training-efficient framework for multimodal metaphor identification. CDGLT incorporates two key innovations: (1) Concept Drift, a mechanism that leverages Spherical Linear Interpolation (SLERP) of cross-modal embeddings from a CLIP encoder to generate a new, divergent concept embedding. This drifted concept helps to alleviate the gap between literal features and the figurative task. (2) A prompt construction strategy, that adapts the method of feature extraction and fusion using pre-trained language models for the multimodal metaphor identification task. CDGLT achieves state-of-the-art performance on the MET-Meme benchmark while significantly reducing training costs compared to existing generative methods. Ablation studies demonstrate the effectiveness of both Concept Drift and our adapted LN Tuning approach. Our method represents a significant step towards efficient and accurate multimodal metaphor understanding. The code is available: \href{https://github.com/Qianvenh/CDGLT}{https://github.com/Qianvenh/CDGLT}.
URLs: https://github.com/Qianvenh/CDGLT, https://github.com/Qianvenh/CDGLT
Authors: Zhongyu Ouyang, Qianlong Wen, Chunhui Zhang, Yanfang Ye, Soroush Vosoughi
Abstract: What enables large language models (LLMs) to effectively model user preferences in sequential recommendation? Our investigation reveals that existing preference-alignment approaches largely rely on binary pairwise comparisons, overlooking two critical factors: preference intensity (the structured strength of affinity or aversion) and temporal context (the extent to which recent interactions better reflect a user's current intent). Through controlled experiments, we show that leveraging comprehensive feedback with structured preference signals substantially improves recommendation performance, indicating that binary modeling discards essential information. Motivated by these findings, we propose RecPO, a unified preference optimization framework that maps both explicit and implicit feedback into a common preference signal and constructs adaptive reward margins that jointly account for preference intensity and interaction recency. Experiments across five datasets show that RecPO consistently outperforms state-of-the-art baselines while exhibiting behavioral patterns aligned with human decision-making, including favoring immediate satisfaction, maintaining preference coherence, and avoiding dispreferred items. Our results highlight that preference intensity and temporal context are fundamental ingredients for effective LLM-based recommendation.
Authors: Dorival Le\~ao, Reiko Aoki, Alberto Ohashi, Teh Led Red
Abstract: This paper presents ``randomized SINDy", a sequential machine learning algorithm designed for dynamic data that has a time-dependent structure. It employs a probabilistic approach, with its PAC learning property rigorously proven through the mathematical theory of functional analysis. The algorithm dynamically predicts using a learned probability distribution of predictors, updating weights via gradient descent and a proximal algorithm to maintain a valid probability density. Inspired by SINDy (Brunton et al. 2016), it incorporates feature augmentation and Tikhonov regularization. For multivariate normal weights, the proximal step is omitted to focus on parameter estimation. The algorithm's effectiveness is demonstrated through experimental results in regression and binary classification using real-world data.
Authors: Ran Tong, Lanruo Wang, Tong Wang, Wei Yan
Abstract: Predicting Parkinson's Disease (PD) progression is crucial for personalized treatment, and voice biomarkers offer a promising non-invasive method for tracking symptom severity through telemonitoring. However, analyzing this longitudinal data is challenging due to inherent within-subject correlations, the small sample sizes typical of clinical trials, and complex patient-specific progression patterns. While deep learning offers high theoretical flexibility, its application to small-cohort longitudinal studies remains under-explored compared to traditional statistical methods. This study presents an application of the Neural Mixed Effects (NME) framework to Parkinson's telemonitoring, benchmarking it against Generalized Neural Network Mixed Models (GNMM) and semi-parametric statistical baseline of Generalized Additive Mixed Models (GAMMs). Using the Oxford Parkinson's telemonitoring voice dataset (), we demonstrate that while neural architectures offer flexibility, they are prone to significant overfitting in small-sample regimes. Our results indicate that GAMMs provide the optimal balance, achieving superior predictive accuracy (MSE 6.56) compared to neural baselines (MSE > 90) while maintaining clinical interpretability. We discuss the critical implications of these findings for developing robust, deployable telemonitoring systems where data scarcity is a constraint, highlighting the necessity for larger, diverse datasets for neural model validation.
Authors: Md Afif Al Mamun, Gias Uddin, Lan Xia, Longyu Zhang
Abstract: Pretrained Language Models or PLMs are transformer-based architectures that can be used in bug triaging tasks. PLMs can better capture token semantics than traditional Machine Learning (ML) models that rely on statistical features (e.g., TF-IDF, bag of words). However, PLMs may still attend to less relevant tokens in a bug report, which can impact their effectiveness. In addition, the model can be sub-optimal with its recommendations when the interaction history of developers around similar bugs is not taken into account. We designed TriagerX to address these limitations. First, to assess token semantics more reliably, we leverage a dual-transformer architecture. Unlike current state-of-the-art (SOTA) baselines that employ a single transformer architecture, TriagerX collects recommendations from two transformers with each offering recommendations via its last three layers. This setup generates a robust content-based ranking of candidate developers. TriagerX then refines this ranking by employing a novel interaction-based ranking methodology, which considers developers' historical interactions with similar fixed bugs. Across five datasets, TriagerX surpasses all nine transformer-based methods, including SOTA baselines, often improving Top-1 and Top-3 developer recommendation accuracy by over 10%. We worked with our large industry partner to successfully deploy TriagerX in their development environment. The partner required both developer and component recommendations, with components acting as proxies for team assignments-particularly useful in cases of developer turnover or team changes. We trained TriagerX on the partner's dataset for both tasks, and it outperformed SOTA baselines by up to 10% for component recommendations and 54% for developer recommendations.
Authors: Yuxuan Zhu, David H. Yang, Mohammad Mohammadi Amiri, Keerthiram Murugesan, Tejaswini Pedapati, Pin-Yu Chen
Abstract: The expanding long-context capabilities of large language models are constrained by a significant memory bottleneck: the key-value (KV) cache required for autoregressive generation. This bottleneck is substantial; for instance, a Llama-3.1-8B model processing a 32K-token prompt at a batch size of 4 requires approximately 16GB for its KV cache, a size exceeding the model's weights. While KV-cache compression via low-rank projection is a promising direction, existing methods rely on a static, offline-learned subspace that performs poorly under data distribution shifts. To overcome these limitations, we introduce OjaKV, a novel framework that integrates a strategic hybrid storage policy with online subspace adaptation. First, OjaKV recognizes that not all tokens are equally important for compression; it preserves the crucial first and most recent tokens in full-rank, maintaining high-fidelity anchors for attention. Second, for the vast majority of intermediate tokens, it applies low-rank compression by incrementally adapting the projection basis using Oja's algorithm for online principal component analysis. This adaptation involves a comprehensive update during prompt prefilling and lightweight periodic updates during decoding, ensuring the subspace remains aligned with the evolving context. Crucially, our framework is fully compatible with modern attention modules like FlashAttention. Experiments demonstrate that OjaKV maintains or even improves zero-shot accuracy at high compression ratios. In particular, OjaKV achieves its strongest gains on very long-context benchmarks that require complex reasoning, highlighting the importance of online subspace adaptation in dynamically tracking context shifts. These results establish our hybrid framework as a practical, plug-and-play solution for memory-efficient long-context inference without requiring model fine-tuning.
Authors: Akira Takahashi, Shusuke Takahashi, Yuki Mitsufuji
Abstract: We introduce MMAudioSep, a generative model for video/text-queried sound separation that is founded on a pretrained video-to-audio model. By leveraging knowledge about the relationship between video/text and audio learned through a pretrained audio generative model, we can train the model more efficiently, i.e., the model does not need to be trained from scratch. We evaluate the performance of MMAudioSep by comparing it to existing separation models, including models based on both deterministic and generative approaches, and find it is superior to the baseline models. Furthermore, we demonstrate that even after acquiring functionality for sound separation via fine-tuning, the model retains the ability for original video-to-audio generation. This highlights the potential of foundational sound generation models to be adopted for sound-related downstream tasks. Our code is available at https://github.com/sony/mmaudiosep.
Authors: Zihan Zhao, Kaushik Pendiyala, Masood Mortazavi, Ning Yan
Abstract: Multi-sensory systems for embodied intelligence, from wearable body-sensor networks to instrumented robotic platforms, routinely face a sensor-asymmetry problem: the richest modality available during laboratory data collection is absent or impractical at deployment time due to cost, fragility, or interference with physical interaction. We introduce PULSE, a general framework for privileged knowledge transfer from an information-rich teacher sensor to a set of cheaper, deployment-ready student sensors. Each student encoder produces shared (modality-invariant) and private (modality-specific) embeddings; the shared subspace is aligned across modalities and then matched to representations of a frozen teacher via multi-layer hidden-state and pooled-embedding distillation. Private embeddings preserve modality-specific structure needed for self-supervised reconstruction, which we show is critical to prevent representational collapse. We instantiate PULSE on the wearable stress-monitoring task, using electrodermal activity (EDA) as the privileged teacher and ECG, BVP, accelerometry, and temperature as students. On the WESAD benchmark under leave-one-subject-out evaluation, PULSE achieves 0.994 AUROC and 0.988 AUPRC (0.965/0.955 on STRESS) without EDA at inference, exceeding all no-EDA baselines and matching the performance of a full-sensor model that retains EDA at test time. We further demonstrate modality-agnostic transfer with ECG as teacher, provide extensive ablations on hidden-state matching depth, shared-private capacity, hinge-loss margin, fusion strategy, and modality dropout, and discuss how the framework generalizes to broader embodied sensing scenarios involving tactile, inertial, and bioelectrical modalities.
Authors: Ritik Raj, Souvik Kundu, Ishita Vohra, Hong Wang, Tushar Krishna
Abstract: Agentic AI serving converts monolithic LLM-based inference to autonomous problem-solvers that can plan, call tools, perform reasoning, and adapt on the fly. Due to diverse task execution need, such serving heavily rely on heterogeneous CPU-GPU systems with majority of the external tools responsible for agentic capability, either run on or are orchestrated by the CPU. Towards having a deeper understanding of its role, this paper aims to characterize and analyze the system bottlenecks introduced by agentic AI workloads from a largely overlooked CPU-centric perspective. We first present a compile-time characterization of agentic AI execution and choose representative workloads to capture the algorithmic diversity. We then perform runtime characterization of the representative workloads analyzing the end-to-end latency and throughput on two different hardware systems to isolate respective architectural bottlenecks. Based on the insights on the bottlenecks, we finally present two scheduling optimizations, namely, 1. CPU-Aware Overlapped Micro-Batching (COMB) and 2. Mixed Agentic Scheduling (MAS) on homogeneous and heterogeneous agentic workloads, respectively. In specific, these methods optimize for improved CPU-GPU concurrent utilization while reducing skewed resource allocation for heterogeneous execution. Experimental evaluations on the two hardware systems demonstrate the efficacy of COMB in yielding up to 1.7x lower P50 latency in standalone homogeneous workload execution and up to 3.9x/1.8x lower service/total latency under homogeneous open-loop load. Additionally, for heterogeneous open-loop load, MAS can reduce the total latency for minority request-type by up to 2.37x/2.49x at P50/P90 percentile.
Authors: Victoria Ebert, Rishabh Singh, Tuochao Chen, Noah A. Smith, Shyamnath Gollakota
Abstract: Conversational AI is constrained in many real-world settings where only one side of a dialogue can be recorded, such as telemedicine, call centers, and smart glasses. We formalize this as the one-sided conversation problem (1SC): inferring and learning from one side of a conversation. We study two tasks: (1) reconstructing the missing speaker's turns for real-time use cases, and (2) generating summaries from one-sided transcripts. Evaluating prompting and finetuned models on MultiWOZ, DailyDialog, and Candor with both human A/B testing and LLM-as-a-judge metrics, we find that access to one future turn and information about utterance length improves reconstruction, placeholder prompting helps to mitigate hallucination, and while large models generate promising reconstructions with prompting, smaller models require finetuning. Further, high-quality summaries can be generated without reconstructing missing turns. We present 1SC as a novel challenge and report promising results that mark a step toward privacy-aware conversational AI.
Authors: Dionysios Adamopoulos, Anastasia Poulopoulou, Georgios Goumas, Christina Giannoula
Abstract: Sparse Convolution (SpC) powers 3D point cloud networks widely used in autonomous driving and augmented/virtual reality. SpC builds a kernel map that stores mappings between input voxel coordinates, output coordinates, and weight offsets, then uses this map to compute feature vectors for output coordinates. Our work identifies three key properties of voxel coordinates: they are integer-valued, bounded within a limited spatial range, and geometrically continuous, i.e., neighboring voxels on the same object surface are highly likely to exist at small spatial offsets from each other. Prior SpC engines do not fully exploit these properties and suffer from high pre-processing and post-processing overheads during kernel map construction. To address this, we design Spira, the first voxel-property-aware SpC engine for GPUs. Spira proposes (i) a high-performance one-shot search algorithm that builds the kernel map with no pre-processing and high data locality, (ii) an effective packed-native processing scheme that accesses packed voxel coordinates at low cost, (iii) a flexible dual-dataflow execution mechanism that efficiently computes output feature vectors by adapting to layer characteristics, and (iv) a network-wide parallelization strategy that builds kernel maps for all SpC layers concurrently at network start. Our evaluation shows that Spira significantly outperforms prior state-of-the-art SpC engines by 1.68x on average and up to 3.04x for end-to-end inference, and by 2.11x on average and up to 3.44x for layer-wise execution across diverse layer configurations. The source code of Spira is freely available at https://github.com/SPIN-Research-Group/Spira.
Authors: Samuel Burles, Enrico Camporeale
Abstract: The requirement for large-scale global simulations of plasma is an ongoing challenge in both space and laboratory plasma physics. Any simulation based on a fluid model inherently requires a closure relation for the high order plasma moments. This review compiles and analyses the recent surge of machine learning approaches developing improved plasma closure models capable of capturing kinetic phenomena within plasma fluid models. The purpose of this review is both to collect and analyse the various methods employed on the plasma closure problem, including both equation discovery methods and neural network surrogate approaches, as well as to provide a general overview of the state of the problem. In particular, we outline the challenges associated with machine learning based closure relations and the direction that future research might take in order to address these challenges.
Authors: Sofiia Chorna, Davide Tisi, Cesare Malosso, Wei Bin How, Michele Ceriotti, Sanggyu Chong
Abstract: The past few years have seen the development of ``universal'' machine-learning interatomic potentials (uMLIPs) capable of approximating the ground-state potential energy surface across a wide range of chemical structures and compositions with reasonable accuracy. While these models differ in the architecture and the dataset used, they share the ability to compress a staggering amount of chemical information into descriptive latent features. Herein, we systematically analyze what the different uMLIPs have learned by quantitatively assessing the relative information content of their latent features with feature reconstruction errors, and observing how the trends are affected by the choice of training set and training protocol. We find that uMLIPs encode the chemical space in significantly distinct ways, with substantial cross-model feature reconstruction errors. When variants of the same model architecture are considered, trends become dependent on the dataset, target, and training protocol of choice. We also observe that fine-tuning of a uMLIP retains a strong pre-training bias in the latent features. Finally, we discuss how atom-level features, which are directly output by MLIPs, can be compressed into global structure-level features via concatenation of progressive cumulants, each adding significantly new information about the variability across the atomic environments within a given system.
Authors: Dongqi Liu, Hang Ding, Qiming Feng, Xurong Xie, Zhucun Xue, Chengjie Wang, Jian Li, Jiangning Zhang, Yabiao Wang
Abstract: Retrieval-Augmented Generation (RAG) has emerged as an important means of enhancing the performance of large language models (LLMs) in knowledge-intensive tasks. However, most existing RAG strategies treat retrieved passages in a flat and unstructured way, which prevents the model from capturing structural cues and constrains its ability to synthesize knowledge from dispersed evidence across documents. To overcome these limitations, we propose Disco-RAG, a discourse-aware framework that explicitly injects discourse signals into the generation process. Our method constructs intra-chunk discourse trees to capture local hierarchies and builds inter-chunk rhetorical graphs to model cross-passage coherence. These structures are jointly integrated into a planning blueprint that conditions the generation. Experiments on question answering and long-document summarization benchmarks show the efficacy of our approach. Disco-RAG achieves state-of-the-art results on the benchmarks without fine-tuning. These findings underscore the important role of discourse structure in advancing RAG systems.
Authors: Xiaoshuai Song, Haofei Chang, Guanting Dong, Yutao Zhu, Ji-Rong Wen, Zhicheng Dou
Abstract: Large language models (LLMs) are expected to be trained to act as agents in various real-world environments, but this process relies on rich and varied tool-interaction sandboxes. However, access to real systems is often restricted; LLM-simulated environments are prone to hallucinations and inconsistencies; and manually built sandboxes are hard to scale. In this paper, we propose EnvScaler, an automated framework for scalable tool-interaction environments via programmatic synthesis. EnvScaler comprises two components. First, SkelBuilder constructs diverse environment skeletons through topic mining, logic modeling, and quality evaluation. Then, ScenGenerator generates multiple task scenarios and rule-based trajectory validation functions for each environment. With EnvScaler, we synthesize 191 environments and about 7K scenarios, and apply them to Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) for Qwen3 series models. Results on three benchmarks show that EnvScaler significantly improves LLMs' ability to solve tasks in complex environments involving multi-turn, multi-tool interactions. We release our code and data at https://github.com/RUC-NLPIR/EnvScaler.
Authors: Alexandra Dragomir, Florin Brad, Radu Tudor Ionescu
Abstract: Large language models (LLMs) have demonstrated competitive performance in zero-shot multilingual machine translation (MT). Some follow-up works further improved MT performance via preference optimization, but they leave a key aspect largely underexplored: the order in which data samples are given during training. We address this topic by integrating curriculum learning into various state-of-the-art preference optimization algorithms to boost MT performance. We introduce a novel curriculum learning strategy with restarts (CLewR), which reiterates easy-to-hard curriculum multiple times during training to effectively mitigate the catastrophic forgetting of easy examples. We demonstrate consistent gains across several model families (Gemma2, Qwen2.5, Llama3.1) and preference optimization techniques. We publicly release our code at https://github.com/alexandra-dragomir/CLewR.
Authors: Xinzi Cao, Jianyang Zhai, Pengfei Li, Zhiheng Hu, Cen Yan, Bingxu Mu, Guanghuan Fang, Bin She, Jiayu Li, Yihan Su, Dongyang Tao, Xiansong Huang, Fan Xu, Feidiao Yang, Yao Lu, Chang-Dong Wang, Yutong Lu, Weicheng Xue, Bin Zhou, Yonghong Tian
Abstract: To meet the ever-increasing demand for computational efficiency, Neural Processing Units (NPUs) have become critical in modern AI infrastructure. However, unlocking their full potential requires developing high-performance compute kernels using vendor-specific Domain-Specific Languages (DSLs), a task that demands deep hardware expertise and is labor-intensive. While Large Language Models (LLMs) have shown promise in general code generation, they struggle with the strict constraints and scarcity of training data in the NPU domain. Our preliminary study reveals that state-of-the-art general-purpose LLMs fail to generate functional complex kernels for Ascend NPUs, yielding a near-zero success rate. To address these challenges, we propose AscendKernelGen, a generation-evaluation integrated framework for NPU kernel development. We introduce Ascend-CoT, a high-quality dataset incorporating chain-of-thought reasoning derived from real-world kernel implementations, and KernelGen-LM, a domain-adaptive model trained via supervised fine-tuning and reinforcement learning with execution feedback. Furthermore, we design NPUKernelBench, a comprehensive benchmark for assessing compilation, correctness, and performance across varying complexity levels. Experimental results demonstrate that our approach significantly bridges the gap between general LLMs and hardware-specific coding. Specifically, the compilation success rate on complex Level-2 kernels improves from 0% to 95.5% (Pass@10), while functional correctness achieves 64.3% compared to the baseline's complete failure. These results highlight the critical role of domain-specific reasoning and rigorous evaluation in automating accelerator-aware code generation. AscendKernGen is available at https://huggingface.co/AscendKernelGen and https://github.com/weich97/NPUKernelBench.
URLs: https://huggingface.co/AscendKernelGen, https://github.com/weich97/NPUKernelBench.
Authors: Ramtin Babaeipour, Fran\c{c}ois Charest, Madison Wright
Abstract: Increasing clinical trial protocol complexity, amendments, and challenges around knowledge management create significant burden for trial teams. Structuring protocol content into standard formats has the potential to improve efficiency, support documentation quality, and strengthen compliance. We evaluate an Artificial Intelligence (AI) system using generative LLMs with Retrieval-Augmented Generation (RAG) for automated clinical trial protocol information extraction. We compare the extraction accuracy of our clinical-trial-specific RAG process against that of publicly available (standalone) LLMs. We also assess the operational impact of AI-assistance on simulated extraction Clinical Research Coordinator (CRC) workflows. Our RAG process shows higher extraction accuracy (89.0%) than standalone LLMs with fine-tuned prompts (62.6%) against expert-supported reference annotations. In simulated extraction workflows, AI-assisted tasks are completed 40% faster, are rated as less cognitively demanding and are strongly preferred by users. While expert oversight remains essential, this suggests that AI-assisted extraction can enable protocol intelligence at scale, motivating the integration of similar methodologies into real-world clinical workflows to further validate its impact on feasibility, study start-up, and post-activation monitoring.
Authors: Yunwei Bai, Ying Kiat Tan, Yao Shu, Tsuhan Chen
Abstract: Few-shot learning (FSL) challenges model generalization to novel classes based on just a few shots of labeled examples, a testbed where traditional test-time augmentations fail to be effective. We introduce 1S-DAug, a one-shot generative augmentation operator that synthesizes diverse yet faithful variants from just one example image at test time. 1S-DAug couples traditional geometric perturbations with controlled noise injection and a denoising diffusion process conditioned on the original image. The generated images are then encoded and aggregated, alongside the original image, into a combined representation for more robust few-shot predictions. Integrated as a training-free model-agnostic plugin, 1S-DAug consistently improves few-shot classification across standard benchmarks of 4 different datasets without any model parameter update, including achieving up to 20\% relative accuracy improvement on the miniImagenet 5-way-1-shot benchmark. Additionally, we provide extension experiments on the larger vision language models as well as theoretical analyses.
Authors: Yulia Alexandr, Hao Duan, Guido Mont\'ufar
Abstract: We study robustness verification of neural networks via metric algebraic geometry. For polynomial neural networks, certifying a robustness radius amounts to computing the distance to the algebraic decision boundary. We use the Euclidean distance (ED) degree as an intrinsic measure of the complexity of this problem, analyze the associated ED discriminant, and introduce a parameter discriminant that detects parameter values at which the ED degree drops. We derive formulas for the ED degree for several network architectures and characterize the expected number of real critical points in the infinite-width limit. We develop symbolic elimination methods to compute these quantities and homotopy-continuation methods for exact robustness certification. Finally, experiments on lightning self-attention modules reveal decision boundaries with strictly smaller ED degree than generic cubic hypersurfaces of the same ambient dimension.
Authors: Alokesh Manna, Neil Spencer, Dipak K. Dey
Abstract: Shoe print evidence recovered from crime scenes plays a key role in forensic investigations. By examining shoe prints, investigators can determine details of the footwear worn by suspects. However, establishing that a suspect's shoes match the make and model of a crime scene print may not be sufficient. Typically, thousands of shoes of the same size, make, and model are manufactured, any of which could be responsible for the print. Accordingly, a popular approach used by investigators is to examine the print for signs of ``accidentals,'' i.e., cuts, scrapes, and other features that accumulate on shoe soles after purchase due to wear. While some patterns of accidentals are common on certain types of shoes, others are highly distinctive, potentially distinguishing the suspect's shoe from all others. Quantifying the rarity of a pattern is thus essential to accurately measuring the strength of forensic evidence. In this study, we address this task by developing a hierarchical Bayesian model. Our improvement over existing methods primarily stems from two advancements. First, we frame our approach in terms of a latent Gaussian model, thus enabling inference to be efficiently scaled to large collections of annotated shoe prints via integrated nested Laplace approximations. Second, we incorporate spatially varying coefficients to model the relationship between shoes' tread patterns and accidental locations. We demonstrate these improvements through superior performance on held-out data, which enhances accuracy and reliability in forensic shoe print analysis.
Authors: Sukumar Kishanthan, Kumar Thushalika, Buddhi Jayasekara, Asela Hevapathige
Abstract: Large language models (LLMs) have achieved strong results in mathematical reasoning, and are increasingly deployed as tutoring and learning support tools in educational settings. However, their reliability for students working in non-English languages, especially low-resource languages, remains poorly understood. We examine this gap by evaluating mathematical reasoning in Sinhala and Tamil -- two languages widely used in South Asian schools but underrepresented in artificial intelligence (AI) research. Using a taxonomy of six math problem types, from basic arithmetic to complex unit conflict and optimization problems, we evaluate four prominent large language models. To avoid translation artifacts that confound language ability with translation quality, we construct a parallel dataset in which each problem is independently authored in Sinhala and Tamil by native speakers, and in English by fluent speakers, all with strong mathematical backgrounds. Our analysis demonstrates that while basic arithmetic reasoning transfers robustly across languages, complex reasoning tasks show significant degradation in Tamil and Sinhala. The pattern of failures varies by model and problem type, suggesting that strong performance in English does not guarantee reliable performance across languages. These findings have direct implications for the deployment of AI tools in multilingual classrooms, and highlight the need for language-specific evaluation before adopting large language models as math tutoring aids in non-English educational contexts.
Authors: Erik Burman, Mats G. Larson, Karl Larsson, Jonatan Vallin
Abstract: We develop an interpolation-based modeling framework for parameter-dependent partial differential equations arising in control, inverse problems, and uncertainty quantification. The solution is discretized in the physical domain using finite element methods, while the dependence on a finite-dimensional parameter is approximated separately. We establish existence, uniqueness, and regularity of the parametric solution and derive rigorous error estimates that explicitly quantify the interplay between spatial discretization and parameter approximation. In low-dimensional parameter spaces, classical interpolation schemes yield algebraic convergence rates based on Sobolev regularity in the parameter variable. In higher-dimensional parameter spaces, we replace classical interpolation by extreme learning machine (ELM) surrogates and obtain error bounds under explicit approximation and stability assumptions. The proposed framework is applied to inverse problems in quantitative photoacoustic tomography, where we derive potential and parameter reconstruction error estimates and demonstrate substantial computational savings compared to standard approaches, without sacrificing accuracy.
Authors: Soroosh Tayebi Arasteh, Marziyeh Mohammadi, Sven Nebelung, Daniel Truhn
Abstract: Differential privacy (DP)'s effect in medical imaging is typically evaluated only through end-to-end performance, leaving the mechanism of privacy-induced utility loss unclear. We introduce Differential Privacy Representation Geometry for Medical Imaging (DP-RGMI), a framework that interprets DP as a structured transformation of representation space and decomposes performance degradation into encoder geometry and task-head utilization. Geometry is quantified by representation displacement from initialization and spectral effective dimension, while utilization is measured as the gap between linear-probe and end-to-end utility. Across over 594,000 images from four chest X-ray datasets and multiple pretrained initializations, we show that DP is consistently associated with a utilization gap even when linear separability is largely preserved. At the same time, displacement and spectral dimension exhibit non-monotonic, initialization- and dataset-dependent reshaping, indicating that DP alters representation anisotropy rather than uniformly collapsing features. Correlation analysis reveals that the association between end-to-end performance and utilization is robust across datasets but can vary by initialization, while geometric quantities capture additional prior- and dataset-conditioned variation. These findings position DP-RGMI as a reproducible framework for diagnosing privacy-induced failure modes and informing privacy model selection.
Authors: Wael Hafez, Cameron Reid, Amit Nazeri
Abstract: Deployed RL agents operate in closed-loop systems where reliable performance depends on maintaining coherent coupling between observations, actions, and outcomes. Current monitoring approaches rely on reward and task metrics, measures that are reactive by design and blind to structural degradation that precedes performance collapse. We argue that deployment monitoring is fundamentally a question about uncertainty resolution: whether the agent's observations and actions continue to reduce uncertainty about outcomes, and whether outcomes constrain what the agent must have done. Information theory directly operationalizes this question, entropy quantifies uncertainty, and mutual information quantifies its resolution across the loop. We introduce Bipredictability (P), the fraction of the total uncertainty budget converted into shared predictability across the observation, action, outcome loop. A theoretical property is a provable classical upper bound P is less than or equal to 0.5, independent of domain, task, or agent, a structural consequence of Shannon entropy rather than an empirical observation. When agency is present, a penalty suppresses P strictly below this ceiling, confirmed at P equals 0.33 across trained agents. To operationalize P as a real time monitoring signal, we introduce the Information Digital Twin (IDT), an auxiliary architecture that computes P and its directional components from the observable interaction stream without access to model internals. Across 168 perturbation trials spanning eight perturbation types and two policy architectures, IDT based monitoring detected 89.3 percent of coupling degradations versus 44.0 percent for reward based monitoring, with 4.4 times lower median latency. These results establish Bipredictability as a principled, bounded, and computable prerequisite signal for closed loop self regulation in deployed reinforcement learning systems.
Authors: Nicola Bariletto, Stephen G. Walker
Abstract: We introduce a novel framework for uncertainty quantification in clustering that combines martingale posterior distributions with density-based clustering. Unlike classical model-based approaches, which define clusters at the latent level of a mixture model, we treat clusters as explicit functionals of the data-generating density, without assuming any specific parametric form. To characterize density uncertainty, we obtain martingale posterior samples via a predictive resampling scheme driven by model score evaluations. This allows us to leverage state-of-the-art differentiable density estimators, such as normalizing flows, making density resampling efficient in large-scale settings and fully parallelizable on modern GPU hardware. Martingale posterior samples of the clustering structure are then obtained by applying density-based clustering to the density draws, enabling principled inference on any clustering-related quantity. Casting the inference target as a density functional further enables a rigorous theoretical analysis of the procedure's convergence properties. We apply our methodology to image and single-cell RNA sequencing data, demonstrating the computational efficiency afforded by its GPU compatibility as well as its ability to recover meaningful clustering structures, with associated uncertainty, across diverse domains.
Authors: Ashwath Vaithinathan Aravindan, Mayank Kejriwal
Abstract: Chain-of-Thought (CoT) prompting has emerged as a foundational technique for eliciting reasoning from Large Language Models (LLMs), yet the robustness of this approach to corruptions in intermediate reasoning steps remains poorly understood. This paper presents a comprehensive empirical evaluation of LLM robustness to a structured taxonomy of 5 CoT perturbation types: \textit{MathError, UnitConversion, Sycophancy, SkippedSteps,} and \textit{ExtraSteps}. We evaluate 13 models spanning three orders of magnitude in parameter count, testing their ability to complete mathematical reasoning tasks despite perturbations injected in the reasoning chain. Our key findings reveal heterogeneous vulnerability patterns: MathError perturbations produce the most severe degradation in small models (50-60\% accuracy loss) but show strong scaling benefits; UnitConversion remains challenging across all scales (>5\% loss even for midsized models); ExtraSteps incur minimal accuracy degradation (0-6\%) even for the smallest of models; Sycophancy and SkippedSteps produce modest effects ($\sim$10\% loss for small models) and slightly improve with scale. Scaling relationships show that model size serve as a protective factor against many perturbations but not always. These findings have direct implications for deploying LLMs in multi-stage reasoning pipelines and underscore the necessity of task-specific robustness assessments and mitigation strategies. The code and results are available at https://github.com/Mystic-Slice/CoTPerturbation
Authors: Zongyue Qin, Raghavv Goel, Mukul Gagrani, Risheek Garrepalli, Mingu Lee, Yizhou Sun
Abstract: Speculative decoding has emerged as a powerful approach to accelerate large language model (LLM) inference by employing lightweight draft models to propose candidate tokens that are subsequently verified by the target model. The effectiveness of this paradigm critically depends on the quality of the draft model. While recent advances such as the EAGLE series achieve state-of-the-art speedup, existing draft models remain limited by error accumulation: they condition only on the current prefix, causing their predictions to drift from the target model over steps. In this work, we propose \textbf{ConFu} (Contemplate the Future), a novel speculative decoding framework that enables draft models to anticipate the future direction of generation. ConFu introduces (i) contemplate tokens and soft prompts that allow the draft model to leverage future-oriented signals from the target model at negligible cost, (ii) a dynamic contemplate token mechanism with MoE to enable context-aware future prediction, and (iii) a training framework with anchor token sampling and future prediction replication that learns robust future prediction. ConFu improves token acceptance rates and generation speed over EAGLE-3 by 8--11\% on Llama-3 3B/8B and by approximately 20\% on Qwen-3 4B across downstream tasks. We believe our work is the first to bridge speculative decoding with continuous reasoning tokens, offering a new direction for accelerating LLM inference.
Authors: Chanhyuk Choi, Taesoo Kim, Donggyu Lee, Siyeol Jung, Taehwan Kim
Abstract: Talking face generation has gained significant attention as a core application of generative models. To enhance the expressiveness and realism of synthesized videos, emotion editing in talking face video plays a crucial role. However, existing approaches often limit expressive flexibility and struggle to generate extended emotions. Label-based methods represent emotions with discrete categories, which fail to capture a wide range of emotions. Audio-based methods can leverage emotionally rich speech signals - and even benefit from expressive text-to-speech (TTS) synthesis - but they fail to express the target emotions because emotions and linguistic contents are entangled in emotional speeches. Images-based methods, on the other hand, rely on target reference images to guide emotion transfer, yet they require high-quality frontal views and face challenges in acquiring reference data for extended emotions (e.g., sarcasm). To address these limitations, we propose Cross-Modal Emotion Transfer (C-MET), a novel approach that generates facial expressions based on speeches by modeling emotion semantic vectors between speech and visual feature spaces. C-MET leverages a large-scale pretrained audio encoder and a disentangled facial expression encoder to learn emotion semantic vectors that represent the difference between two different emotional embeddings across modalities. Extensive experiments on the MEAD and CREMA-D datasets demonstrate that our method improves emotion accuracy by 14% over state-of-the-art methods, while generating expressive talking face videos - even for unseen extended emotions. Code, checkpoint, and demo are available at https://chanhyeok-choi.github.io/C-MET/
Authors: Yannis Montreuil, Axel Carlier, Lai Xing Ng, Wei Tsang Ooi
Abstract: Existing multi-expert learning-to-defer surrogates are statistically consistent, yet they can underfit, suppress useful experts, or degrade as the expert pool grows. We trace these failures to a shared architectural choice: casting classes and experts as actions inside one augmented prediction geometry. Consistency governs the population target; it says nothing about how the surrogate distributes gradient mass during training. We analyze five surrogates along both axes and show that each trades a fix on one for a failure on the other. We then introduce a decoupled surrogate that estimates the class posterior with a softmax and each expert utility with an independent sigmoid. It admits an $\mathcal{H}$-consistency bound whose constant is $J$-independent for fixed per-expert weight $\beta{=}\lambda/J$, and its gradients are free of the amplification, starvation, and coupling pathologies of the augmented family. Experiments on synthetic benchmarks, CIFAR-10, CIFAR-10H, and Covertype confirm that the decoupled surrogate is the only method that avoids amplification under redundancy, preserves rare specialists, and consistently improves over a standalone classifier across all settings.
Authors: Kyle Waters, Lucas Nuzzi, Tadhg Looram, Alessandro Tomasiello, Ariel Ghislain Kemogne Kamdoum, Bikun Li, Damien Sileo, Egor Kretov, Francesco Fournier-Facio, Georgios Soloupis, Haile Kassahun, Hew Wolff, Jiaqi Cai, Lianghui Li, Marc Roth, Mohinder Naiya, Naixu Guo, Qicheng Tang, Richard Wheeler, Samuele Sala, Serguei Popov, Steven Dillmann, Yuqi Li
Abstract: AI agents hold growing promise for accelerating scientific discovery; yet, a lack of frontier evaluations hinders adoption into real workflows. Expert-written benchmarks have proven effective at measuring AI reasoning, but most at this stage have become saturated and only measure performance on constrained outputs. To help address this gap, we introduce COMPOSITE-STEM, a benchmark of 70 expert-written tasks in physics, biology, chemistry, and mathematics, curated by doctoral-level researchers. Our benchmark combines exact-match grading and criterion-based rubrics with an LLM-as-a-jury grading protocol, allowing more flexible assessment of scientifically meaningful outputs. Using an adapted multimodal Terminus-2 agent harness within the Harbor agentic evaluation framework, we evaluate four frontier models. The top-performing model achieves 21%, demonstrating that COMPOSITE-STEM captures capabilities beyond current agent reach. All tasks are open-sourced with contributor permission to support reproducibility and to promote additional research towards AI's acceleration of scientific progress in these domains.
Authors: Zae Myung Kim, Dongseok Lee, Jaehyung Kim, Vipul Raheja, Dongyeop Kang
Abstract: Existing tool-use benchmarks for LLM agents are overwhelmingly linear: our analysis of six benchmarks shows 55 to 100% of instances are simple chains of 2 to 5 steps. We introduce The Amazing Agent Race (AAR), a benchmark featuring directed acyclic graph (DAG) puzzles (or "legs") with fork-merge tool chains. We release 1,400 instances across two variants: sequential (800 legs) and compositional (600 DAG legs). Agents must navigate Wikipedia, execute multi-step tool chains, and aggregate results into a verifiable answer. Legs are procedurally generated from Wikipedia seeds across four difficulty levels with live-API validation. Three complementary metrics (finish-line accuracy, pit-stop visit rate, and roadblock completion rate) separately diagnose navigation, tool-use, and arithmetic failures. Evaluating three agent frameworks on 1,400 legs, the best achieves only 37.2% accuracy. Navigation errors dominate (27 to 52% of trials) while tool-use errors remain below 17%, and agent architecture matters as much as model scale (Claude Code matches Codex CLI at 37% with 6x fewer tokens). The compositional structure of AAR reveals that agents fail not at calling tools but at navigating to the right pages, a blind spot invisible to linear benchmarks. The project page can be accessed at: https://minnesotanlp.github.io/the-amazing-agent-race
Authors: Olga Chetverina
Abstract: Standard subword tokenization methods fragment numbers inconsistently, causing large language models (LLMs) to lose positional and decimal structure - a primary driver of errors in arithmetic and scientific reasoning. We introduce Triadic Suffix Tokenization (TST), a deterministic scheme that partitions digits into three-digit triads and annotates each triad with an explicit magnitude marker. Critically, the scheme defines a fixed, one-to-one mapping between suffixes and orders of magnitude for the integer part (thousands, millions, billions, etc.) and a parallel system of replicated markers for fractional depth (tenths, thousandths, millionths, etc.). Unlike approaches that rely on positional inference, this method provides a consistent gradient signal, which should ensure stable convergence. Two implementation variants are proposed: (1) a vocabulary-based approach that adds at most 10,000 fixed tokens to an existing vocabulary, covering 33 orders of magnitude ($10^{-15}$ to $10^{18}$); and (2) a suffix-marker approach that uses a small set of special tokens to denote magnitude dynamically. Both variants preserve exact digits while making order-of-magnitude relationships transparent at the token level. While we focus on 3-digit groups (Triadic), the framework is inherently scalable to any group size for precise vocabulary optimization. Furthermore, it allows for linear vocabulary expansion to accommodate arbitrary precision and range. TST is architecture-agnostic and can be integrated as a drop-in preprocessing step. Experimental validation is deferred to future work.
Authors: Nahyun Lee, Guijin Son, Hyunwoo Ko, Chanyoung Kim, JunYoung An, Kyubeen Han, Il-Youp Kwak
Abstract: We introduce KMMMU, a native Korean benchmark for evaluating multimodal understanding in Korean cultural and institutional settings. KMMMU contains 3,466 questions from exams natively written in Korean, covering nine disciplines and nine visual modality categories, along with a 300-item Korean-specific subset and a hard subset of 627 questions. Unlike translated or English-centric benchmarks, KMMMU targets information-dense problems shaped by local conventions, official standards, and discipline-specific visual formats. Experiments show that the strongest open-source model reaches only 42.05% accuracy on the full set, while the best proprietary model achieves 52.42% on the hard subset. Performance varies across disciplines, with some disciplines emerging as bottlenecks, and Korean-specific questions showing gaps of up to 13.43%. Error analysis suggests that these failures stem less from insufficient reasoning depth than from weak convention-to-label mapping, few-shot symbolic induction, localized knowledge recall, and domain-specific standards understanding. KMMMU provides a testbed for multimodal evaluation beyond English-centric benchmarks and for developing more reliable systems for expert real-world tasks.
Authors: Bryan Sanchez
Abstract: Alignment-tuned language models frequently suppress factual log-probabilities on politically sensitive topics despite retaining the knowledge in their hidden representations. We show that a 786K-parameter (approximately 0.02% of the base model) post-transformer adapter, trained on frozen hidden states, corrects this suppression on 31 ideology-discriminating facts across Qwen3-4B, 8B, and 14B. The adapter memorizes all 15 training facts and generalizes to 11--39% of 16 held-out facts across 5 random splits per scale, with zero knowledge regressions via anchored training. Both gated (SwiGLU) and ungated (linear bottleneck) adapters achieve comparable results; neither consistently outperforms the other (Fisher exact p > 0.09 at all scales). On instruct models, the adapter corrects log-probability rankings. When applied at all token positions during generation, the adapter produces incoherent output; however, when applied only at the current prediction position (last-position-only), the adapter produces coherent, less censored text. A logit-space adapter operating after token projection fails to produce coherent generation at any application mode, suggesting hidden-state intervention is the correct level for generation correction. A previously undocumented silent gradient bug in Apple MLX explains all null results in earlier iterations of this work: the standard pattern nn.value_and_grad(model, fn)(model.parameters()) returns zero gradients without error; the correct pattern nn.value_and_grad(model, fn)(model, data) resolves this. We provide a minimal reproduction and discuss implications for other adapter research using MLX.
Authors: Aditi Agrawal, Celine John Philip, Giancarlo K. Sagastume, Marcus A. Battraw, Wilsaan M. Joiner, Jonathon S. Schofield, Lee M. Miller, Richard S. Whittle
Abstract: Neuromotor decoding from upper-limb electromyography (sEMG) can enhance human-machine interfaces and offer a more natural means of controlling prosthetic limbs, virtual reality, and household electronics. Unfortunately, current sEMG technology does not always perform consistently across users because individual differences such as age and body mass index, among many others, can substantially alter signal quality. This variability makes sEMG characteristics highly idiosyncratic, often necessitating laborious personalization and iterative tuning to achieve reliable performance. This variability has particular import for sEMG-based assistive devices and neural interfaces, where demographic biases in sEMG features could undermine broad and fair deployment. In this study, we explore how demographic differences affect the sEMG signals produced and their implications for machine learning-based gesture decoding. We analyze the data set provided by, in which we derive 147 common sEMG features extracted from 81 demographically diverse individuals performing discrete hand gestures. Using mixed-effects linear models and partial least squares (PLS) analysis, which take into consideration demographic variables (including age, sex, height, weight, skin properties, subcutaneous fat, and hair density), we identify that 33\% (49 of 147) of commonly used sEMG features show significant associations with demographic characteristics. These results may help guide the development of fair and unbiased sEMG-based neural interfaces across a diverse population.
Authors: Elies Gil-Fuster, Seongwook Shin, Sofiene Jerbi, Jens Eisert, Maximilian J. Kramer
Abstract: Quantum kernel methods are among the leading candidates for achieving quantum advantage in supervised learning. A key bottleneck is the cost of inference: evaluating a trained model on new data requires estimating a weighted sum $\sum_{i=1}^N \alpha_i k(x,x_i)$ of $N$ kernel values to additive precision $\varepsilon$, where $\alpha$ is the vector of trained coefficients. The standard approach estimates each term independently via sampling, yielding a query complexity of $O(N\lVert\alpha\rVert_2^2/\varepsilon^2)$. In this work, we identify two independent axes for improvement: (1) How individual kernel values are estimated (sampling versus quantum amplitude estimation), and (2) how the sum is approximated (term-by-term versus via a single observable), and systematically analyze all combinations thereof. The query-optimal combination, encoding the full inference sum as the expectation value of a single observable and applying quantum amplitude estimation, achieves a query complexity of $O(\lVert\alpha\rVert_1/\varepsilon)$, removing the dependence on $N$ from the query count and yielding a quadratic improvement in both $\lVert\alpha\rVert_1$ and $\varepsilon$. We prove a matching lower bound of $\Omega(\lVert\alpha\rVert_1/\varepsilon)$, establishing query-optimality of our approach up to logarithmic factors. Beyond query complexity, we also analyze how these improvements translate into gate costs and show that the query-optimal strategy is not always optimal in practice from the perspective of gate complexity. Our results provide both a query-optimal algorithm and a practically optimal choice of strategy depending on hardware capabilities, along with a complete landscape of intermediate methods to guide practitioners. All algorithms require only amplitude estimation as a subroutine and are thus natural candidates for early-fault-tolerant implementations.