Authors: Jiyuan Zhang, Yining Liu, Siqi Yan, Lisen Deng, Jennifer Cao, Shuqi Yang, Min Ni, Bi Xue, Shen Li
Abstract: The pervasive "memory wall" bottleneck is significantly amplified in modern large-scale Mixture-of-Experts (MoE) architectures. MoE's inherent architectural sparsity leads to sparse arithmetic compute and also introduces substantial activation memory overheads -- driven by large token routing buffers and the need to materialize and buffer intermediate tensors. This memory pressure limits the maximum batch size and sequence length that can fit on GPUs, and also results in excessive data movements that hinders performance and efficient model scaling. We present MoEBlaze, a memory-efficient MoE training framework that addresses these issues through a co-designed system approach: (i) an end-to-end token dispatch and MoE training method with optimized data structures to eliminate intermediate buffers and activation materializing, and (ii) co-designed kernels with smart activation checkpoint to mitigate memory footprint while simultaneously achieving better performance. We demonstrate that MoEBlaze can achieve over 4x speedups and over 50% memory savings compared to existing MoE frameworks.
Authors: Susmit Das
Abstract: Reasoning oriented large language models often expose explicit "thinking" as long, turn-global traces at the start of every response, either always on or toggled externally at inference time. While useful for arithmetic, programming, and problem solving, this design is costly, blurs claim level auditability, and cannot re-trigger explicit reasoning once the model begins presenting. Dialogue models are also largely blind to temporal structure, treating replies after seconds and replies after weeks as equivalent unless time is stated in text. We introduce TIME, the Temporally Intelligent Meta-reasoning Engine, a behavioral alignment framework that treats explicit reasoning as a context sensitive resource driven by discourse and temporal cues. TIME augments dialogue with optional ISO 8601
URLs: https://github.com/The-Coherence-Initiative/TIME, https://github.com/The-Coherence-Initiative/TIMEBench
Authors: Jaehong Oh
Abstract: Neuro-symbolic reasoning systems face fundamental challenges in maintaining semantic coherence while satisfying physical and logical constraints. Building upon our previous work on Ontology Neural Networks, we present an enhanced framework that integrates topological conditioning with gradient stabilization mechanisms. The approach employs Forman-Ricci curvature to capture graph topology, Deep Delta Learning for stable rank-one perturbations during constraint projection, and Covariance Matrix Adaptation Evolution Strategy for parameter optimization. Experimental evaluation across multiple problem sizes demonstrates that the method achieves mean energy reduction to 1.15 compared to baseline values of 11.68, with 95 percent success rate in constraint satisfaction tasks. The framework exhibits seed-independent convergence and graceful scaling behavior up to twenty-node problems, suggesting that topological structure can inform gradient-based optimization without sacrificing interpretability or computational efficiency.
Authors: Tianrun Yu, Kaixiang Zhao, Cheng Zhang, Anjun Gao, Yueyang Quan, Zhuqing Liu, Minghong Fang
Abstract: Federated learning (FL) has emerged as a transformative distributed learning paradigm, enabling multiple clients to collaboratively train a global model under the coordination of a central server without sharing their raw training data. While FL offers notable advantages, it faces critical challenges in ensuring fairness across diverse demographic groups. To address these fairness concerns, various fairness-aware debiasing methods have been proposed. However, many of these approaches either require modifications to clients' training protocols or lack flexibility in their aggregation strategies. In this work, we address these limitations by introducing EquFL, a novel server-side debiasing method designed to mitigate bias in FL systems. EquFL operates by allowing the server to generate a single calibrated update after receiving model updates from the clients. This calibrated update is then integrated with the aggregated client updates to produce an adjusted global model that reduces bias. Theoretically, we establish that EquFL converges to the optimal global model achieved by FedAvg and effectively reduces fairness loss over training rounds. Empirically, we demonstrate that EquFL significantly mitigates bias within the system, showcasing its practical effectiveness.
Authors: Shovito Barua Soumma, Hassan Ghasemzadeh
Abstract: Accurate forecasting of blood glucose from CGM is essential for preventing dysglycemic events, thus enabling proactive diabetes management. However, current forecasting models treat blood glucose readings captured using CGMs as a numerical sequence, either ignoring context or relying on additional sensors/modalities that are difficult to collect and deploy at scale. Recently, LLMs have shown promise for time-series forecasting tasks, yet their role as agentic context extractors in diabetes care remains largely unexplored. To address these limitations, we propose GlyRAG, a context-aware, retrieval-augmented forecasting framework that derives semantic understanding of blood glucose dynamics directly from CGM traces without requiring additional sensor modalities. GlyRAG employs an LLM as a contextualization agent to generate clinical summaries. These summaries are embedded by a language model and fused with patch-based glucose representations in a multimodal transformer architecture with a cross translation loss aligining textual and physiological embeddings. A retrieval module then identifies similar historical episodes in the learned embedding space and uses cross-attention to integrate these case-based analogues prior to making a forecasting inference. Extensive evaluations on two T1D cohorts show that GlyRAG consistently outperforms state-of-the art methods, achieving up to 39% lower RMSE and a further 1.7% reduction in RMSE over the baseline. Clinical evaluation shows that GlyRAG places 85% predictions in safe zones and achieves 51% improvement in predicting dysglycemic events across both cohorts. These results indicate that LLM-based contextualization and retrieval over CGM traces can enhance the accuracy and clinical reliability of long-horizon glucose forecasting without the need for extra sensors, thus supporting future agentic decision-support tools for diabetes management.
Authors: Md Shafiqul Islam, Shakti Prasad Padhy, Douglas Allaire, Raymundo Arr\'oyave
Abstract: Gaussian Process (GP) regression is a powerful nonparametric Bayesian framework, but its performance depends critically on the choice of covariance kernel. Selecting an appropriate kernel is therefore central to model quality, yet remains one of the most challenging and computationally expensive steps in probabilistic modeling. We present a Bayesian optimization framework built on kernel-of-kernels geometry, using expected divergence-based distances between GP priors to explore kernel space efficiently. A multidimensional scaling (MDS) embedding of this distance matrix maps a discrete kernel library into a continuous Euclidean manifold, enabling smooth BO. In this formulation, the input space comprises kernel compositions, the objective is the log marginal likelihood, and featurization is given by the MDS coordinates. When the divergence yields a valid metric, the embedding preserves geometry and produces a stable BO landscape. We demonstrate the approach on synthetic benchmarks, real-world time-series datasets, and an additive manufacturing case study predicting melt-pool geometry, achieving superior predictive accuracy and uncertainty calibration relative to baselines including Large Language Model (LLM)-guided search. This framework establishes a reusable probabilistic geometry for kernel search, with direct relevance to GP modeling and deep kernel learning.
Authors: Sebastian J. Wetzel
Abstract: Non-injective functions are not invertible. However, non-injective functions can be restricted to sub-domains on which they are locally injective and surjective and thus invertible if the dimensionality between input and output spaces are the same. Further, even if the dimensionalities do not match it is often possible to choose a preferred solution from many possible solutions. Twin neural network regression is naturally capable of incorporating these properties to invert non-injective functions. Twin neural network regression is trained to predict adjustments to well known input variables $\mathbf{x}^{\text{anchor}}$ to obtain an estimate for an unknown $\mathbf{x}^{\text{new}}$ under a change of the target variable from $\mathbf{y}^{\text{anchor}}$ to $\mathbf{y}^{\text{new}}$. In combination with k-nearest neighbor search, I propose a deterministic framework that finds input parameters to a given target variable of non-injective functions. The method is demonstrated by inverting non-injective functions describing toy problems and robot arm control that are a) defined by data or b) known as mathematical formula.
Authors: Prakash Gawas, Antoine Legrain, Louis-Martin Rousseau
Abstract: Imitation learning (IL) provides a data-driven framework for approximating policies for large-scale combinatorial optimisation problems formulated as sequential decision problems (SDPs), where exact solution methods are computationally intractable. A central but underexplored aspect of IL in this context is the role of the \emph{expert} that generates training demonstrations. Existing studies employ a wide range of expert constructions, yet lack a unifying framework to characterise their modelling assumptions, computational properties, and impact on learning performance. This paper introduces a systematic taxonomy of experts for IL in combinatorial optimisation under uncertainty. Experts are classified along three dimensions: (i) their treatment of uncertainty, including myopic, deterministic, full-information, two-stage stochastic, and multi-stage stochastic formulations; (ii) their level of optimality, distinguishing task-optimal and approximate experts; and (iii) their interaction mode with the learner, ranging from one-shot supervision to iterative, interactive schemes. Building on this taxonomy, we propose a generalised Dataset Aggregation (DAgger) algorithm that supports multiple expert queries, expert aggregation, and flexible interaction strategies. The proposed framework is evaluated on a dynamic physician-to-patient assignment problem with stochastic arrivals and capacity constraints. Computational experiments compare learning outcomes across expert types and interaction regimes. The results show that policies learned from stochastic experts consistently outperform those learned from deterministic or full-information experts, while interactive learning improves solution quality using fewer expert demonstrations. Aggregated deterministic experts provide an effective alternative when stochastic optimisation becomes computationally challenging.
Authors: Namrata Banerji, Tanya Berger-Wolf
Abstract: Accurate multistep forecasting of node-level attributes on dynamic graphs is critical for applications ranging from financial trust networks to biological networks. Existing spatiotemporal graph neural networks typically assume a static adjacency matrix. In this work, we propose an end-to-end dynamic edge-biased spatiotemporal model that ingests a multi-dimensional timeseries of node attributes and a timeseries of adjacency matrices, to predict multiple future steps of node attributes. At each time step, our transformer-based model injects the given adjacency as an adaptable attention bias, allowing the model to focus on relevant neighbors as the graph evolves. We further deploy a masked node-time pretraining objective that primes the encoder to reconstruct missing features, and train with scheduled sampling and a horizon-weighted loss to mitigate compounding error over long horizons. Unlike prior work, our model accommodates dynamic graphs that vary across input samples, enabling forecasting in multi-system settings such as brain networks across different subjects, financial systems in different contexts, or evolving social systems. Empirical results demonstrate that our method consistently outperforms strong baselines on Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE).
Authors: Minwoo Cho, Batuhan Altundas, Matthew Gombolay
Abstract: Knowledge distillation (KD) has the potential to accelerate MARL by employing a centralized teacher for decentralized students but faces key bottlenecks. Specifically, there are (1) challenges in synthesizing high-performing teaching policies in complex domains, (2) difficulties when teachers must reason in out-of-distribution (OOD) states, and (3) mismatches between the decentralized students' and the centralized teacher's observation spaces. To address these limitations, we propose HINT (Hierarchical INteractive Teacher-based transfer), a novel KD framework for MARL in a centralized training, decentralized execution setup. By leveraging hierarchical RL, HINT provides a scalable, high-performing teacher. Our key innovation, pseudo off-policy RL, enables the teacher policy to be updated using both teacher and student experience, thereby improving OOD adaptation. HINT also applies performance-based filtering to retain only outcome-relevant guidance, reducing observation mismatches. We evaluate HINT on challenging cooperative domains (e.g., FireCommander for resource allocation, MARINE for tactical combat). Across these benchmarks, HINT outperforms baselines, achieving improvements of 60% to 165% in success rate.
Authors: Yiqun T Chen, Sizhu Lu, Sijia Li, Moran Guo, Shengyi Li
Abstract: Large language models (LLMs) are increasingly used as automatic evaluators of generative AI outputs, a paradigm often referred to as "LLM-as-a-judge." In practice, LLM judges are imperfect predictions for the underlying truth and can exhibit systematic, non-random errors. Two main approaches have recently been proposed to address this issue: (i) direct measurementerror correction based on misclassification models such as Rogan-Gladen-style estimators, and (ii) surrogate-outcome approaches such as prediction-powered inference (PPI), which correct bias by calibrating prediction residuals on a small set of gold-standard human labels. In this paper, we systematically study the performance of these two approaches for estimating mean parameters (e.g., average benchmark scores or pairwise win rates). Leveraging tools from semiparametric efficiency theory, we unify the two classes of estimators by deriving explicit forms of efficient influence function (EIF)-based efficient estimators and characterize conditions under which PPI-style estimators attain strictly smaller asymptotic variance than measurement-error corrections. We verify our theoretical results in simulations and demonstrate the methods on real-data examples. We provide an implementation of the benchmarked methods and comparison utilities at https://github.com/yiqunchen/debias-llm-as-a-judge.
Authors: Xiaowen He, Su Jiang, Louis J. Durlofsky
Abstract: Accurately assessing the potential for fault slip is essential in many subsurface operations. Conventional model-based history matching methods, which entail the generation of posterior geomodels calibrated to observed data, can be challenging to apply in coupled flow-geomechanics problems with faults. In this work, we implement a variational autoencoder (VAE)-based data-space inversion (DSI) framework to predict pressure, stress and strain fields, and fault slip tendency, in CO${_2}$ storage projects. The main computations required by the DSI workflow entail the simulation of O(1000) prior geomodels. The posterior distributions for quantities of interest are then inferred directly from prior simulation results and observed data, without the need to generate posterior geomodels. The model used here involves a synthetic 3D system with two faults. Realizations of heterogeneous permeability and porosity fields are generated using geostatistical software, and uncertain geomechanical and fault parameters are sampled for each realization from prior distributions. Coupled flow-geomechanics simulations for these geomodels are conducted using GEOS. A VAE with stacked convolutional long short-term memory layers is trained, using the prior simulation results, to represent pressure, strain, effective normal stress and shear stress fields in terms of latent variables. The VAE parameterization is used with DSI for posterior predictions, with monitoring wells providing observed pressure and strain data. Posterior results for synthetic true models demonstrate that the DSI-VAE framework gives accurate predictions for pressure, strain, and stress fields and for fault slip tendency. The framework is also shown to reduce uncertainty in key geomechanical and fault parameters.
Authors: Marko Sterbentz, Kevin Cushing, Cameron Barrie, Kristian J. Hammond
Abstract: Recent advances in text-to-SQL systems have been driven by larger models and improved datasets, yet progress is still limited by the scarcity of high-quality training data. Manual data creation is expensive, and existing synthetic methods trade off reliability and scalability. Template-based approaches ensure correct SQL but require schema-specific templates, while LLM-based generation scales easily but lacks quality and correctness guarantees. We introduce RingSQL, a hybrid data generation framework that combines schema-independent query templates with LLM-based paraphrasing of natural language questions. This approach preserves SQL correctness across diverse schemas while providing broad linguistic variety. In our experiments, we find that models trained using data produced by RingSQL achieve an average gain in accuracy of +2.3% across six text-to-SQL benchmarks when compared to models trained on other synthetic data. We make our code available at https://github.com/nu-c3lab/RingSQL.
Authors: Pingchuan Ma, Qixin Zhang, Shuai Wang, Dacheng Tao
Abstract: Recently, differentiable causal discovery has emerged as a promising approach to improve the accuracy and efficiency of existing methods. However, when applied to high-dimensional data or data with latent confounders, these methods, often based on off-the-shelf continuous optimization algorithms, struggle with the vast search space, the complexity of the objective function, and the nontrivial nature of graph-theoretical constraints. As a result, there has been a surge of interest in leveraging super-structures to guide the optimization process. Nonetheless, learning an appropriate super-structure at the right level of granularity, and doing so efficiently across various settings, presents significant challenges. In this paper, we propose ALVGL, a novel and general enhancement to the differentiable causal discovery pipeline. ALVGL employs a sparse and low-rank decomposition to learn the precision matrix of the data. We design an ADMM procedure to optimize this decomposition, identifying components in the precision matrix that are most relevant to the underlying causal structure. These components are then combined to construct a super-structure that is provably a superset of the true causal graph. This super-structure is used to initialize a standard differentiable causal discovery method with a more focused search space, thereby improving both optimization efficiency and accuracy. We demonstrate the versatility of ALVGL by instantiating it across a range of structural causal models, including both Gaussian and non-Gaussian settings, with and without unmeasured confounders. Extensive experiments on synthetic and real-world datasets show that ALVGL not only achieves state-of-the-art accuracy but also significantly improves optimization efficiency, making it a reliable and effective solution for differentiable causal discovery.
Authors: Jiefu Ou, Sapana Chaudhary, Kaj Bostrom, Nathaniel Weir, Shuai Zhang, Huzefa Rangwala, George Karypis
Abstract: Large Language Models (LLMs) demonstrate strong capabilities in general coding tasks but encounter two key challenges when optimizing code: (i) the complexity of writing optimized code (such as performant CUDA kernels and competition-level CPU code) requires expertise in systems, algorithms and specific languages and (ii) requires interpretation of performance metrics like timing and device utilization beyond binary correctness. In this work, we explore inference-time search algorithms that guide the LLM to discover better solutions through iterative refinement based on execution feedback. Our approach, called MaxCode unifies existing search methods under a max-reward reinforcement learning framework, making the observation and action-value functions modular for modification. To enhance the observation space, we integrate a natural language critique model that converts raw execution feedback into diagnostic insights about errors and performance bottlenecks, and the best-discounted reward seen so far. Together, these provide richer input to the code proposal function. To improve exploration during search, we train a generative reward-to-go model using action values from rollouts to rerank potential solutions. Testing on the KernelBench (CUDA) and PIE (C++) optimization benchmarks shows that MaxCode improves optimized code performance compared to baselines, achieving 20.3% and 10.1% relative improvements in absolute speedup value and relative speedup ranking, respectively.
Authors: Feihu Jin, Ying Tan
Abstract: Fine-tuning large language models (LLMs) using standard first-order (FO) optimization often drives training toward sharp, poorly generalizing minima. Conversely, zeroth-order (ZO) methods offer stronger exploratory behavior without relying on explicit gradients, yet suffer from slow convergence. More critically, our analysis reveals that in generative tasks, the vast output and search space significantly amplify estimation variance, rendering ZO methods both noisy and inefficient. To address these challenges, we propose \textbf{Hi-ZFO} (\textbf{Hi}erarchical \textbf{Z}eroth- and \textbf{F}irst-\textbf{O}rder optimization), a hybrid framework designed to synergize the precision of FO gradients with the exploratory capability of ZO estimation. Hi-ZFO adaptively partitions the model through layer-wise importance profiling, applying precise FO updates to critical layers while leveraging ZO optimization for less sensitive ones. Notably, ZO in Hi-ZFO is not merely a memory-saving surrogate; it is intentionally introduced as a source of "beneficial stochasticity" to help the model escape the local minima where pure FO optimization tends to stagnate. Validated across diverse generative, mathematical, and code reasoning tasks, Hi-ZFO consistently achieves superior performance while significantly reducing the training time. These results demonstrate the effectiveness of hierarchical hybrid optimization for LLM fine-tuning.
Authors: Roy Xie, Deepak Gopinath, David Qiu, Dong Lin, Haitian Sun, Saloni Potdar, Bhuwan Dhingra
Abstract: Search-augmented large language models (LLMs) excel at knowledge-intensive tasks by integrating external retrieval. However, they often over-search -- unnecessarily invoking search tool even when it does not improve response quality, which leads to computational inefficiency and hallucinations by incorporating irrelevant context. In this work, we conduct a systematic evaluation of over-searching across multiple dimensions, including query types, model categories, retrieval conditions, and multi-turn conversations. Our finding shows: (i) search generally improves answer accuracy on answerable queries but harms abstention on unanswerable ones; (ii) over-searching is more pronounced in complex reasoning models and deep research systems, is exacerbated by noisy retrieval, and compounds across turns in multi-turn conversations; and (iii) the composition of retrieved evidence is crucial, as the presence of negative evidence improves abstention. To quantify over-searching, we introduce Tokens Per Correctness (TPC), an evaluation metric that captures the performance-cost trade-off for search-augmented LLMs. Lastly, we investigate mitigation approaches at both the query and retrieval levels and release the OverSearchQA to foster continued research into efficient search-augmented LLMs.
Authors: Jiayu Fang, Zhiqi Shao, Haoning Xi, Boris Choy, Junbin Gao
Abstract: The development of a cross-city accident prevention system is particularly challenging due to the heterogeneity, inconsistent reporting, and inherently clustered, sparse, cyclical, and noisy nature of urban accident data. These intrinsic data properties, combined with fragmented governance and incompatible reporting standards, have long hindered the creation of an integrated, cross-city accident prevention framework. To address this gap, we propose the Mamba Local-ttention Spatial-Temporal Network MLA-STNet, a unified system that formulates accident risk prediction as a multi-task learning problem across multiple cities. MLA-STNet integrates two complementary modules: (i)the Spatio-Temporal Geographical Mamba-Attention (STG-MA), which suppresses unstable spatio-temporal fluctuations and strengthens long-range temporal dependencies; and (ii) the Spatio-Temporal Semantic Mamba-Attention (STS-MA), which mitigates cross-city heterogeneity through a shared-parameter design that jointly trains all cities while preserving individual semantic representation spaces. We validate the proposed framework through 75 experiments under two forecasting scenarios, full-day and high-frequency accident periods, using real-world datasets from New York City and Chicago. Compared with the state-of-the-art baselines, MLA-STNet achieves up to 6% lower RMSE, 8% higher Recall, and 5% higher MAP, while maintaining less than 1% performance variation under 50% input noise. These results demonstrate that MLA-STNet effectively unifies heterogeneous urban datasets within a scalable, robust, and interpretable Cross-City Accident Prevention System, paving the way for coordinated and data-driven urban safety management.
Authors: Rui An, Haohao Qu, Wenqi Fan, Xuequn Shang, Qing Li
Abstract: Accurate and efficient multivariate time series (MTS) analysis is increasingly critical for a wide range of intelligent applications. Within this realm, Transformers have emerged as the predominant architecture due to their strong ability to capture pairwise dependencies. However, Transformer-based models suffer from quadratic computational complexity and high memory overhead, limiting their scalability and practical deployment in long-term and large-scale MTS modeling. Recently, Mamba has emerged as a promising linear-time alternative with high expressiveness. Nevertheless, directly applying vanilla Mamba to MTS remains suboptimal due to three key limitations: (i) the lack of explicit cross-variate modeling, (ii) difficulty in disentangling the entangled intra-series temporal dynamics and inter-series interactions, and (iii) insufficient modeling of latent time-lag interaction effects. These issues constrain its effectiveness across diverse MTS tasks. To address these challenges, we propose DeMa, a dual-path delay-aware Mamba backbone. DeMa preserves Mamba's linear-complexity advantage while substantially improving its suitability for MTS settings. Specifically, DeMa introduces three key innovations: (i) it decomposes the MTS into intra-series temporal dynamics and inter-series interactions; (ii) it develops a temporal path with a Mamba-SSD module to capture long-range dynamics within each individual series, enabling series-independent, parallel computation; and (iii) it designs a variate path with a Mamba-DALA module that integrates delay-aware linear attention to model cross-variate dependencies. Extensive experiments on five representative tasks, long- and short-term forecasting, data imputation, anomaly detection, and series classification, demonstrate that DeMa achieves state-of-the-art performance while delivering remarkable computational efficiency.
Authors: Wei Zhou, Hong Huang, Ruize Shi, Bang Liu
Abstract: Heterogeneous Graph Neural Networks(HGNNs) have advanced mainly through better encoders, yet their decoding/projection stage still relies on a single shared linear head, assuming it can map rich node embeddings to labels. We call this the Linear Projection Bottleneck: in heterogeneous graphs, contextual diversity and long-tail shifts make a global head miss fine semantics, overfit hub nodes, and underserve tail nodes. While Mixture-of-Experts(MoE) could help, naively applying it clashes with structural imbalance and risks expert collapse. We propose a Heterogeneous-aware Orthogonal Prototype Experts framework named HOPE, a plug-and-play replacement for the standard prediction head. HOPE uses learnable prototype-based routing to assign instances to experts by similarity, letting expert usage follow the natural long-tail distribution, and adds expert orthogonalization to encourage diversity and prevent collapse. Experiments on four real datasets show consistent gains across SOTA HGNN backbones with minimal overhead.
Authors: Moe Shiina, Shunnosuke Ikeda, Yuichi Takano
Abstract: A scoring system is a linear classifier composed of a small number of explanatory variables, each assigned a small integer coefficient. This system is highly interpretable and allows predictions to be made with simple manual calculations without the need for a calculator. Several previous studies have used mixed-integer optimization (MIO) techniques to develop scoring systems for binary classification; however, they have not focused on directly maximizing AUC (i.e., area under the receiver operating characteristic curve), even though AUC is recognized as an essential evaluation metric for scoring systems. Our goal herein is to establish an effective MIO framework for constructing scoring systems that directly maximize the buffered AUC (bAUC) as the tightest concave lower bound on AUC. Our optimization model is formulated as a mixed-integer linear optimization (MILO) problem that maximizes bAUC subject to a group sparsity constraint for limiting the number of questions in the scoring system. Computational experiments using publicly available real-world datasets demonstrate that our MILO method can build scoring systems with superior AUC values compared to the baseline methods based on regularization and stepwise regression. This research contributes to the advancement of MIO techniques for developing highly interpretable classification models.
Authors: Xue Feng, Li Wang, Deanna Needell, Rongjie Lai
Abstract: The Jordan-Kinderlehrer-Otto (JKO) scheme provides a stable variational framework for computing Wasserstein gradient flows, but its practical use is often limited by the high computational cost of repeatedly solving the JKO subproblems. We propose a self-supervised approach for learning a JKO solution operator without requiring numerical solutions of any JKO trajectories. The learned operator maps an input density directly to the minimizer of the corresponding JKO subproblem, and can be iteratively applied to efficiently generate the gradient-flow evolution. A key challenge is that only a number of initial densities are typically available for training. To address this, we introduce a Learn-to-Evolve algorithm that jointly learns the JKO operator and its induced trajectories by alternating between trajectory generation and operator updates. As training progresses, the generated data increasingly approximates true JKO trajectories. Meanwhile, this Learn-to-Evolve strategy serves as a natural form of data augmentation, significantly enhancing the generalization ability of the learned operator. Numerical experiments demonstrate the accuracy, stability, and robustness of the proposed method across various choices of energies and initial conditions.
Authors: Shufei Ge, Shijia Wang, Lloyd Elliott
Abstract: Neural networks have shown state-of-the-art performances in various classification and regression tasks. Rectified linear units (ReLU) are often used as activation functions for the hidden layers in a neural network model. In this article, we establish the connection between the Poisson hyperplane processes (PHP) and two-layer ReLU neural networks. We show that the PHP with a Gaussian prior is an alternative probabilistic representation to a two-layer ReLU neural network. In addition, we show that a two-layer neural network constructed by PHP is scalable to large-scale problems via the decomposition propositions. Finally, we propose an annealed sequential Monte Carlo algorithm for Bayesian inference. Our numerical experiments demonstrate that our proposed method outperforms the classic two-layer ReLU neural network. The implementation of our proposed model is available at https://github.com/ShufeiGe/Pois_Relu.git.
Authors: Jingcheng Hu, Yinmin Zhang, Shijie Shang, Xiaobo Yang, Yue Peng, Zhewei Huang, Hebin Zhou, Xin Wu, Jie Cheng, Fanqi Wan, Xiangwen Kong, Chengyuan Yao, Kaiwen Yan, Ailin Huang, Hongyu Zhou, Qi Han, Zheng Ge, Daxin Jiang, Xiangyu Zhang, Heung-Yeung Shum
Abstract: We introduce Parallel Coordinated Reasoning (PaCoRe), a training-and-inference framework designed to overcome a central limitation of contemporary language models: their inability to scale test-time compute (TTC) far beyond sequential reasoning under a fixed context window. PaCoRe departs from the traditional sequential paradigm by driving TTC through massive parallel exploration coordinated via a message-passing architecture in multiple rounds. Each round launches many parallel reasoning trajectories, compacts their findings into context-bounded messages, and synthesizes these messages to guide the next round and ultimately produce the final answer. Trained end-to-end with large-scale, outcome-based reinforcement learning, the model masters the synthesis abilities required by PaCoRe and scales to multi-million-token effective TTC without exceeding context limits. The approach yields strong improvements across diverse domains, and notably pushes reasoning beyond frontier systems in mathematics: an 8B model reaches 94.5% on HMMT 2025, surpassing GPT-5's 93.2% by scaling effective TTC to roughly two million tokens. We open-source model checkpoints, training data, and the full inference pipeline to accelerate follow-up work.
Authors: S\'ilvia Casacuberta, Moritz Hardt
Abstract: Conditional average treatment effect (CATE) estimation is the de facto gold standard for targeting a treatment to a heterogeneous population. The method estimates treatment effects up to an error $\epsilon > 0$ in each of $M$ different strata of the population, targeting individuals in decreasing order of estimated treatment effect until the budget runs out. In general, this method requires $O(M/\epsilon^2)$ samples. This is best possible if the goal is to estimate all treatment effects up to an $\epsilon$ error. In this work, we show how to achieve the same total treatment effect as CATE with only $O(M/\epsilon)$ samples for natural distributions of treatment effects. The key insight is that coarse estimates suffice for near-optimal treatment allocations. In addition, we show that budget flexibility can further reduce the sample complexity of allocation. Finally, we evaluate our algorithm on various real-world RCT datasets. In all cases, it finds nearly optimal treatment allocations with surprisingly few samples. Our work highlights the fundamental distinction between treatment effect estimation and treatment allocation: the latter requires far fewer samples.
Authors: Zijun Min, Bingshuai Liu, Ante Wang, Long Zhang, Anxiang Zeng, Haibo Zhang, Jinsong Su
Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) offers a promising framework for optimizing large language models in reasoning tasks. However, existing RLVR algorithms focus on different granularities, and each has complementary strengths and limitations. Group Relative Policy Optimization (GRPO) updates the policy with token-level importance ratios, which preserves fine-grained credit assignment but often suffers from high variance and instability. In contrast, Group Sequence Policy Optimization (GSPO) applies single sequence-level importance ratios across all tokens in a response that better matches sequence-level rewards, but sacrifices token-wise credit assignment. In this paper, we propose Dynamic Hybrid Policy Optimization (DHPO) to bridge GRPO and GSPO within a single clipped surrogate objective. DHPO combines token-level and sequence-level importance ratios using weighting mechanisms. We explore two variants of the mixing mechanism, including an averaged mixing and an entropy-guided mixing. To further stabilize training, we employ a branch-specific clipping strategy that constrains token-level and sequence-level ratios within separate trust regions before mixing, preventing outliers in either branch from dominating the update. Across seven challenging mathematical reasoning benchmarks, experiments on both dense and MoE models from the Qwen3 series show that DHPO consistently outperforms GRPO and GSPO. We will release our code upon acceptance of this paper.
Authors: Yiming Zhou, Mingyue Cheng, Hao Wang, Enhong Chen
Abstract: Time series are highly valuable and rarely shareable across nodes, making federated learning a promising paradigm to leverage distributed temporal data. However, different sampling standards lead to diverse time granularities and variable sets across nodes, hindering classical federated learning. We propose PiXTime, a novel time series forecasting model designed for federated learning that enables effective prediction across nodes with multi-granularity and heterogeneous variable sets. PiXTime employs a personalized Patch Embedding to map node-specific granularity time series into token sequences of a unified dimension for processing by a subsequent shared model, and uses a global VE Table to align variable category semantics across nodes, thereby enhancing cross-node transferability. With a transformer-based shared model, PiXTime captures representations of auxiliary series with arbitrary numbers of variables and uses cross-attention to enhance the prediction of the target series. Experiments show PiXTime achieves state-of-the-art performance in federated settings and demonstrates superior performance on eight widely used real-world traditional benchmarks.
Authors: ShaoZhen Liu, Xinting Huang, Houwen Peng, Xin Chen, Xinyang Song, Qi Li, Zhenan Sun
Abstract: In recent years, large language models (LLMs) have demonstrated significant potential in complex reasoning tasks like mathematical problem-solving. However, existing research predominantly relies on reinforcement learning (RL) frameworks while overlooking supervised fine-tuning (SFT) methods. This paper proposes a new two-stage training framework that enhances models' self-correction capabilities through self-generated long chain-of-thought (CoT) data. During the first stage, a multi-turn dialogue strategy guides the model to generate CoT data incorporating verification, backtracking, subgoal decomposition, and backward reasoning, with predefined rules filtering high-quality samples for supervised fine-tuning. The second stage employs a difficulty-aware rejection sampling mechanism to dynamically optimize data distribution, strengthening the model's ability to handle complex problems. The approach generates reasoning chains extended over 4 times longer while maintaining strong scalability, proving that SFT effectively activates models' intrinsic reasoning capabilities and provides a resource-efficient pathway for complex task optimization. Experimental results demonstrate performance improvements on mathematical benchmarks including GSM8K and MATH500, with the fine-tuned model achieving a substantial improvement on competition-level problems like AIME24. Code will be open-sourced.
Authors: Zhi Wang, Zhongbin Wu, Yanni Li, Bing Liu, Guangxi Li, Yuping Wang
Abstract: Existing research on continual learning (CL) of a sequence of tasks focuses mainly on dealing with catastrophic forgetting (CF) to balance the learning plasticity of new tasks and the memory stability of old tasks. However, an ideal CL agent should not only be able to overcome CF, but also encourage positive forward and backward knowledge transfer (KT), i.e., using the learned knowledge from previous tasks for the new task learning (namely FKT), and improving the previous tasks' performance with the knowledge of the new task (namely BKT). To this end, this paper first models CL as an optimization problem in which each sequential learning task aims to achieve its optimal performance under the constraint that both FKT and BKT should be positive. It then proposes a novel Enhanced Task Continual Learning (ETCL) method, which achieves forgetting-free and positive KT. Furthermore, the bounds that can lead to negative FKT and BKT are estimated theoretically. Based on the bounds, a new strategy for online task similarity detection is also proposed to facilitate positive KT. To overcome CF, ETCL learns a set of task-specific binary masks to isolate a sparse sub-network for each task while preserving the performance of a dense network for the task. At the beginning of a new task learning, ETCL tries to align the new task's gradient with that of the sub-network of the previous most similar task to ensure positive FKT. By using a new bi-objective optimization strategy and an orthogonal gradient projection method, ETCL updates only the weights of previous similar tasks at the classification layer to achieve positive BKT. Extensive evaluations demonstrate that the proposed ETCL markedly outperforms strong baselines on dissimilar, similar, and mixed task sequences.
Authors: Xinyue Wang, Stephen Wang, Biwei Huang
Abstract: We reveal that transformers trained in an autoregressive manner naturally encode time-delayed causal structures in their learned representations. When predicting future values in multivariate time series, the gradient sensitivities of transformer outputs with respect to past inputs directly recover the underlying causal graph, without any explicit causal objectives or structural constraints. We prove this connection theoretically under standard identifiability conditions and develop a practical extraction method using aggregated gradient attributions. On challenging cases such as nonlinear dynamics, long-term dependencies, and non-stationary systems, this approach greatly surpasses the performance of state-of-the-art discovery algorithms, especially as data heterogeneity increases, exhibiting scaling potential where causal accuracy improves with data volume and heterogeneity, a property traditional methods lack. This unifying view lays the groundwork for a future paradigm where causal discovery operates through the lens of foundation models, and foundation models gain interpretability and enhancement through the lens of causality.
Authors: Miguel Matey-Sanz, Joaqu\'in Torres-Sospedra, Joaqu\'in Huerta, Sergio Trilles
Abstract: Wi-Fi fingerprinting remains one of the most practical solutions for indoor positioning, however, its performance is often limited by the size and heterogeneity of fingerprint datasets, strong Received Signal Strength Indicator variability, and the ambiguity introduced in large and multi-floor environments. These factors significantly degrade localisation accuracy, particularly when global models are applied without considering structural constraints. This paper introduces a clustering-based method that structures the fingerprint dataset prior to localisation. Fingerprints are grouped using either spatial or radio features, and clustering can be applied at the building or floor level. In the localisation phase, a clustering estimation procedure based on the strongest access points assigns unseen fingerprints to the most relevant cluster. Localisation is then performed only within the selected clusters, allowing learning models to operate on reduced and more coherent subsets of data. The effectiveness of the method is evaluated on three public datasets and several machine learning models. Results show a consistent reduction in localisation errors, particularly under building-level strategies, but at the cost of reducing the floor detection accuracy. These results demonstrate that explicitly structuring datasets through clustering is an effective and flexible approach for scalable indoor positioning.
Authors: George Ma, Zhongyuan Liang, Irene Y. Chen, Somayeh Sojoudi
Abstract: We investigate whether sparse autoencoders (SAEs) identify genuine reasoning features in large language models (LLMs). Starting from features selected using standard contrastive activation methods, we introduce a falsification-oriented framework that combines causal token injection experiments and LLM-guided falsification to test whether feature activation reflects reasoning processes or superficial linguistic correlates. Across 20 configurations spanning multiple model families, layers, and reasoning datasets, we find that identified reasoning features are highly sensitive to token-level interventions. Injecting a small number of feature-associated tokens into non-reasoning text is sufficient to elicit strong activation for 59% to 94% of features, indicating reliance on lexical artifacts. For the remaining features that are not explained by simple token triggers, LLM-guided falsification consistently produces non-reasoning inputs that activate the feature and reasoning inputs that do not, with no analyzed feature satisfying our criteria for genuine reasoning behavior. Steering these features yields minimal changes or slight degradations in benchmark performance. Together, these results suggest that SAE features identified by contrastive approaches primarily capture linguistic correlates of reasoning rather than the underlying reasoning computations themselves.
Authors: Yeonsang Shin, Insoo Kim, Bongkeun Kim, Keonwoo Bae, Bohyung Han
Abstract: Transformer-based autoregressive models excel in data generation but are inherently constrained by their reliance on discretized tokens, which limits their ability to represent continuous values with high precision. We analyze the scalability limitations of existing discretization-based approaches for generating hybrid discrete-continuous sequences, particularly in high-precision domains such as semiconductor circuit designs, where precision loss can lead to functional failure. To address the challenge, we propose AGDC, a novel unified framework that jointly models discrete and continuous values for variable-length sequences. AGDC employs a hybrid approach that combines categorical prediction for discrete values with diffusion-based modeling for continuous values, incorporating two key technical components: an end-of-sequence (EOS) logit adjustment mechanism that uses an MLP to dynamically adjust EOS token logits based on sequence context, and a length regularization term integrated into the loss function. Additionally, we present ContLayNet, a large-scale benchmark comprising 334K high-precision semiconductor layout samples with specialized evaluation metrics that capture functional correctness where precision errors significantly impact performance. Experiments on semiconductor layouts (ContLayNet), graphic layouts, and SVGs demonstrate AGDC's superior performance in generating high-fidelity hybrid vector representations compared to discretization-based and fixed-schema baselines, achieving scalable high-precision generation across diverse domains.
Authors: Hongyaoxing Gul, Lijuan Hu, Shuzi Niu, Fangfang Liu
Abstract: Traditional post-training quantization (PTQ) is considered an effective approach to reduce model size and accelerate inference of large-scale language models (LLMs). However, existing low-rank PTQ methods require costly fine-tuning to determine a compromise rank for diverse data and layers in large models, failing to exploit their full potential. Additionally, the current SVD-based low-rank approximation compounds the computational overhead. In this work, we thoroughly analyze the varying effectiveness of low-rank approximation across different layers in representative models. Accordingly, we introduce \underline{F}lexible \underline{L}ow-\underline{R}ank \underline{Q}uantization (FLRQ), a novel solution designed to quickly identify the accuracy-optimal ranks and aggregate them to achieve minimal storage combinations. FLRQ comprises two powerful components, Rank1-Sketch-based Flexible Rank Selection (R1-FLR) and Best Low-rank Approximation under Clipping (BLC). R1-FLR applies the R1-Sketch with Gaussian projection for the fast low-rank approximation, enabling outlier-aware rank extraction for each layer. Meanwhile, BLC aims at minimizing the low-rank quantization error under the scaling and clipping strategy through an iterative method. FLRQ demonstrates strong effectiveness and robustness in comprehensive experiments, achieving state-of-the-art performance in both quantization quality and algorithm efficiency.
Authors: Yongyi Yang, Jianyang Gao
Abstract: Hyper-Connections (HC) generalizes residual connections by introducing dynamic residual matrices that mix information across multiple residual streams, accelerating convergence in deep neural networks. However, unconstrained residual matrices can compromise training stability. To address this, DeepSeek's Manifold-Constrained Hyper-Connections (mHC) approximately projects these matrices onto the Birkhoff polytope via iterative Sinkhorn--Knopp (SK) normalization. We identify two limitations of this approach: (i) finite SK iterations do not guarantee exact doubly stochasticity, leaving an approximation gap that can accumulate through network depth and undermine stability; (ii) efficient SK implementation requires highly specialized CUDA kernels, raising engineering barriers and reducing portability. Motivated by the Birkhoff--von Neumann theorem, we propose mHC-lite, a simple reparameterization that explicitly constructs doubly stochastic matrices as convex combinations of permutation matrices. This approach guarantees exact doubly stochasticity by construction and can be implemented using only native matrix operations. Extensive experiments demonstrate that mHC-lite matches or exceeds mHC in performance while achieving higher training throughput with a naive implementation and eliminating the residual instabilities observed in both HC and mHC. The code is publicly available at https://github.com/FFTYYY/mhc-lite.
Authors: Turkan Simge Ispak, Salih Tileylioglu, Erdem Akagunduz
Abstract: Accurate P-wave detection is critical for earthquake early warning, yet strong-motion records pose challenges due to high noise levels, limited labeled data, and complex waveform characteristics. This study reframes P-wave arrival detection as a self-supervised anomaly detection task to evaluate how architectural variations regulate the trade-off between reconstruction fidelity and anomaly discrimination. Through a comprehensive grid search of 492 Variational Autoencoder configurations, we show that while skip connections minimize reconstruction error (Mean Absolute Error approximately 0.0012), they induce "overgeneralization", allowing the model to reconstruct noise and masking the detection signal. In contrast, attention mechanisms prioritize global context over local detail and yield the highest detection performance with an area-under-the-curve of 0.875. The attention-based Variational Autoencoder achieves an area-under-the-curve of 0.91 in the 0 to 40-kilometer near-source range, demonstrating high suitability for immediate early warning applications. These findings establish that architectural constraints favoring global context over pixel-perfect reconstruction are essential for robust, self-supervised P-wave detection.
Authors: Yifan Zhang, Wei Bi, Kechi Zhang, Dongming Jin, Jie Fu, Zhi Jin
Abstract: Algorithm extraction aims to synthesize executable programs directly from models trained on specific algorithmic tasks, enabling de novo algorithm discovery without relying on human-written code. However, extending this paradigm to Transformer is hindered by superposition, where entangled features encoded in overlapping directions obstruct the extraction of symbolic expressions. In this work, we propose the Discrete Transformer, an architecture explicitly engineered to bridge the gap between continuous representations and discrete symbolic logic. By enforcing a strict functional disentanglement, which constrains Numerical Attention to information routing and Numerical MLP to element-wise arithmetic, and employing temperature-annealed sampling, our method effectively facilitates the extraction of human-readable programs. Empirically, the Discrete Transformer not only achieves performance comparable to RNN-based baselines but crucially extends interpretability to continuous variable domains. Moreover, our analysis of the annealing process shows that the efficient discrete search undergoes a clear phase transition from exploration to exploitation. We further demonstrate that our method enables fine-grained control over synthesized programs by imposing inductive biases. Collectively, these findings establish the Discrete Transformer as a robust framework for demonstration-free algorithm discovery, offering a rigorous pathway toward Transformer interpretability.
Authors: Manel Gil-Sorribes, J\'ulia Vilalta-Mor, Isaac Filella-Merc\`e, Robert Soliva, \'Alvaro Ciudad, V\'ictor Guallar, Alexis Molina
Abstract: Accurate drug-target interaction (DTI) prediction is essential for computational drug discovery, yet existing models often rely on single-modality predefined molecular descriptors or sequence-based embeddings with limited representativeness. We propose Tensor-DTI, a contrastive learning framework that integrates multimodal embeddings from molecular graphs, protein language models, and binding-site predictions to improve interaction modeling. Tensor-DTI employs a siamese dual-encoder architecture, enabling it to capture both chemical and structural interaction features while distinguishing interacting from non-interacting pairs. Evaluations on multiple DTI benchmarks demonstrate that Tensor-DTI outperforms existing sequence-based and graph-based models. We also conduct large-scale inference experiments on CDK2 across billion-scale chemical libraries, where Tensor-DTI produces chemically plausible hit distributions even when CDK2 is withheld from training. In enrichment studies against Glide docking and Boltz-2 co-folder, Tensor-DTI remains competitive on CDK2 and improves the screening budget required to recover moderate fractions of high-affinity ligands on out-of-family targets under strict family-holdout splits. Additionally, we explore its applicability to protein-RNA and peptide-protein interactions. Our findings highlight the benefits of integrating multimodal information with contrastive objectives to enhance interaction-prediction accuracy and to provide more interpretable and reliability-aware models for virtual screening.
Authors: Mohamed Amine Hallam, Kuo-Kun Tseng
Abstract: Transformers require positional encodings to represent sequence order, yet most prior work focuses on designing new positional encodings rather than examining how positional information is fused with token embeddings. In this paper, we study whether the fusion mechanism itself affects performance, particularly in long-sequence settings. We conduct a controlled empirical study comparing three canonical fusion strategies--element-wise addition, concatenation with projection, and scalar gated fusion--under identical Transformer architectures, data splits, and random seeds. Experiments on three text classification datasets spanning short (AG News), medium (IMDB), and long (ArXiv) sequences show that fusion choice has negligible impact on short texts but produces consistent gains on long documents. To verify that these gains are structural rather than stochastic, we perform paired-seed analysis and cross-dataset comparison across sequence-length regimes. Additional experiments on the ArXiv dataset indicate that the benefit of learnable fusion generalizes across multiple positional encoding families. Finally, we explore a lightweight convolutional gating mechanism that introduces local inductive bias at the fusion level, evaluated on long documents only. Our results indicate that positional-encoding fusion is a non-trivial design choice for long-sequence Transformers and should be treated as an explicit modeling decision rather than a fixed default.
Authors: Enrique Feito-Casares, Francisco M. Melgarejo-Meseguer, Jos\'e-Luis Rojo-\'Alvarez
Abstract: Motivated by the growing interest in representation learning approaches that uncover the latent structure of high-dimensional data, this work proposes new algorithms for reconstruction-based manifold learning within Reproducing-Kernel Hilbert Spaces (RKHS). Each observation is first reconstructed as a linear combination of the other samples in the RKHS, by optimizing a vector form of the Representer Theorem for their autorepresentation property. A separable operator-valued kernel extends the formulation to vector-valued data while retaining the simplicity of a single scalar similarity function. A subsequent kernel-alignment task projects the data into a lower-dimensional latent space whose Gram matrix aims to match the high-dimensional reconstruction kernel, thus transferring the auto-reconstruction geometry of the RKHS to the embedding. Therefore, the proposed algorithms represent an extended approach to the autorepresentation property, exhibited by many natural data, by using and adapting well-known results of Kernel Learning Theory. Numerical experiments on both simulated (concentric circles and swiss-roll) and real (cancer molecular activity and IoT network intrusions) datasets provide empirical evidence of the practical effectiveness of the proposed approach.
Authors: Zhanpei Huang, Taochen chen, Fangqing Gu, Yiqun Zhang
Abstract: Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder characterized by deficits in social communication and behavioral patterns. Eye movement data offers a non-invasive diagnostic tool for ASD detection, as it is inherently discrete and exhibits short-term temporal dependencies, reflecting localized gaze focus between fixation points. These characteristics enable the data to provide deeper insights into subtle behavioral markers, distinguishing ASD-related patterns from typical development. Eye movement signals mainly contain short-term and localized dependencies. However, despite the widespread application of stacked attention layers in Transformer-based models for capturing long-range dependencies, our experimental results indicate that this approach yields only limited benefits when applied to eye movement data. This may be because discrete fixation points and short-term dependencies in gaze focus reduce the utility of global attention mechanisms, making them less efficient than architectures focusing on local temporal patterns. To efficiently capture subtle and complex eye movement patterns, distinguishing ASD from typically developing (TD) individuals, a discrete short-term sequential (DSTS) modeling framework is designed with Class-aware Representation and Imbalance-aware Mechanisms. Through extensive experiments on several eye movement datasets, DSTS outperforms both traditional machine learning techniques and more sophisticated deep learning models.
Authors: Md Sultanul Islam Ovi, Muhsina Tarannum Munfa, Miftahul Alam Adib, Syed Sabbir Hasan
Abstract: Accurate classification of sleep disorders, particularly insomnia and sleep apnea, is important for reducing long term health risks and improving patient quality of life. However, clinical sleep studies are resource intensive and are difficult to scale for population level screening. This paper presents a Dual Pipeline Machine Learning Framework for multi class sleep disorder screening using the Sleep Health and Lifestyle dataset. The framework consists of two parallel processing streams: a statistical pipeline that targets linear separability using Mutual Information and Linear Discriminant Analysis, and a wrapper based pipeline that applies Boruta feature selection with an autoencoder for non linear representation learning. To address class imbalance, we use the hybrid SMOTETomek resampling strategy. In experiments, Extra Trees and K Nearest Neighbors achieved an accuracy of 98.67%, outperforming recent baselines on the same dataset. Statistical testing using the Wilcoxon Signed Rank Test indicates that the improvement over baseline configurations is significant, and inference latency remains below 400 milliseconds. These results suggest that the proposed dual pipeline design supports accurate and efficient automated screening for non invasive sleep disorder risk stratification.
Authors: Eric Weine, Peter Carbonetto, Rafael A. Irizarry, Matthew Stephens
Abstract: Poisson non-negative matrix factorization (NMF) is a widely used method to find interpretable "parts-based" decompositions of count data. While many variants of Poisson NMF exist, existing methods assume that the "parts" in the decomposition combine additively. This assumption may be natural in some settings, but not in others. Here we introduce Poisson NMF with the shifted-log link function to relax this assumption. The shifted-log link function has a single tuning parameter, and as this parameter varies the model changes from assuming that parts combine additively (i.e., standard Poisson NMF) to assuming that parts combine more multiplicatively. We provide an algorithm to fit this model by maximum likelihood, and also an approximation that substantially reduces computation time for large, sparse datasets (computations scale with the number of non-zero entries in the data matrix). We illustrate these new methods on a variety of real datasets. Our examples show how the choice of link function in Poisson NMF can substantively impact the results, and how in some settings the use of a shifted-log link function may improve interpretability compared with the standard, additive link.
Authors: Huilin Deng, Hongchen Luo, Yue Zhu, Long Li, Zhuoyue Chen, Xinghao Zhao, Ming Li, Jihai Zhang, Mengchang Wang, Yang Cao, Yu Kang
Abstract: Recent advances in Reinforcement Learning with Verifiable Rewards (RLVR) for Large Language Model (LLM) reasoning have been hindered by a persistent challenge: exploration collapse. The semantic homogeneity of random rollouts often traps models in narrow, over-optimized behaviors. While existing methods leverage policy entropy to encourage exploration, they face inherent limitations. Global entropy regularization is susceptible to reward hacking, which can induce meaningless verbosity, whereas local token-selective updates struggle with the strong inductive bias of pre-trained models. To address this, we propose Latent Policy Optimization via Iterative Information Bottleneck (IIB-LPO), a novel approach that shifts exploration from statistical perturbation of token distributions to topological branching of reasoning trajectories. IIB-LPO triggers latent branching at high-entropy states to diversify reasoning paths and employs the Information Bottleneck principle both as a trajectory filter and a self-reward mechanism, ensuring concise and informative exploration. Empirical results across four mathematical reasoning benchmarks demonstrate that IIB-LPO achieves state-of-the-art performance, surpassing prior methods by margins of up to 5.3% in accuracy and 7.4% in diversity metrics.
Authors: Doyoung Kim, Donghee Lee, Hye-Sung Lee, Jiheon Lee, Jaeok Yi
Abstract: In many problems in physics and engineering, one encounters complicated differential equations with strongly scale-dependent terms for which exact analytical or numerical solutions are not available. A common strategy is to divide the domain into several regions (patches) and simplify the equation in each region. When approximate analytic solutions can be obtained in each patch, they are then matched at the interfaces to construct a global solution. However, this patching procedure can fail to reproduce the correct solution, since the approximate forms may break down near the matching boundaries. In this work, we propose a learning framework in which the integration constants of asymptotic analytic solutions are promoted to scale-dependent functions. By constraining these coefficient functions with the original differential equation over the domain, the network learns a globally valid solution that smoothly interpolates between asymptotic regimes, eliminating the need for arbitrary boundary matching. We demonstrate the effectiveness of this framework in representative problems from chemical kinetics and cosmology, where it accurately reproduces global solutions and outperforms conventional matching procedures.
Authors: Ayoub Ajarra, Debabrota Basu
Abstract: As machine learning models become increasingly embedded in societal infrastructure, auditing them for bias is of growing importance. However, in real-world deployments, auditing is complicated by the fact that model owners may adaptively update their models in response to changing environments, such as financial markets. These updates can alter the underlying model class while preserving certain properties of interest, raising fundamental questions about what can be reliably audited under such shifts. In this work, we study group fairness auditing under arbitrary updates. We consider general shifts that modify the pre-audit model class while maintaining invariance of the audited property. Our goals are two-fold: (i) to characterize the information complexity of allowable updates, by identifying which strategic changes preserve the property under audit; and (ii) to efficiently estimate auditing properties, such as group fairness, using a minimal number of labeled samples. We propose a generic framework for PAC auditing based on an Empirical Property Optimization (EPO) oracle. For statistical parity, we establish distribution-free auditing bounds characterized by the SP dimension, a novel combinatorial measure that captures the complexity of admissible strategic updates. Finally, we demonstrate that our framework naturally extends to other auditing objectives, including prediction error and robust risk.
Authors: Pattarawat Chormai, Ali Hashemi, Klaus-Robert M\"uller, Gr\'egoire Montavon
Abstract: Knowledge distillation involves transferring the predictive capabilities of large, high-performing AI models (teachers) to smaller models (students) that can operate in environments with limited computing power. In this paper, we address the scenario in which only a few classes and their associated intermediate concepts are relevant to distill. This scenario is common in practice, yet few existing distillation methods explicitly focus on the relevant subtask. To address this gap, we introduce 'SubDistill', a new distillation algorithm with improved numerical properties that only distills the relevant components of the teacher model at each layer. Experiments on CIFAR-100 and ImageNet with Convolutional and Transformer models demonstrate that SubDistill outperforms existing layer-wise distillation techniques on a representative set of subtasks. Our benchmark evaluations are complemented by Explainable AI analyses showing that our distilled student models more closely match the decision structure of the original teacher model.
Authors: Sidney Shapiro, Burhanuddin Panvelwala
Abstract: Reproducibility remains a persistent challenge in forecasting research and practice, particularly in business and financial analytics where forecasts inform high-stakes decisions. Traditional forecasting methods, while theoretically interpretable, often require extensive manual tuning and are difficult to replicate in proprietary environments. Machine learning approaches offer predictive flexibility but introduce challenges related to interpretability, stochastic training procedures, and cross-environment reproducibility. This paper examines Prophet, an open-source forecasting framework developed by Meta, as a reproducibility-enabling solution that balances interpretability, standardized workflows, and accessibility. Rather than proposing a new algorithm, this study evaluates how Prophet's additive structure, open-source implementation, and standardized workflow contribute to transparent and replicable forecasting practice. Using publicly available financial and retail datasets, we compare Prophet's performance and interpretability with multiple ARIMA specifications (auto-selected, manually specified, and seasonal variants) and Random Forest under a controlled and fully documented experimental design. This multi-model comparison provides a robust assessment of Prophet's relative performance and reproducibility advantages. Through concrete Python examples, we demonstrate how Prophet facilitates efficient forecasting workflows and integration with analytical pipelines. The study positions Prophet within the broader context of reproducible research. It highlights Prophet's role as a methodological building block that supports verification, auditability, and methodological rigor. This work provides researchers and practitioners with a practical reference framework for reproducible forecasting in Python-based research workflows.
Authors: Juaren Steiger, Bin Li
Abstract: The constrained combinatorial multi-armed bandit model has been widely employed to solve problems in wireless networking and related areas, including the problem of wireless scheduling for throughput optimization under unknown channel conditions. Most work in this area uses an algorithm design strategy that combines a bandit learning algorithm with the virtual queue technique to track the throughput constraint violation. These algorithms seek to minimize the virtual queue length in their algorithm design. However, in networks where channel conditions change abruptly, the resulting constraints may become infeasible, leading to unbounded growth in virtual queue lengths. In this paper, we make the key observation that the dynamics of the head-of-line age, i.e. the age of the oldest packet in the virtual queue, make it more robust when used in algorithm design compared to the virtual queue length. We therefore design a learning-based scheduling policy that uses the head-of-line age in place of the virtual queue length. We show that our policy matches state-of-the-art performance under i.i.d. network conditions. Crucially, we also show that the system remains stable even under abrupt changes in channel conditions and can rapidly recover from periods of constraint infeasibility.
Authors: Sahibzada Saadoon Hammad, Joaqu\'in Huerta Guijarro, Francisco Ramos, Michael Gould Carlson, Sergio Trilles Oliver
Abstract: The rapid deployment of Internet of Things (IoT) devices has led to large-scale sensor networks that monitor environmental and urban phenomena in real time. Communities of Interest (CoIs) provide a promising paradigm for organising heterogeneous IoT sensor networks by grouping devices with similar operational and environmental characteristics. This work presents an anomaly detection framework based on the CoI paradigm by grouping sensors into communities using a fused similarity matrix that incorporates temporal correlations via Spearman coefficients, spatial proximity using Gaussian distance decay, and elevation similarities. For each community, representative stations based on the best silhouette are selected and three autoencoder architectures (BiLSTM, LSTM, and MLP) are trained using Bayesian hyperparameter optimization with expanding window cross-validation and tested on stations from the same cluster and the best representative stations of other clusters. The models are trained on normal temperature patterns of the data and anomalies are detected through reconstruction error analysis. Experimental results show a robust within-community performance across the evaluated configurations, while variations across communities are observed. Overall, the results support the applicability of community-based model sharing in reducing computational overhead and to analyse model generalisability across IoT sensor networks.
Authors: {\TH}\'or Sverrisson, Steinn Gu{\dh}mundsson
Abstract: Automated seizure detection from electroencephalography (EEG) remains difficult due to the large variability of seizure dynamics across patients, recording conditions, and clinical settings. We introduce LookAroundNet, a transformer-based seizure detector that uses a wider temporal window of EEG data to model seizure activity. The seizure detector incorporates EEG signals before and after the segment of interest, reflecting how clinicians use surrounding context when interpreting EEG recordings. We evaluate the proposed method on multiple EEG datasets spanning diverse clinical environments, patient populations, and recording modalities, including routine clinical EEG and long-term ambulatory recordings, in order to study performance across varying data distributions. The evaluation includes publicly available datasets as well as a large proprietary collection of home EEG recordings, providing complementary views of controlled clinical data and unconstrained home-monitoring conditions. Our results show that LookAroundNet achieves strong performance across datasets, generalizes well to previously unseen recording conditions, and operates with computational costs compatible with real-world clinical deployment. The results indicate that extended temporal context, increased training data diversity, and model ensembling are key factors for improving performance. This work contributes to moving automatic seizure detection models toward clinically viable solutions.
Authors: Zequan Liang, Ruoyu Zhang, Wei Shao, krishna Karthik, Ehsan Kourkchi, Setareh Rafatirad, Houman Homayoun
Abstract: Blood oxygen saturation (SpO2) is a vital marker for healthcare monitoring. Traditional SpO2 estimation methods often rely on complex clinical calibration, making them unsuitable for low-power, wearable applications. In this paper, we propose a transfer learning-based framework for the rapid adaptation of SpO2 estimation to energy-efficient wearable devices using low-sampling-rate (25Hz) dual-channel photoplethysmography (PPG). We first pretrain a bidirectional Long Short-Term Memory (BiLSTM) model with self-attention on a public clinical dataset, then fine-tune it using data collected from our wearable We-Be band and an FDA-approved reference pulse oximeter. Experimental results show that our approach achieves a mean absolute error (MAE) of 2.967% on the public dataset and 2.624% on the private dataset, significantly outperforming traditional calibration and non-transferred machine learning baselines. Moreover, using 25Hz PPG reduces power consumption by 40% compared to 100Hz, excluding baseline draw. Our method also attains an MAE of 3.284% in instantaneous SpO2 prediction, effectively capturing rapid fluctuations. These results demonstrate the rapid adaptation of accurate, low-power SpO2 monitoring on wearable devices without the need for clinical calibration.
Authors: Zequan Liang, Ruoyu Zhang, Wei Shao, Mahdi Pirayesh Shirazi Nejad, Ehsan Kourkchi, Setareh Rafatirad, Houman Homayoun
Abstract: Accurate and generalizable blood pressure (BP) estimation is vital for the early detection and management of cardiovascular diseases. In this study, we enforce subject-level data splitting on a public multi-wavelength photoplethysmography (PPG) dataset and propose a generalizable BP estimation framework based on curriculum-adversarial learning. Our approach combines curriculum learning, which transitions from hypertension classification to BP regression, with domain-adversarial training that confuses subject identity to encourage the learning of subject-invariant features. Experiments show that multi-channel fusion consistently outperforms single-channel models. On the four-wavelength PPG dataset, our method achieves strong performance under strict subject-level splitting, with mean absolute errors (MAE) of 14.2mmHg for systolic blood pressure (SBP) and 6.4mmHg for diastolic blood pressure (DBP). Additionally, ablation studies validate the effectiveness of both the curriculum and adversarial components. These results highlight the potential of leveraging complementary information in multi-wavelength PPG and curriculum-adversarial strategies for accurate and robust BP estimation.
Authors: Armin Gerami, Kazem Faghih, Ramani Duraiswami
Abstract: Retrieval Augmented Generation (RAG) enhances Large Language Models (LLMs) by connecting them to external knowledge, improving accuracy and reducing outdated information. However, this introduces challenges such as factual inconsistencies, source conflicts, bias propagation, and security vulnerabilities, which undermine the trustworthiness of RAG systems. A key gap in current RAG evaluation is the lack of a metric to quantify the contribution of individual retrieved documents to the final output. To address this, we introduce the Influence Score (IS), a novel metric based on Partial Information Decomposition that measures the impact of each retrieved document on the generated response. We validate IS through two experiments. First, a poison attack simulation across three datasets demonstrates that IS correctly identifies the malicious document as the most influential in $86\%$ of cases. Second, an ablation study shows that a response generated using only the top-ranked documents by IS is consistently judged more similar to the original response than one generated from the remaining documents. These results confirm the efficacy of IS in isolating and quantifying document influence, offering a valuable tool for improving the transparency and reliability of RAG systems.
Authors: Muhammad Mufti, Omar Hammad, Mahfuzur Rahman
Abstract: Online consumer reviews play a crucial role in guiding purchase decisions by offering insights into product quality, usability, and performance. However, the increasing volume of user-generated reviews has led to information overload, making it difficult for consumers to identify content that aligns with their specific preferences. Existing review ranking systems typically rely on metrics such as helpfulness votes, star ratings, and recency, but these fail to capture individual user interests and often treat textual sentiment and rating signals separately. This research addresses these limitations by proposing a personalized framework that integrates review ranking and abstractive summarization to enhance decision-making efficiency. The proposed system begins by modeling each user's sentiment through a hybrid analysis of star ratings and review content. Simultaneously, user preferences were derived from historical reviews using sentence embeddings and clustering, forming semantic profiles aligned with thematic and sentiment dimensions. A relevance scoring algorithm matched these profiles with unseen reviews based on sentiment and aspect similarity. Top-matched reviews were then summarized to reflect individual interests. A user study with 70 participants demonstrated that the personalized approach improved satisfaction, perceived relevance, and decision-making confidence, while reducing time spent reading. The results highlight the method's effectiveness in alleviating information overload and delivering content tailored to user-specific preferences, emphasizing its value in enhancing user experience in review-rich decision-making environments.
Authors: Muzakkiruddin Ahmed Mohammed, John R. Talburt, Leon Claasssens, Adriaan Marais
Abstract: Industrial part specification extraction from unstructured text remains a persistent challenge in manufacturing, procurement, and maintenance, where manual processing is both time-consuming and error-prone. This paper introduces a retrieval-augmented multi-LLM ensemble framework that orchestrates nine state-of-the-art Large Language Models (LLMs) within a structured three-phase pipeline. RAGsemble addresses key limitations of single-model systems by combining the complementary strengths of model families including Gemini (2.0, 2.5, 1.5), OpenAI (GPT-4o, o4-mini), Mistral Large, and Gemma (1B, 4B, 3n-e4b), while grounding outputs in factual data using FAISS-based semantic retrieval. The system architecture consists of three stages: (1) parallel extraction by diverse LLMs, (2) targeted research augmentation leveraging high-performing models, and (3) intelligent synthesis with conflict resolution and confidence-aware scoring. RAG integration provides real-time access to structured part databases, enabling the system to validate, refine, and enrich outputs through similarity-based reference retrieval. Experimental results using real industrial datasets demonstrate significant gains in extraction accuracy, technical completeness, and structured output quality compared to leading single-LLM baselines. Key contributions include a scalable ensemble architecture for industrial domains, seamless RAG integration throughout the pipeline, comprehensive quality assessment mechanisms, and a production-ready solution suitable for deployment in knowledge-intensive manufacturing environments.
Authors: Geonwoo Bang, Dongho Kim, Moohong Min
Abstract: Evaluating complex texts across domains requires converting user defined criteria into quantitative, explainable indicators, which is a persistent challenge in search and recommendation systems. Single prompt LLM evaluations suffer from complexity and latency issues, while criterion specific decomposition approaches rely on naive averaging or opaque black-box aggregation methods. We present an interpretable aggregation framework combining LLM scoring with the Analytic Hierarchy Process. Our method generates criterion specific scores via LLM as judge, measures discriminative power using Jensen Shannon distance, and derives statistically grounded weights through AHP pairwise comparison matrices. Experiments on Amazon review quality assessment and depression related text scoring demonstrate that our approach achieves high explainability and operational efficiency while maintaining comparable predictive power, making it suitable for real time latency sensitive web services.
Authors: Yoav Evron, Michal Bar-Asher Siegal, Michael Fire
Abstract: The recent Artificial Intelligence (AI) revolution has opened transformative possibilities for the humanities, particularly in unlocking the visual content embedded in historical manuscripts. While digital archives now offer unprecedented access to these materials, the ability to systematically study illustrations at a large scale remains challenging. Our study presents a fast and scalable AI approach for detecting, extracting, and describing illustrations in digitized manuscripts. Focusing on collections like the Vatican Library, our system enables efficient visual analysis across millions of pages. Our pipeline consists of three stages: (1) a fine-tuned image classification model filters out text-only pages; (2) an efficient object detection model identifies and crops illustrations; and (3) a multimodal image captioning model generates concise, human-readable descriptions. These are stored in a searchable database, allowing scholars to retrieve relevant visual materials through keyword queries. By harnessing the power of recent AI advancements, we enable large-scale visual research that was previously impractical, empowering scholars in historical studies, art history, and cultural heritage to explore visual motifs, artistic styles, and cross-cultural influences with new precision and speed. Applying our pipeline to over three million digitized manuscript pages, we automatically identified and extracted more than 200,000 unique illustrations. This scale of processing in under 0.06 seconds per page, dramatically outperforms traditional segmentation techniques in both efficiency and accessibility for visual scholarship. Our work demonstrates how cutting-edge AI tools can profoundly reshape scholarly workflows and open new avenues for multidisciplinary research in the age of digital manuscripts.
Authors: Xiran Fan, Zhimeng Jiang, Chin-Chia Michael Yeh, Yuzhong Chen, Yingtong Dou, Menghai Pan, Yan Zheng
Abstract: The ubiquity of payment networks generates vast transactional data encoding rich consumer and merchant behavioral patterns. Recent foundation models for transaction analysis process tabular data sequentially but rely on index-based representations for categorical merchant fields, causing substantial semantic information loss by converting rich textual data into discrete tokens. While Large Language Models (LLMs) can address this limitation through superior semantic understanding, their computational overhead challenges real-time financial deployment. We introduce a hybrid framework that uses LLM-generated embeddings as semantic initializations for lightweight transaction models, balancing interpretability with operational efficiency. Our approach employs multi-source data fusion to enrich merchant categorical fields and a one-word constraint principle for consistent embedding generation across LLM architectures. We systematically address data quality through noise filtering and context-aware enrichment. Experiments on large-scale transaction datasets demonstrate significant performance improvements across multiple transaction understanding tasks.
Authors: Benjamin Avanzi, Matthew Lambrianidis, Greg Taylor, Bernard Wong
Abstract: The use of neural networks trained on individual claims data has become increasingly popular in the actuarial reserving literature. We consider how to best input historical payment data in neural network models. Additionally, case estimates are also available in the format of a time series, and we extend our analysis to assessing their predictive power. In this paper, we compare a feed-forward neural network trained on summarised transactions to a recurrent neural network equipped to analyse a claim's entire payment history and/or case estimate development history. We draw conclusions from training and comparing the performance of the models on multiple, comparable highly complex datasets simulated from SPLICE (Avanzi, Taylor and Wang, 2023). We find evidence that case estimates will improve predictions significantly, but that equipping the neural network with memory only leads to meagre improvements. Although the case estimation process and quality will vary significantly between insurers, we provide a standardised methodology for assessing their value.
Authors: Nicholas R. Rasmussen, Rodrigue Rizk, Longwei Wang, Arun Singh, KC Santosh
Abstract: The early detection of Parkinsons disease remains a critical challenge in clinical neuroscience, with electroencephalography offering a noninvasive and scalable pathway toward population level screening. While machine learning has shown promise in this domain, many reported results suffer from methodological flaws, most notably patient level data leakage, inflating performance estimates and limiting clinical translation. To address these modeling pitfalls, we propose a unified evaluation framework grounded in nested cross validation and incorporating three complementary safeguards: (i) patient level stratification to eliminate subject overlap and ensure unbiased generalization, (ii) multi layered windowing to harmonize heterogeneous EEG recordings while preserving temporal dynamics, and (iii) inner loop channel selection to enable principled feature reduction without information leakage. Applied across three independent datasets with a heterogeneous number of channels, a convolutional neural network trained under this framework achieved 80.6% accuracy and demonstrated state of the art performance under held out population block testing, comparable to other methods in the literature. This performance underscores the necessity of nested cross validation as a safeguard against bias and as a principled means of selecting the most relevant information for patient level decisions, providing a reproducible foundation that can extend to other biomedical signal analysis domains.
Authors: Hector Zenil
Abstract: We formalise recursive self-training in Large Language Models (LLMs) and Generative AI as a discrete-time dynamical system and prove that, as training data become increasingly self-generated ($\alpha_t \to 0$), the system undergoes inevitably degenerative dynamics. We derive two fundamental failure modes: (1) Entropy Decay, where finite sampling effects cause a monotonic loss of distributional diversity (mode collapse), and (2) Variance Amplification, where the loss of external grounding causes the model's representation of truth to drift as a random walk, bounded only by the support diameter. We show these behaviours are not contingent on architecture but are consequences of distributional learning on finite samples. We further argue that Reinforcement Learning with imperfect verifiers suffers similar semantic collapse. To overcome these limits, we propose a path involving symbolic regression and program synthesis guided by Algorithmic Probability. The Coding Theorem Method (CTM) allows for identifying generative mechanisms rather than mere correlations, escaping the data-processing inequality that binds standard statistical learning. We conclude that while purely distributional learning leads to model collapse, hybrid neurosymbolic approaches offer a coherent framework for sustained self-improvement.
Authors: Luigi Favaro, Andrea Giammanco, Claudius Krause
Abstract: The high-dimensional complex nature of detectors makes fast calorimeter simulations a prime application for modern generative machine learning. Vision transformers (ViTs) can emulate the Geant4 response with unmatched accuracy and are not limited to regular geometries. Starting from the CaloDREAM architecture, we demonstrate the robustness and scalability of ViTs on regular and irregular geometries, and multiple detectors. Our results show that ViTs generate electromagnetic and hadronic showers statistically indistinguishable from Geant4 in multiple evaluation metrics, while maintaining the generation time in the $\mathcal{O}(10-100)$ ms on a single GPU. Furthermore, we show that pretraining on a large dataset and fine-tuning on the target geometry leads to reduced training costs and higher data efficiency, or altogether improves the fidelity of generated showers.
Authors: Rohan Vitthal Thorat, Rajdip Nayek
Abstract: Accurate prediction of structural dynamics is imperative for preserving digital twin fidelity throughout operational lifetimes. Parametric models with fixed nominal parameters often omit critical physical effects due to simplifications in geometry, material behavior, damping, or boundary conditions, resulting in model form errors (MFEs) that impair predictive accuracy. This work introduces a comprehensive framework for MFE estimation and correction in high-dimensional finite element (FE) based structural dynamical systems. The Gaussian Process Latent Force Model (GPLFM) represents discrepancies non-parametrically in the reduced modal domain, allowing a flexible data-driven characterization of unmodeled dynamics. A linear Bayesian filtering approach jointly estimates system states and discrepancies, incorporating epistemic and aleatoric uncertainties. To ensure computational tractability, the FE system is projected onto a reduced modal basis, and a mesh-invariant neural network maps modal states to discrepancy estimates, permitting model rectification across different FE discretizations without retraining. Validation is undertaken across five MFE scenarios-including incorrect beam theory, damping misspecification, misspecified boundary condition, unmodeled material nonlinearity, and local damage demonstrating the surrogate model's substantial reduction of displacement and rotation prediction errors under unseen excitations. The proposed methodology offers a potential means to uphold digital twin accuracy amid inherent modeling uncertainties.
Authors: Chung Han Tsai, ChengTo Lin, Chung Han Tsai, ChengTo Lin, Baowen Zhang, Qingyue Deng, Yunhui Zhao, Zhijia Song, Baowen Zhang, Qingyue Deng, Yunhui Zhao, Zhijia Song
Abstract: Amid the rapid digitalization of judicial systems, the integration of big data into adjudication remains underexplored, particularly in uncovering the structural logic of legal applications. This study bridges this gap by employing social network analysis (SNA) to examine credit card disputes involving personal information protection adjudicated in Beijing. By constructing a legal citation network, we reveal the latent patterns of substantive and procedural law application. The findings demonstrate that SNA can effectively identify core legal norms and typify cases, offering a robust methodological framework for optimizing 'Digital Court' systems. These insights provide practical pathways for enhancing judicial efficiency and consistency through data-driven case retrieval and holistic judicial information networks.
Authors: Alex Mulrooney, David Hong
Abstract: Canonical Polyadic (CP) tensor decomposition is a workhorse algorithm for discovering underlying low-dimensional structure in tensor data. This is accomplished in conventional CP decomposition by fitting a low-rank tensor to data with respect to the least-squares loss. Generalized CP (GCP) decompositions generalize this approach by allowing general loss functions that can be more appropriate, e.g., to model binary and count data or to improve robustness to outliers. However, GCP decompositions do not explicitly account for any symmetry in the tensors, which commonly arises in modern applications. For example, a tensor formed by stacking the adjacency matrices of a dynamic graph over time will naturally exhibit symmetry along the two modes corresponding to the graph nodes. In this paper, we develop a symmetric GCP (SymGCP) decomposition that allows for general forms of symmetry, i.e., symmetry along any subset of the modes. SymGCP accounts for symmetry by enforcing the corresponding symmetry in the decomposition. We derive gradients for SymGCP that enable its efficient computation via all-at-once optimization with existing tensor kernels. The form of the gradients also leads to various stochastic approximations that enable us to develop stochastic SymGCP algorithms that can scale to large tensors. We demonstrate the utility of the proposed SymGCP algorithms with a variety of experiments on both synthetic and real data.
Authors: Qiao Liu, Wing Hung Wong
Abstract: Modern data analysis increasingly requires flexible conditional inference P(X_B | X_A) where (X_A, X_B) is an arbitrary partition of observed variable X. Existing conditional inference methods lack this flexibility as they are tied to a fixed conditioning structure and cannot perform new conditional inference once trained. To solve this, we propose a Bayesian generative modeling (BGM) approach for arbitrary conditional inference without retraining. BGM learns a generative model of X through an iterative Bayesian updating algorithm where model parameters and latent variables are updated until convergence. Once trained, any conditional distribution can be obtained without retraining. Empirically, BGM achieves superior prediction performance with well calibrated predictive intervals, demonstrating that a single learned model can serve as a universal engine for conditional prediction with uncertainty quantification. We provide theoretical guarantees for the convergence of the stochastic iterative algorithm, statistical consistency and conditional-risk bounds. The proposed BGM framework leverages the power of AI to capture complex relationships among variables while adhering to Bayesian principles, emerging as a promising framework for advancing various applications in modern data science. The code for BGM is freely available at https://github.com/liuq-lab/bayesgm.
Authors: Zheng Luo, T Pranav Kutralingam, Ogochukwu N Okoani, Wanpeng Xu, Hua Wei, Xiyang Hu
Abstract: Large Language Models (LLMs) are increasingly deployed as agents that invoke external tools through structured function calls. While recent work reports strong tool-calling performance under standard English-centric evaluations, the robustness of tool calling under multilingual user interactions remains underexplored. In this work, we introduce MLCL, a diagnostic benchmark, and conduct a systematic evaluation of multilingual tool calling across Chinese, Hindi, and the low-resource language Igbo. Through fine-grained error analysis, we show that many failures occur despite correct intent understanding and tool selection. We identify parameter value language mismatch as a dominant failure mode, where models generate semantically appropriate parameter values in the user's language, violating language-invariant execution conventions. We further evaluate several inference-time system strategies and find that while these strategies substantially reduce language-induced execution errors, none of them can fully recover English-level performance.
Authors: Aleix Alcacer, Irene Epifanio
Abstract: Archetypal analysis serves as an exploratory tool that interprets a collection of observations as convex combinations of pure (extreme) patterns. When these patterns correspond to actual observations within the sample, they are termed archetypoids. For the first time, we propose applying archetypoid analysis to nominal observations, specifically for identifying archetypal cases from questionnaires featuring nominal multiple-choice questions with a single possible answer. This approach can enhance our understanding of a nominal data set, similar to its application in multivariate contexts. We compare this methodology with the use of archetype analysis and probabilistic archetypal analysis and demonstrate the benefits of this methodology using a real-world example: the German credit dataset.
Authors: Roberto Garrone
Abstract: This paper proposes a portfolio construction framework designed to remain robust under estimation error, non-stationarity, and realistic trading constraints. The methodology combines dynamic asset eligibility, deterministic rebalancing, and bounded multi-factor tilts applied to an equal-weight baseline. Asset eligibility is formalized as a state-dependent constraint on portfolio construction, allowing factor exposure to adjust endogenously in response to observable market conditions such as liquidity, volatility, and cross-sectional breadth. Rather than estimating expected returns or covariances, the framework relies on cross-sectional rankings and hard structural bounds to control concentration, turnover, and fragility. The resulting approach is fully algorithmic, transparent, and directly implementable. It provides a robustness-oriented alternative to parametric optimization and unconstrained multi-factor models, particularly suited for long-horizon allocations where stability and operational feasibility are primary objectives.
Authors: Getachew K. Befekadu
Abstract: This brief note considers the problem of learning with dynamic-optimizing principal-agent setting, in which the agents are allowed to have global perspectives about the learning process, i.e., the ability to view things according to their relative importances or in their true relations based-on some aggregated information shared by the principal. Whereas, the principal, which is exerting an influence on the learning process of the agents in the aggregation, is primarily tasked to solve a high-level optimization problem posed as an empirical-likelihood estimator under conditional moment restrictions model that also accounts information about the agents' predictive performances on out-of-samples as well as a set of private datasets available only to the principal. In particular, we present a coherent mathematical argument which is necessary for characterizing the learning process behind this abstract principal-agent learning framework, although we acknowledge that there are a few conceptual and theoretical issues still need to be addressed.
Authors: Dohyeong Ki, Adityanand Guntuboyina
Abstract: This paper establishes a rigorous theoretical foundation for the function class implicitly learned by XGBoost, bridging the gap between its empirical success and our theoretical understanding. We introduce an infinite-dimensional function class $\mathcal{F}^{d, s}_{\infty-\text{ST}}$ that extends finite ensembles of bounded-depth regression trees, together with a complexity measure $V^{d, s}_{\infty-\text{XGB}}(\cdot)$ that generalizes the $L^1$ regularization penalty used in XGBoost. We show that every optimizer of the XGBoost objective is also an optimizer of an equivalent penalized regression problem over $\mathcal{F}^{d, s}_{\infty-\text{ST}}$ with penalty $V^{d, s}_{\infty-\text{XGB}}(\cdot)$, providing an interpretation of XGBoost as implicitly targeting a broader function class. We also develop a smoothness-based interpretation of $\mathcal{F}^{d, s}_{\infty-\text{ST}}$ and $V^{d, s}_{\infty-\text{XGB}}(\cdot)$ in terms of Hardy--Krause variation. We prove that the least squares estimator over $\{f \in \mathcal{F}^{d, s}_{\infty-\text{ST}}: V^{d, s}_{\infty-\text{XGB}}(f) \le V\}$ achieves a nearly minimax-optimal rate of convergence $n^{-2/3} (\log n)^{4(\min(s, d) - 1)/3}$, thereby avoiding the curse of dimensionality. Our results provide the first rigorous characterization of the function space underlying XGBoost, clarify its connection to classical notions of variation, and identify an important open problem: whether the XGBoost algorithm itself achieves minimax optimality over this class.
Authors: Songze Li, Ruishi He, Xiaojun Jia, Jun Wang, Zhihui Fu
Abstract: Large Language Models (LLMs) face a significant threat from multi-turn jailbreak attacks, where adversaries progressively steer conversations to elicit harmful outputs. However, the practical effectiveness of existing attacks is undermined by several critical limitations: they struggle to maintain a coherent progression over long interactions, often losing track of what has been accomplished and what remains to be done; they rely on rigid or pre-defined patterns, and fail to adapt to the LLM's dynamic and unpredictable conversational state. To address these shortcomings, we introduce Mastermind, a multi-turn jailbreak framework that adopts a dynamic and self-improving approach. Mastermind operates in a closed loop of planning, execution, and reflection, enabling it to autonomously build and refine its knowledge of model vulnerabilities through interaction. It employs a hierarchical planning architecture that decouples high-level attack objectives from low-level tactical execution, ensuring long-term focus and coherence. This planning is guided by a knowledge repository that autonomously discovers and refines effective attack patterns by reflecting on interactive experiences. Mastermind leverages this accumulated knowledge to dynamically recombine and adapt attack vectors, dramatically improving both effectiveness and resilience. We conduct comprehensive experiments against state-of-the-art models, including GPT-5 and Claude 3.7 Sonnet. The results demonstrate that Mastermind significantly outperforms existing baselines, achieving substantially higher attack success rates and harmfulness ratings. Moreover, our framework exhibits notable resilience against multiple advanced defense mechanisms.
Authors: Sahil Wadhwa, Himanshu Kumar, Guanqun Yang, Abbaas Alif Mohamed Nishar, Pranab Mohanty, Swapnil Shinde, Yue Wu
Abstract: Large Language Models (LLMs) are powerful candidates for complex decision-making, leveraging vast encoded knowledge and remarkable zero-shot abilities. However, their adoption in high-stakes environments is hindered by their opacity; their outputs lack faithful explanations and cannot be effectively contested to correct errors, undermining trustworthiness. In this paper, we propose ART (Adaptive Reasoning Trees), a hierarchical method for claim verification. The process begins with a root claim, which branches into supporting and attacking child arguments. An argument's strength is determined bottom-up via a pairwise tournament of its children, adjudicated by a judge LLM, allowing a final, transparent and contestable verdict to be systematically derived which is missing in methods like Chain-of-Thought (CoT). We empirically validate ART on multiple datasets, analyzing different argument generators and comparison strategies. Our findings show that ART's structured reasoning outperforms strong baselines, establishing a new benchmark for explainable claim verification which is more reliable and ensures clarity in the overall decision making step.
Authors: Samuel E. Johnny, Bernes L. Atabonfack, Israel Alagbe, Assane Gueye
Abstract: Accurate tumor segmentation and classification in breast ultrasound (BUS) imaging remain challenging due to low contrast, speckle noise, and diverse lesion morphology. This study presents a multi-task deep learning framework that jointly performs lesion segmentation and diagnostic classification using embeddings from the Segment Anything Model (SAM) vision encoder. Unlike prompt-based SAM variants, our approach employs a prompt-free, fully supervised adaptation where high-dimensional SAM features are decoded through either a lightweight convolutional head or a UNet-inspired decoder for pixel-wise segmentation. The classification branch is enhanced via mask-guided attention, allowing the model to focus on lesion-relevant features while suppressing background artifacts. Experiments on the PRECISE 2025 breast ultrasound dataset, split per class into 80 percent training and 20 percent testing, show that the proposed method achieves a Dice Similarity Coefficient (DSC) of 0.887 and an accuracy of 92.3 percent, ranking among the top entries on the PRECISE challenge leaderboard. These results demonstrate that SAM-based representations, when coupled with segmentation-guided learning, significantly improve both lesion delineation and diagnostic prediction in breast ultrasound imaging.
Authors: Ricardo Vinuesa, Steven L. Brunton, Gianmarco Mengaldo
Abstract: Artificial intelligence now outperforms humans in several scientific and engineering tasks, yet its internal representations often remain opaque. In this Perspective, we argue that explainable artificial intelligence (XAI), combined with causal reasoning, enables {\it learning from the learners}. Focusing on discovery, optimization and certification, we show how the combination of foundation models and explainability methods allows the extraction of causal mechanisms, guides robust design and control, and supports trust and accountability in high-stakes applications. We discuss challenges in faithfulness, generalization and usability of explanations, and propose XAI as a unifying framework for human-AI collaboration in science and engineering.
Authors: Kamyar Barakati, Haochen Zhu, C Charlotte Buchanan, Dustin A Gilbert, Philip Rack, Sergei V. Kalinin
Abstract: Combinatorial materials libraries are an efficient route to generate large families of candidate compositions, but their impact is often limited by the speed and depth of characterization and by the difficulty of extracting actionable structure-property relations from complex characterization data. Here we develop an autonomous scanning probe microscopy (SPM) framework that integrates automated atomic force and magnetic force microscopy (AFM/MFM) to rapidly explore magnetic and structural properties across combinatorial spread libraries. To enable automated exploration of systems without a clear optimization target, we introduce a combination of a static physics-informed bag-of-features (BoF) representation of measured surface morphology and magnetic structure with multi-objective Bayesian optimization (MOBO) to discover the relative significance and robustness of features. The resulting closed-loop workflow selectively samples the compositional gradient and reconstructs feature landscapes consistent with dense grid "ground truth" measurements. The resulting Pareto structure reveals where multiple nanoscale objectives are simultaneously optimized, where trade-offs between roughness, coherence, and magnetic contrast are unavoidable, and how families of compositions cluster into distinct functional regimes, thereby turning multi-feature imaging data into interpretable maps of competing structure-property trends. While demonstrated for Au-Co-Ni and AFM/MFM, the approach is general and can be extended to other combinatorial systems, imaging modalities, and feature sets, illustrating how feature-based MOBO and autonomous SPM can transform microscopy images from static data products into active feedback for real-time, multi-objective materials discovery.
Authors: Eliatan Niktab, Hardip Patel
Abstract: Tokenization sits at the boundary between high-throughput genomic input and GPU compute, posing challenges in both algorithm design and system throughput. Overlapping k-mer tokenization can introduce information leakage under masked language modeling (MLM) and may degrade downstream accuracy. Single-nucleotide tokenization avoids leakage and preserves per-base fidelity, but it greatly increases sequence length for attention-based architectures. Non-overlapping k-mers and byte-pair encoding (BPE) provide compression and avoid leakage, at the cost of boundary sensitivity or reduced interpretability. Empirically, the choice of tokenization interacts strongly with model architecture and task requirements. At the system level, however, standard string tokenizers and host-bound vocabulary lookups dominate wall-clock time once inputs reach billions of bases, regardless of the tokenization algorithm. We present DNATok, a high-performance, GPU-first tokenization system that replaces general-purpose string processing with byte lookup table (LUT)-based identifier streaming and an overlapped host-to-device (H2D)/compute pipeline using pinned memory and architectural parallelism. DNATok is vocabulary-agnostic: it accelerates single-nucleotide, non-overlapping k-mer, and BPE tokenization, and integrates as a drop-in systems layer beneath genomic foundation models. DNATok achieves 84-95x higher encoding throughput than optimized Hugging Face baselines and up to 1.9x higher H2D throughput. End-to-end streaming reaches 1.27-1.84e8 tokens/s depending on configuration, effectively removing tokenization as a bottleneck for production-scale training and inference.
Authors: Adam Bodicoat, Gunel Jahangirova, Valerio Terragni
Abstract: Automated unit test generation aims to improve software quality while reducing the time and effort required for creating tests manually. However, existing techniques primarily generate regression oracles that predicate on the implemented behavior of the class under test. They do not address the oracle problem: the challenge of distinguishing correct from incorrect program behavior. With the rise of Foundation Models (FMs), particularly Large Language Models (LLMs), there is a new opportunity to generate test oracles that reflect intended behavior. This positions LLMs as enablers of Promptware, where software creation and testing are driven by natural-language prompts. This paper presents an empirical study on the effectiveness of LLMs in generating test oracles that expose software failures. We investigate how different prompting strategies and levels of contextual input impact the quality of LLM-generated oracles. Our findings offer insights into the strengths and limitations of LLM-based oracle generation in the FM era, improving our understanding of their capabilities and fostering future research in this area.
Authors: Hai Man, Chaobo Wang, Jia-Rui Li, Yuping Tian, Shu-Gang Chen
Abstract: Traditional methods for determining critical parameters are often influenced by human factors. This research introduces a physics-inspired adaptive reinforcement learning framework that enables agents to autonomously interact with physical environments, simultaneously identifying both the critical temperature and various types of critical exponents in the Ising model with precision. Interestingly, our algorithm exhibits search behavior reminiscent of phase transitions, efficiently converging to target parameters regardless of initial conditions. Experimental results demonstrate that this method significantly outperforms traditional approaches, particularly in environments with strong perturbations. This study not only incorporates physical concepts into machine learning to enhance algorithm interpretability but also establishes a new paradigm for scientific exploration, transitioning from manual analysis to autonomous AI discovery.
Authors: Benjamin Rozonoyer, Chong You, Michael Boratko, Himanshu Jain, Nilesh Gupta, Srinadh Bhojanapalli, Andrew McCallum, Felix Yu
Abstract: Dual and cross encoders have long been mainstays of information retrieval (IR), but are being challenged by the emergent capabilities of LLMs. An LLM-based approach we term pointwise generative ranking - generating tokens the length of a single docID as opposed to a list in order to enable ranking via beam search - combines efficiency and expressivity benefits while leveraging the in-context capabilities of Causal Transformers. Although there is ample evidence to suggest that pretrained LLMs are well-suited for ranking, we find that the vast majority of LLM-based approaches rely on next-token prediction, a loss function which is fundamentally rank-agnostic (and especially so with pointwise supervision). In this paper, we first prove that the expressivity of pointwise generative ranking with multi-token docIDs is superior to that of dual encoders. We then propose SToICaL - a Simple Token-Item Calibrated Loss - which can incorporate rank-aware supervision at both the item and token levels within the pointwise setup. We run a suite of experiments on ranking tasks derived from WordNet (Fellbaum, 1998) and ESCI (Reddy et al., arXiv:2206.06588). Two variants of SToICaL successfully suppress the probability of invalid docID generations and improve on common ranking metrics beyond top-1 retrieval.
Authors: Takito Sawada, Akinori Iwata, Masahiro Okuda
Abstract: Convolutional Neural Networks (CNNs) are known to exhibit a strong texture bias, favoring local patterns over global shape information--a tendency inherent to their convolutional architecture. While this bias is beneficial for texture-rich natural images, it often degrades performance on shape-dominant data such as illustrations and sketches. Although prior work has proposed shape-biased models to mitigate this issue, these approaches lack a quantitative metric for identifying which datasets would actually benefit from such modifications. To address this gap, we propose a data-driven metric that quantifies the shape-texture balance of a dataset by computing the Structural Similarity Index (SSIM) between each image's luminance channel and its L0-smoothed counterpart. Building on this metric, we further introduce a computationally efficient adaptation method that promotes shape bias by modifying the dilation of max-pooling operations while keeping convolutional weights frozen. Experimental results show that this approach consistently improves classification accuracy on shape-dominant datasets, particularly in low-data regimes where full fine-tuning is impractical, requiring training only the final classification layer.
Authors: Chuhan Wang, Xintong Li, Jennifer Yuntong Zhang, Junda Wu, Chengkai Huang, Lina Yao, Julian McAuley, Jingbo Shang
Abstract: Multimodal large language models often struggle with faithful reasoning in complex visual scenes, where intricate entities and relations require precise visual grounding at each step. This reasoning unfaithfulness frequently manifests as hallucinated entities, mis-grounded relations, skipped steps, and over-specified reasoning. Existing preference-based approaches, typically relying on textual perturbations or answer-conditioned rationales, fail to address this challenge as they allow models to exploit language priors to bypass visual grounding. To address this, we propose SceneAlign, a framework that leverages scene graphs as structured visual information to perform controllable structural interventions. By identifying reasoning-critical nodes and perturbing them through four targeted strategies that mimic typical grounding failures, SceneAlign constructs hard negative rationales that remain linguistically plausible but are grounded in inaccurate visual facts. These contrastive pairs are used in Direct Preference Optimization to steer models toward fine-grained, structure-faithful reasoning. Across seven visual reasoning benchmarks, SceneAlign consistently improves answer accuracy and reasoning faithfulness, highlighting the effectiveness of grounding-aware alignment for multimodal reasoning.
Authors: Emily Cheng, Carmen Amo Alonso, Federico Danieli, Arno Blaas, Luca Zappella, Pau Rodriguez, Xavier Suau
Abstract: As generative models become ubiquitous, there is a critical need for fine-grained control over the generation process. Yet, while controlled generation methods from prompting to fine-tuning proliferate, a fundamental question remains unanswered: are these models truly controllable in the first place? In this work, we provide a theoretical framework to formally answer this question. Framing human-model interaction as a control process, we propose a novel algorithm to estimate the controllable sets of models in a dialogue setting. Notably, we provide formal guarantees on the estimation error as a function of sample complexity: we derive probably-approximately correct bounds for controllable set estimates that are distribution-free, employ no assumptions except for output boundedness, and work for any black-box nonlinear control system (i.e., any generative model). We empirically demonstrate the theoretical framework on different tasks in controlling dialogue processes, for both language models and text-to-image generation. Our results show that model controllability is surprisingly fragile and highly dependent on the experimental setting. This highlights the need for rigorous controllability analysis, shifting the focus from simply attempting control to first understanding its fundamental limits.
Authors: Caroline Mazini Rodrigues (IRISA, CNRS), Nicolas Keriven (CNRS, IRISA, COMPACT), Thomas Maugey (COMPACT)
Abstract: Deep learning models for image compression often face practical limitations in hardware-constrained applications. Although these models achieve high-quality reconstructions, they are typically complex, heavyweight, and require substantial training data and computational resources. We propose a methodology to partially compress these networks by reducing the size of their encoders. Our approach uses a simplified knowledge distillation strategy to approximate the latent space of the original models with less data and shorter training, yielding lightweight encoders from heavyweight ones. We evaluate the resulting lightweight encoders across two different architectures on the image compression task. Experiments show that our method preserves reconstruction quality and statistical fidelity better than training lightweight encoders with the original loss, making it practical for resource-limited environments.
Authors: Alireza Dehghanpour Farashah, Aditi Khandelwal, Marylou Fauchard, Zhuan Shi, Negar Rostamzadeh, Golnoosh Farnadi
Abstract: As multilingual large language models become more widely used, ensuring their safety and fairness across diverse linguistic contexts presents unique challenges. While existing research on machine unlearning has primarily focused on monolingual settings, typically English, multilingual environments introduce additional complexities due to cross-lingual knowledge transfer and biases embedded in both pretraining and fine-tuning data. In this work, we study multilingual unlearning using the Aya-Expanse 8B model under two settings: (1) data unlearning and (2) concept unlearning. We extend benchmarks for factual knowledge and stereotypes to ten languages through translation: English, French, Arabic, Japanese, Russian, Farsi, Korean, Hindi, Hebrew, and Indonesian. These languages span five language families and a wide range of resource levels. Our experiments show that unlearning in high-resource languages is generally more stable, with asymmetric transfer effects observed between typologically related languages. Furthermore, our analysis of linguistic distances indicates that syntactic similarity is the strongest predictor of cross-lingual unlearning behavior.
Authors: Haoran Wang, Xuanyi Zhang, Shuangsang Fang, Longke Ran, Ziqing Deng, Yong Zhang, Yuxiang Li, Shaoshuai Li
Abstract: Recent advancements in single-cell multi-omics, particularly RNA-seq, have provided profound insights into cellular heterogeneity and gene regulation. While pre-trained language model (PLM) paradigm based single-cell foundation models have shown promise, they remain constrained by insufficient integration of in-depth individual profiles and neglecting the influence of noise within multi-modal data. To address both issues, we propose an Open-world Language Knowledge-Aided Robust Single-Cell Foundation Model (OKR-CELL). It is built based on a cross-modal Cell-Language pre-training framework, which comprises two key innovations: (1) leveraging Large Language Models (LLMs) based workflow with retrieval-augmented generation (RAG) enriches cell textual descriptions using open-world knowledge; (2) devising a Cross-modal Robust Alignment (CRA) objective that incorporates sample reliability assessment, curriculum learning, and coupled momentum contrastive learning to strengthen the model's resistance to noisy data. After pretraining on 32M cell-text pairs, OKR-CELL obtains cutting-edge results across 6 evaluation tasks. Beyond standard benchmarks such as cell clustering, cell-type annotation, batch-effect correction, and few-shot annotation, the model also demonstrates superior performance in broader multi-modal applications, including zero-shot cell-type annotation and bidirectional cell-text retrieval.
Authors: Gianmario Voria, Moses Openja, Foutse Khomh, Gemma Catolino, Fabio Palomba
Abstract: The advent of transformer-based language models has reshaped how AI systems process and generate text. In software engineering (SE), these models now support diverse activities, accelerating automation and decision-making. Yet, evidence shows that these models can reproduce or amplify social biases, raising fairness concerns. Recent work on neuron editing has shown that internal activations in pre-trained transformers can be traced and modified to alter model behavior. Building on the concept of knowledge neurons, neurons that encode factual information, we hypothesize the existence of biased neurons that capture stereotypical associations within pre-trained transformers. To test this hypothesis, we build a dataset of biased relations, i.e., triplets encoding stereotypes across nine bias types, and adapt neuron attribution strategies to trace and suppress biased neurons in BERT models. We then assess the impact of suppression on SE tasks. Our findings show that biased knowledge is localized within small neuron subsets, and suppressing them substantially reduces bias with minimal performance loss. This demonstrates that bias in transformers can be traced and mitigated at the neuron level, offering an interpretable approach to fairness in SE.
Authors: Thomas Fabian
Abstract: Understanding how large language models (LLMs) represent natural language is a central challenge in natural language processing (NLP) research. Many existing methods extract word embeddings from an LLM, visualise the embedding space via point-plots, and compare the relative positions of certain words. However, this approach only considers single words and not whole natural language expressions, thus disregards the context in which a word is used. Here we present a novel tool for analysing and visualising information flow in natural language expressions by applying diffusion tensor imaging (DTI) to word embeddings. We find that DTI reveals how information flows between word embeddings. Tracking information flows within the layers of an LLM allows for comparing different model structures and revealing opportunities for pruning an LLM's under-utilised layers. Furthermore, our model reveals differences in information flows for tasks like pronoun resolution and metaphor detection. Our results show that our model permits novel insights into how LLMs represent actual natural language expressions, extending the comparison of isolated word embeddings and improving the interpretability of NLP models.
Authors: Guray Ozgur, Eduarda Caldeira, Tahar Chettaoui, Jan Niklas Kolf, Marco Huber, Naser Damer, Fadi Boutros
Abstract: Face Image Quality Assessment (FIQA) is essential for reliable face recognition systems. Current approaches primarily exploit only final-layer representations, while training-free methods require multiple forward passes or backpropagation. We propose ViTNT-FIQA, a training-free approach that measures the stability of patch embedding evolution across intermediate Vision Transformer (ViT) blocks. We demonstrate that high-quality face images exhibit stable feature refinement trajectories across blocks, while degraded images show erratic transformations. Our method computes Euclidean distances between L2-normalized patch embeddings from consecutive transformer blocks and aggregates them into image-level quality scores. We empirically validate this correlation on a quality-labeled synthetic dataset with controlled degradation levels. Unlike existing training-free approaches, ViTNT-FIQA requires only a single forward pass without backpropagation or architectural modifications. Through extensive evaluation on eight benchmarks (LFW, AgeDB-30, CFP-FP, CALFW, Adience, CPLFW, XQLFW, IJB-C), we show that ViTNT-FIQA achieves competitive performance with state-of-the-art methods while maintaining computational efficiency and immediate applicability to any pre-trained ViT-based face recognition model.
Authors: Tianwang Jia, Xiaoqing Chen, Dongrui Wu
Abstract: Electroencephalogram (EEG)-based brain-computer interfaces (BCIs) are widely adopted due to their efficiency and portability; however, their decoding algorithms still face multiple challenges, including inadequate generalization, adversarial vulnerability, and privacy leakage. This paper proposes Secure and Accurate FEderated learning (SAFE), a federated learning-based approach that protects user privacy by keeping data local during model training. SAFE employs local batch-specific normalization to mitigate cross-subject feature distribution shifts and hence improves model generalization. It further enhances adversarial robustness by introducing perturbations in both the input space and the parameter space through federated adversarial training and adversarial weight perturbation. Experiments on five EEG datasets from motor imagery (MI) and event-related potential (ERP) BCI paradigms demonstrated that SAFE consistently outperformed 14 state-of-the-art approaches in both decoding accuracy and adversarial robustness, while ensuring privacy protection. Notably, it even outperformed centralized training approaches that do not consider privacy protection at all. To our knowledge, SAFE is the first algorithm to simultaneously achieve high decoding accuracy, strong adversarial robustness, and reliable privacy protection without using any calibration data from the target subject, making it highly desirable for real-world BCIs.
Authors: Eilam Cohen, Itamar Bul, Danielle Inbar, Omri Loewenbach
Abstract: Large language models (LLMs) enable strong text generation, and in general there is a practical tradeoff between fine-tuning and prompt engineering. We introduce Simplify-This, a comparative study evaluating both paradigms for text simplification with encoder-decoder LLMs across multiple benchmarks, using a range of evaluation metrics. Fine-tuned models consistently deliver stronger structural simplification, whereas prompting often attains higher semantic similarity scores yet tends to copy inputs. A human evaluation favors fine-tuned outputs overall. We release code, a cleaned derivative dataset used in our study, checkpoints of fine-tuned models, and prompt templates to facilitate reproducibility and future work.
Authors: Xiaoshuai Song, Haofei Chang, Guanting Dong, Yutao Zhu, Zhicheng Dou, Ji-Rong Wen
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: ChunTeng Chen, YiChen Hsu, YiWen Liu, WeiFang Sun, TsaiChing Ni, ChunYi Lee, Min Sun, YuanFu Yang
Abstract: The ability to automatically generate large-scale, interactive, and physically realistic 3D environments is crucial for advancing robotic learning and embodied intelligence. However, existing generative approaches often fail to capture the functional complexity of real-world interiors, particularly those containing articulated objects with movable parts essential for manipulation and navigation. This paper presents SceneFoundry, a language-guided diffusion framework that generates apartment-scale 3D worlds with functionally articulated furniture and semantically diverse layouts for robotic training. From natural language prompts, an LLM module controls floor layout generation, while diffusion-based posterior sampling efficiently populates the scene with articulated assets from large-scale 3D repositories. To ensure physical usability, SceneFoundry employs differentiable guidance functions to regulate object quantity, prevent articulation collisions, and maintain sufficient walkable space for robotic navigation. Extensive experiments demonstrate that our framework generates structurally valid, semantically coherent, and functionally interactive environments across diverse scene types and conditions, enabling scalable embodied AI research.
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: Kaichen Shen, Peng Chen
Abstract: Sequential Bayesian optimal experimental design (SBOED) for PDE-governed inverse problems is computationally challenging, especially for infinite-dimensional random field parameters. High-fidelity approaches require repeated forward and adjoint PDE solves inside nested Bayesian inversion and design loops. We formulate SBOED as a finite-horizon Markov decision process and learn an amortized design policy via policy-gradient reinforcement learning (PGRL), enabling online design selection from the experiment history without repeatedly solving an SBOED optimization problem. To make policy training and reward evaluation scalable, we combine dual dimension reduction -- active subspace projection for the parameter and principal component analysis for the state -- with an adjusted derivative-informed latent attention neural operator (LANO) surrogate that predicts both the parameter-to-solution map and its Jacobian. We use a Laplace-based D-optimality reward while noting that, in general, other expected-information-gain utilities such as KL divergence can also be used within the same framework. We further introduce an eigenvalue-based evaluation strategy that uses prior samples as proxies for maximum a posteriori (MAP) points, avoiding repeated MAP solves while retaining accurate information-gain estimates. Numerical experiments on sequential multi-sensor placement for contaminant source tracking demonstrate approximately $100\times$ speedup over high-fidelity finite element methods, improved performance over random sensor placements, and physically interpretable policies that discover an ``upstream'' tracking strategy.
Authors: Constantinos Karouzos, Xingwei Tan, Nikolaos Aletras
Abstract: Preference tuning aligns pretrained language models to human judgments of quality, helpfulness, or safety by optimizing over explicit preference signals rather than likelihood alone. Prior work has shown that preference-tuning degrades performance and reduces helpfulness when evaluated outside the training domain. However, the extent to which adaptation strategies mitigate this domain shift remains unexplored. We address this challenge by conducting a comprehensive and systematic study of alignment generalization under domain shift. We compare five popular alignment objectives and various adaptation strategies from source to target, including target-domain supervised fine-tuning and pseudo-labeling, across summarization and question-answering helpfulness tasks. Our findings reveal systematic differences in generalization across alignment objectives under domain shift. We show that adaptation strategies based on pseudo-labeling can substantially reduce domain-shift degradation
Authors: Haoming Xu, Ningyuan Zhao, Yunzhi Yao, Weihong Xu, Hongru Wang, Xinle Deng, Shumin Deng, Jeff Z. Pan, Huajun Chen, Ningyu Zhang
Abstract: As Large Language Models (LLMs) are increasingly deployed in real-world settings, correctness alone is insufficient. Reliable deployment requires maintaining truthful beliefs under contextual perturbations. Existing evaluations largely rely on point-wise confidence like Self-Consistency, which can mask brittle belief. We show that even facts answered with perfect self-consistency can rapidly collapse under mild contextual interference. To address this gap, we propose Neighbor-Consistency Belief (NCB), a structural measure of belief robustness that evaluates response coherence across a conceptual neighborhood. To validate the efficiency of NCB, we introduce a new cognitive stress-testing protocol that probes outputs stability under contextual interference. Experiments across multiple LLMs show that the performance of high-NCB data is relatively more resistant to interference. Finally, we present Structure-Aware Training (SAT), which optimizes context-invariant belief structure and reduces long-tail knowledge brittleness by approximately 30%. Code will be available at https://github.com/zjunlp/belief.
Authors: Yigitcan Comlek, R. Murali Krishnan, Sandipp Krishnan Ravi, Amin Moghaddas, Rafael Giorjao, Michael Eff, Anirban Samaddar, Nesar S. Ramachandra, Sandeep Madireddy, Liping Wang
Abstract: Modern engineering and scientific workflows often require simultaneous predictions across related tasks and fidelity levels, where high-fidelity data is scarce and expensive, while low-fidelity data is more abundant. This paper introduces an Multi-Task Gaussian Processes (MTGP) framework tailored for engineering systems characterized by multi-source, multi-fidelity data, addressing challenges of data sparsity and varying task correlations. The proposed framework leverages inter-task relationships across outputs and fidelity levels to improve predictive performance and reduce computational costs. The framework is validated across three representative scenarios: Forrester function benchmark, 3D ellipsoidal void modeling, and friction-stir welding. By quantifying and leveraging inter-task relationships, the proposed MTGP framework offers a robust and scalable solution for predictive modeling in domains with significant computational and experimental costs, supporting informed decision-making and efficient resource utilization.
Authors: E. Middell, L. Carlton, S. Moradi, T. Codina, T. Fischer, J. Cutler, S. Kelley, J. Behrendt, T. Dissanayake, N. Harmening, M. A. Y\"ucel, D. A. Boas, A. von L\"uhmann
Abstract: Functional near-infrared spectroscopy (fNIRS) and diffuse optical tomography (DOT) are rapidly evolving toward wearable, multimodal, and data-driven, AI-supported neuroimaging in the everyday world. However, current analytical tools are fragmented across platforms, limiting reproducibility, interoperability, and integration with modern machine learning (ML) workflows. Cedalion is a Python-based open-source framework designed to unify advanced model-based and data-driven analysis of multimodal fNIRS and DOT data within a reproducible, extensible, and community-driven environment. Cedalion integrates forward modelling, photogrammetric optode co-registration, signal processing, GLM Analysis, DOT image reconstruction, and ML-based data-driven methods within a single standardized architecture based on the Python ecosystem. It adheres to SNIRF and BIDS standards, supports cloud-executable Jupyter notebooks, and provides containerized workflows for scalable, fully reproducible analysis pipelines that can be provided alongside original research publications. Cedalion connects established optical-neuroimaging pipelines with ML frameworks such as scikit-learn and PyTorch, enabling seamless multimodal fusion with EEG, MEG, and physiological data. It implements validated algorithms for signal-quality assessment, motion correction, GLM modelling, and DOT reconstruction, complemented by modules for simulation, data augmentation, and multimodal physiology analysis. Automated documentation links each method to its source publication, and continuous-integration testing ensures robustness. This tutorial paper provides seven fully executable notebooks that demonstrate core features. Cedalion offers an open, transparent, and community extensible foundation that supports reproducible, scalable, cloud- and ML-ready fNIRS/DOT workflows for laboratory-based and real-world neuroimaging.
Authors: Jingsheng Zheng, Jintian Zhang, Yujie Luo, Yuren Mao, Yunjun Gao, Lun Du, Huajun Chen, Ningyu Zhang
Abstract: Autonomous machine learning agents have revolutionized scientific discovery, yet they remain constrained by a Generate-Execute-Feedback paradigm. Previous approaches suffer from a severe Execution Bottleneck, as hypothesis evaluation relies strictly on expensive physical execution. To bypass these physical constraints, we internalize execution priors to substitute costly runtime checks with instantaneous predictive reasoning, drawing inspiration from World Models. In this work, we formalize the task of Data-centric Solution Preference and construct a comprehensive corpus of 18,438 pairwise comparisons. We demonstrate that LLMs exhibit significant predictive capabilities when primed with a Verified Data Analysis Report, achieving 61.5% accuracy and robust confidence calibration. Finally, we instantiate this framework in FOREAGENT, an agent that employs a Predict-then-Verify loop, achieving a 6x acceleration in convergence while surpassing execution-based baselines by +6%. Our code and dataset will be publicly available soon at https://github.com/zjunlp/predict-before-execute.
Authors: Pankaj Gupta, Priya Mudgil, Niharika Dutta, Kartik Bose, Nitish Kumar, Anupam Kumar, Jimil Shah, Vaneet Jearth, Jayanta Samanta, Vishal Sharma, Harshal Mandavdhare, Surinder Rana, Saroj K Sinha, Usha Dutta
Abstract: Background: Pancreatic cancer is one of the most aggressive cancers, with poor survival rates. Endoscopic ultrasound (EUS) is a key diagnostic modality, but its effectiveness is constrained by operator subjectivity. This study evaluates a Vision Transformer-based deep learning segmentation model for pancreatic tumors. Methods: A segmentation model using the USFM framework with a Vision Transformer backbone was trained and validated with 17,367 EUS images (from two public datasets) in 5-fold cross-validation. The model was tested on an independent dataset of 350 EUS images from another public dataset, manually segmented by radiologists. Preprocessing included grayscale conversion, cropping, and resizing to 512x512 pixels. Metrics included Dice similarity coefficient (DSC), intersection over union (IoU), sensitivity, specificity, and accuracy. Results: In 5-fold cross-validation, the model achieved a mean DSC of 0.651 +/- 0.738, IoU of 0.579 +/- 0.658, sensitivity of 69.8%, specificity of 98.8%, and accuracy of 97.5%. For the external validation set, the model achieved a DSC of 0.657 (95% CI: 0.634-0.769), IoU of 0.614 (95% CI: 0.590-0.689), sensitivity of 71.8%, and specificity of 97.7%. Results were consistent, but 9.7% of cases exhibited erroneous multiple predictions. Conclusions: The Vision Transformer-based model demonstrated strong performance for pancreatic tumor segmentation in EUS images. However, dataset heterogeneity and limited external validation highlight the need for further refinement, standardization, and prospective studies.
Authors: Akhil S. Nair, Lucas Foppa
Abstract: Active learning (AL) plays a critical role in materials science, enabling applications such as the construction of machine-learning interatomic potentials for atomistic simulations and the operation of self-driving laboratories. Despite its widespread use, the reliability and effectiveness of AL workflows depend on implicit design assumptions that are rarely examined systematically. Here, we critically assess AL workflows deployed in materials science and investigate how key design choices, such as surrogate models, sampling strategies, uncertainty quantification and evaluation metrics, relate to their performance. By identifying common pitfalls and discussing practical mitigation strategies, we provide guidance to practitioners for the efficient design, assessment, and interpretation of AL workflows in materials science.
Authors: Kieran Wood, Stephen J. Roberts, Stefan Zohren
Abstract: We propose DeePM (Deep Portfolio Manager), a structured deep-learning macro portfolio manager trained end-to-end to maximize a robust, risk-adjusted utility. DeePM addresses three fundamental challenges in financial learning: (1) it resolves the asynchronous "ragged filtration" problem via a Directed Delay (Causal Sieve) mechanism that prioritizes causal impulse-response learning over information freshness; (2) it combats low signal-to-noise ratios via a Macroeconomic Graph Prior, regularizing cross-asset dependence according to economic first principles; and (3) it optimizes a distributionally robust objective where a smooth worst-window penalty serves as a differentiable proxy for Entropic Value-at-Risk (EVaR) - a window-robust utility encouraging strong performance in the most adverse historical subperiods. In large-scale backtests from 2010-2025 on 50 diversified futures with highly realistic transaction costs, DeePM attains net risk-adjusted returns that are roughly twice those of classical trend-following strategies and passive benchmarks, solely using daily closing prices. Furthermore, DeePM improves upon the state-of-the-art Momentum Transformer architecture by roughly fifty percent. The model demonstrates structural resilience across the 2010s "CTA (Commodity Trading Advisor) Winter" and the post-2020 volatility regime shift, maintaining consistent performance through the pandemic, inflation shocks, and the subsequent higher-for-longer environment. Ablation studies confirm that strictly lagged cross-sectional attention, graph prior, principled treatment of transaction costs, and robust minimax optimization are the primary drivers of this generalization capability.
Authors: Dror Jacoby, Yanzhi Li, Shuyue Yu, Nicola Di Cicco, Hagit Messer, Gil Zussman, Igor Kadota
Abstract: As emerging applications demand higher throughput and lower latencies, operators are increasingly deploying millimeter-wave (mmWave) links within x-haul transport networks, spanning fronthaul, midhaul, and backhaul segments. However, the inherent susceptibility of mmWave frequencies to weather-related attenuation, particularly rain fading, complicates the maintenance of stringent Quality of Service (QoS) requirements. This creates a critical challenge: making admission decisions under uncertainty regarding future network capacity. To address this, we develop a proactive slice admission control framework for mmWave x-haul networks subject to rain-induced fluctuations. Our objective is to improve network performance, ensure QoS, and optimize revenue, thereby surpassing the limitations of standard reactive approaches. The proposed framework integrates a deep learning predictor of future network conditions with a proactive Q-learning-based slice admission control mechanism. We validate our solution using real-world data from a mmWave x-haul deployment in a dense urban area, incorporating realistic models of link capacity attenuation and dynamic slice demands. Extensive evaluations demonstrate that our proactive solution achieves 2-3x higher long-term average revenue under dynamic link conditions, providing a scalable and resilient framework for adaptive admission control.
Authors: Isaiah J. King, Bernardo Trindade, Benjamin Bowman, H. Howie Huang
Abstract: Representing networks as a graph and training a link prediction model using benign connections is an effective method of anomaly-based intrusion detection. Existing works using this technique have shown great success using temporal graph neural networks and skip-gram-based approaches on random walks. However, random walk-based approaches are unable to incorporate rich edge data, while the GNN-based approaches require large amounts of memory to train. In this work, we propose extending the original insight from random walk-based skip-grams--that random walks through a graph are analogous to sentences in a corpus--to the more modern transformer-based foundation models. Using language models that take advantage of GPU optimizations, we can quickly train a graph foundation model to predict missing tokens in random walks through a network of computers. The graph foundation model is then finetuned for link prediction and used as a network anomaly detector. This new approach allows us to combine the efficiency of random walk-based methods and the rich semantic representation of deep learning methods. This system, which we call CyberGFM, achieved state-of-the-art results on three widely used network anomaly detection datasets, delivering a up to 2$\times$ improvement in average precision. We found that CyberGFM outperforms all prior works in unsupervised link prediction for network anomaly detection, using the same number of parameters, and with equal or better efficiency than the previous best approaches.
Authors: Sunia Tanweer, Firas A. Khasawneh
Abstract: We develop a practical framework for distinguishing diffusive stochastic processes from deterministic signals using only a single discrete time series. Our approach is based on classical excursion and crossing theorems for continuous semimartingales, which correlates number $N_\varepsilon$ of excursions of magnitude at least $\varepsilon$ with the quadratic variation $[X]_T$ of the process. The scaling law holds universally for all continuous semimartingales with finite quadratic variation, including general Ito diffusions with nonlinear or state-dependent volatility, but fails sharply for deterministic systems -- thereby providing a theoretically-certfied method of distinguishing between these dynamics, as opposed to the subjective entropy or recurrence based state of the art methods. We construct a robust data-driven diffusion test. The method compares the empirical excursion counts against the theoretical expectation. The resulting ratio $K(\varepsilon)=N_{\varepsilon}^{\mathrm{emp}}/N_{\varepsilon}^{\mathrm{theory}}$ is then summarized by a log-log slope deviation measuring the $\varepsilon^{-2}$ law that provides a classification into diffusion-like or not. We demonstrate the method on canonical stochastic systems, some periodic and chaotic maps and systems with additive white noise, as well as the stochastic Duffing system. The approach is nonparametric, model-free, and relies only on the universal small-scale structure of continuous semimartingales.
Authors: Johanna Tengler, Christoph Brune, Jos\'e A. Iglesias
Abstract: We study the discrete-to-continuum consistency of the training of shallow graph convolutional neural networks (GCNNs) on proximity graphs of sampled point clouds under a manifold assumption. Graph convolution is defined spectrally via the graph Laplacian, whose low-frequency spectrum approximates that of the Laplace-Beltrami operator of the underlying smooth manifold, and shallow GCNNs of possibly infinite width are linear functionals on the space of measures on the parameter space. From this functional-analytic perspective, graph signals are seen as spatial discretizations of functions on the manifold, which leads to a natural notion of training data consistent across graph resolutions. To enable convergence results, the continuum parameter space is chosen as a weakly compact product of unit balls, with Sobolev regularity imposed on the output weight and bias, but not on the convolutional parameter. The corresponding discrete parameter spaces inherit the corresponding spectral decay, and are additionally restricted by a frequency cutoff adapted to the informative spectral window of the graph Laplacians. Under these assumptions, we prove $\Gamma$-convergence of regularized empirical risk minimization functionals and corresponding convergence of their global minimizers, in the sense of weak convergence of the parameter measures and uniform convergence of the functions over compact sets. This provides a formalization of mesh and sample independence for the training of such networks.
Authors: Jinshuai Bai, Laith Alzubaidi, Qingxia Wang, Ellen Kuhl, Mohammed Bennamoun, Yuantong Gu
Abstract: Deep learning (DL) relies heavily on data, and the quality of data influences its performance significantly. However, obtaining high-quality, well-annotated datasets can be challenging or even impossible in many real-world applications, such as structural risk estimation and medical diagnosis. This presents a significant barrier to the practical implementation of DL in these fields. Physics-guided deep learning (PGDL) is a novel type of DL that can integrate physics laws to train neural networks. This can be applied to any systems that are controlled or governed by physics laws, such as mechanics, finance and medical applications. It has been demonstrated that, with the additional information provided by physics laws, PGDL achieves great accuracy and generalisation in the presence of data scarcity. This review provides a detailed examination of PGDL and offers a structured overview of its use in addressing data scarcity across various fields, including physics, engineering and medical applications. Moreover, the review identifies the current limitations and opportunities for PGDL in relation to data scarcity and offers a thorough discussion on the future prospects of PGDL.
Authors: Yuanzhi Li, Raghu Meka, Rina Panigrahy, Kulin Shah
Abstract: Supervised and unsupervised learning using deep neural networks typically aims to exploit the underlying structure in the training data; this structure is often explained using a latent generative process that produces the data, and the generative process is often hierarchical, involving latent concepts. Despite the significant work on understanding the learning of the latent structure and underlying concepts using theory and experiments, a framework that mathematically captures the definition of a concept and provides ways to operate on concepts is missing. In this work, we propose to characterize a simple primitive concept by the zero set of a collection of polynomials and use moment statistics of the data to uniquely represent the concepts; we show how this view can be used to obtain a signature of the concept. These signatures can be used to discover a common structure across the set of concepts and could recursively produce the signature of higher-level concepts from the signatures of lower-level concepts. To utilize such desired properties, we propose a method by keeping a dictionary of concepts and show that the proposed method can learn different types of hierarchical structures of the data.
Authors: Xinhao Zhang, Jinghan Zhang, Banafsheh Rekabdar, Yuanchun Zhou, Pengfei Wang, Kunpeng Liu
Abstract: The representation of feature space is a crucial environment where data points get vectorized and embedded for subsequent modeling. Thus the efficacy of machine learning (ML) algorithms is closely related to the quality of feature engineering. As one of the most important techniques, feature generation transforms raw data into an optimized feature space conducive to model training and further refines the space. Despite the advancements in automated feature engineering and feature generation, current methodologies often suffer from three fundamental issues: lack of explainability, limited applicability, and inflexible strategy. These shortcomings frequently hinder and limit the deployment of ML models across varied scenarios. Our research introduces a novel approach adopting large language models (LLMs) and feature-generating prompts to address these challenges. We propose a dynamic and adaptive feature generation method that enhances the interpretability of the feature generation process. Our approach broadens the applicability across various data types and tasks and offers advantages over strategic flexibility. A broad range of experiments showcases that our approach is significantly superior to existing methods.
Authors: Stefan Haufe, Rick Wilming, Benedict Clark, Rustam Zhumagambetov, Ahc\`ene Boubekki, J\"org Martin, Danny Panknin
Abstract: The field of "explainable artificial intelligence" (XAI) seemingly addresses the desire that decisions of machine learning systems should be human-understandable. However, in its current state, XAI itself needs scrutiny. Popular methods cannot reliably answer relevant questions about ML models, their training data, or test inputs, because they systematically attribute importance to input features that are independent of the prediction target. This limits the utility of XAI for diagnosing and correcting data and models, for scientific discovery, and for identifying intervention targets. The fundamental reason for this is that current XAI methods do not address well-defined problems and are not evaluated against targeted criteria of explanation correctness. Researchers should formally define the problems they intend to solve and design methods accordingly. This will lead to diverse use-case-dependent notions of explanation correctness and objective metrics of explanation performance that can be used to validate XAI algorithms.
Authors: M. Rostami, S. S. Kia
Abstract: Federated learning (FL) has gained considerable popularity for distributed machine learning due to its ability to preserve the privacy of participating agents by eliminating the need for data aggregation. Nevertheless, communication costs between agents and the central server in FL are substantial in large-scale problems and remain a limiting factor for this algorithm. This paper introduces an innovative algorithm, called FedScalar, within the FL framework aimed at improving communication efficiency. Unlike traditional FL methods that require agents to send high-dimensional vectors to the server, FedScalar enables agents to communicate updates using only two scalar values. Each agent encodes its updated model parameters into a scalar via the inner product between its local update difference and a random vector, which is transmitted to the server along with the agent's local random seed value. The server then averages the received scalar values and decodes the information by projecting the averaged scalar onto the regenerated random vector using the corresponding agent seed values. Our method thereby significantly reduces communication overhead. Technically, we demonstrate that the proposed algorithm achieves a convergence rate of O(d/\sqrt(K)) to a stationary point for smooth, non-convex loss functions. Additionally, our analysis shows that changing the underlying distribution of the random vector generated by the server from Gaussian to Rademacher distribution reduces the variance during the aggregation step of the algorithm. Finally, we validate the performance and communication efficiency of our algorithm with numerical simulations.
Authors: Xiaoxing Ren, Nicola Bastianello, Karl H. Johansson, Thomas Parisini
Abstract: We address distributed learning problems, both nonconvex and convex, over undirected networks. In particular, we design a novel algorithm based on the distributed Alternating Direction Method of Multipliers (ADMM) to address the challenges of high communication costs, and large datasets. Our design tackles these challenges i) by enabling the agents to perform multiple local training steps between each round of communications; and ii) by allowing the agents to employ stochastic gradients while carrying out local computations. We show that the proposed algorithm converges to a neighborhood of a stationary point, for nonconvex problems, and of an optimal point, for convex problems. We also propose a variant of the algorithm to incorporate variance reduction thus achieving exact convergence. We show that the resulting algorithm indeed converges to a stationary (or optimal) point, and moreover that local training accelerates convergence. We thoroughly compare the proposed algorithms with the state of the art, both theoretically and through numerical results.
Authors: Xinhao Zhang, Jinghan Zhang, Fengran Mo, Dongjie Wang, Yanjie Fu, Kunpeng Liu
Abstract: Humans excel in analogical learning and knowledge transfer and, more importantly, possess a unique understanding of identifying appropriate sources of knowledge. From a model's perspective, this presents an interesting challenge. If models could autonomously retrieve knowledge useful for transfer or decision-making to solve problems, they would transition from passively acquiring to actively accessing and learning from knowledge. However, filling models with knowledge is relatively straightforward -- it simply requires more training and accessible knowledge bases. The more complex task is teaching models about which knowledge can be analogized and transferred. Therefore, we design a knowledge augmentation method, LEKA, for knowledge transfer that actively searches for suitable knowledge sources that can enrich the target domain's knowledge. This LEKA method extracts key information from the target domain's textual information, retrieves pertinent data from external data libraries, and harmonizes retrieved data with the target domain data in feature space and marginal probability measures. We validate the effectiveness of our approach through extensive experiments across various domains and demonstrate significant improvements over traditional methods in reducing computational costs, automating data alignment, and optimizing transfer learning outcomes.
Authors: Samuele Bortolotti, Emanuele Marconato, Paolo Morettin, Andrea Passerini, Stefano Teso
Abstract: Concept-based Models are neural networks that learn a concept extractor to map inputs to high-level concepts and an inference layer to translate these into predictions. Ensuring these modules produce interpretable concepts and behave reliably in out-of-distribution is crucial, yet the conditions for achieving this remain unclear. We study this problem by establishing a novel connection between Concept-based Models and reasoning shortcuts (RSs), a common issue where models achieve high accuracy by learning low-quality concepts, even when the inference layer is fixed and provided upfront. Specifically, we extend RSs to the more complex setting of Concept-based Models and derive theoretical conditions for identifying both the concepts and the inference layer. Our empirical results highlight the impact of RSs and show that existing methods, even combined with multiple natural mitigation strategies, often fail to meet these conditions in practice.
Authors: Lorenzo Brigato, Rafael Morand, Knut Str{\o}mmen, Maria Panagiotou, Markus Schmidt, Stavroula Mougiakakou
Abstract: Recent advances in long-term time series forecasting have introduced numerous complex supervised prediction models that consistently outperform previously published architectures. However, this rapid progression raises concerns regarding inconsistent benchmarking and reporting practices, which may undermine the reliability of these comparisons. In this study, we first perform a broad, thorough, and reproducible evaluation of the top-performing supervised models on the most popular benchmark and additional baselines representing the most active architecture families. This extensive evaluation assesses eight models on 14 datasets, encompassing $\sim$5,000 trained networks for the hyperparameter (HP) searches. Then, through a comprehensive analysis, we find that slight changes to experimental setups or current evaluation metrics drastically shift the common belief that newly published results are advancing the state of the art. Our findings emphasize the need to shift focus away from pursuing ever-more complex models, towards enhancing benchmarking practices through rigorous and standardized evaluations that enable more substantiated claims, including reproducible HP setups and statistical testing. We offer recommendations for future research.
Authors: Daniil Medyakov, Gleb Molodtsov, Savelii Chezhegov, Alexey Rebrikov, Aleksandr Beznosikov
Abstract: Stochastic optimization algorithms are widely used for machine learning with large-scale data. However, their convergence often suffers from non-vanishing variance. Variance Reduction (VR) methods, such as SVRG and SARAH, address this issue but introduce a bottleneck by requiring periodic full gradient computations. In this paper, we explore popular VR techniques and propose an approach that eliminates the necessity for expensive full gradient calculations. To avoid these computations and make our approach memory-efficient, we employ two key techniques: the shuffling heuristic and the concept of SAG/SAGA methods. For non-convex objectives, our convergence rates match those of standard shuffling methods, while under strong convexity, they demonstrate an improvement. We empirically validate the efficiency of our approach and demonstrate its scalability on large-scale machine learning tasks including image classification problem on CIFAR-10 and CIFAR-100 datasets.
Authors: Haoqi He, Yan Xiao, Wenzhi Xu, Ruoying Liu, Xiaokai Lin, Kai Wen
Abstract: Estimating the global Lipschitz constant of neural networks is crucial for understanding and improving their robustness and generalization capabilities. However, precise calculations are NP-hard, and current semidefinite programming (SDP) methods face challenges such as high memory usage and slow processing speeds. In this paper, we propose HiQ-Lip, a hybrid quantum-classical hierarchical method that leverages quantum computing to estimate the global Lipschitz constant. We tackle the estimation by converting it into a Quadratic Unconstrained Binary Optimization problem and implement a multilevel graph coarsening and refinement strategy to adapt to the constraints of contemporary quantum hardware. Our experimental evaluations on fully connected neural networks demonstrate that HiQ-Lip not only provides estimates comparable to state-of-the-art methods but also significantly accelerates the computation process. In specific tests involving two-layer neural networks with 256 hidden neurons, HiQ-Lip doubles the solving speed and offers more accurate upper bounds than the existing best method, LiPopt. These findings highlight the promising utility of small-scale quantum devices in advancing the estimation of neural network robustness.
Authors: Zihao Zheng, Xiuping Cui, Size Zheng, Maoliang Li, Jiayu Chen, Yun Liang, Xiang Chen
Abstract: As the Mix-of-Experts (MoE) architecture increases the number of parameters in large models, there is an even greater need for model quantization. However, existing quantization methods overlook the expert dynamics of MoE across multiple datasets. Moreover, the existing static quantization cannot adapt MoE to various data change scenarios. In this paper, we perform a multi-level analysis to reveal MoE dynamics and define the significance of each channel/each expert. Based on the analysis results, we propose \textit{DynaMo}, an end-to-end MoE quantization framework. DynaMo adopts an expert-level mixed-precision baseline quantization strategy, which ensures the quantized MoEs are compatible with multiple existing datasets. Furthermore, DynaMo incorporates a channel-level dynamic switching mechanism to adapt these quantized MoE models to novel datasets. Experiments show that DynaMo achieves a 2.78~4.54 PPL decrease and a 1.85%~3.77% accuracy improvement in various datasets, with ~3x inference speedup and negligible overhead.
Authors: Jakub Adamczyk, Jakub Poziemski, Pawel Siedlecki
Abstract: Small molecules play a critical role in the biomedical, environmental, and agrochemical domains, each with distinct physicochemical requirements and success criteria. Although biomedical research benefits from extensive datasets and established benchmarks, agrochemical data remain scarce, particularly with respect to species-specific toxicity. This work focuses on ApisTox, the most comprehensive dataset of experimentally validated chemical toxicity to the honey bee (\textit{Apis mellifera}), an ecologically vital pollinator. The primary goal of this study was to determine the suitability of diverse machine learning approaches for modeling such toxicity, including molecular fingerprints, graph kernels, and graph neural networks, as well as pretrained models. Comparative analysis with medicinal datasets from the MoleculeNet benchmark reveals that ApisTox represents a distinct chemical space. Performance degradation on non-medicinal datasets, such as \mbox{ApisTox}, demonstrates their limited generalizability of current state-of-the-art algorithms trained solely on biomedical data. Our study highlights the need for more diverse datasets and for targeted model development geared toward the agrochemical domain.
Authors: Hao Wu, Yuan Gao, Ruijian Gou, Xian Wu, Chuhan Wu, Huahui Yi, Johannes Brandstetter, Fan Xu, Kun Wang, Penghao Zhao, Hao Jia, Qi Song, Xinliang Liu, Juncai He, Shuhao Cao, Huanshuo Dong, Yanfei Xiang, Fan Zhang, Haixin Wang, Xingjian Shi, Qiufeng Wang, Shuaipeng Li, Ruobing Xie, Feng Tao, Yuxu Lu, Yu Guo, Yuntian Chen, Yuxuan Liang, Qingsong Wen, Wanli Ouyang, Deliang Chen, Niklas Boers, Xiaomeng Huang
Abstract: Reliable long-term forecasting of Earth system dynamics is fundamentally limited by instabilities in current artificial intelligence (AI) models during extended autoregressive simulations. These failures often originate from inherent spectral bias, leading to inadequate representation of critical high-frequency, small-scale processes and subsequent uncontrolled error amplification. Inspired by the nested grids in numerical models used to resolve small scales, we present TritonCast. At the core of its design is a dedicated latent dynamical core, which ensures the long-term stability of the macro-evolution at a coarse scale. An outer structure then fuses this stable trend with fine-grained local details. This design effectively mitigates the spectral bias caused by cross-scale interactions. In atmospheric science, it achieves state-of-the-art accuracy on the WeatherBench 2 benchmark while demonstrating exceptional long-term stability: executing year-long autoregressive global forecasts and completing multi-year climate simulations that span the entire available $2500$-day test period without drift. In oceanography, it extends skillful eddy forecast to $120$ days and exhibits unprecedented zero-shot cross-resolution generalization. Ablation studies reveal that this performance stems from the synergistic interplay of the architecture's core components. TritonCast thus offers a promising pathway towards a new generation of trustworthy, AI-driven simulations. This significant advance has the potential to accelerate discovery in climate and Earth system science, enabling more reliable long-term forecasting and deeper insights into complex geophysical dynamics.
Authors: Parsa Mirtaheri, Ezra Edelman, Samy Jelassi, Eran Malach, Enric Boix-Adsera
Abstract: Inference-time computation has emerged as a promising scaling axis for improving large language model reasoning. However, despite yielding impressive performance, the optimal allocation of inference-time computation remains poorly understood. A central question is whether to prioritize sequential scaling (e.g., longer chains of thought) or parallel scaling (e.g., majority voting across multiple short chains of thought). In this work, we seek to illuminate the landscape of test-time scaling by demonstrating the existence of reasoning settings where sequential scaling offers an exponential advantage over parallel scaling. These settings are based on graph connectivity problems in challenging distributions of graphs. We validate our theoretical findings with comprehensive experiments across a range of language models, including models trained from scratch for graph connectivity with different chain of thought strategies as well as large reasoning models.
Authors: Sanna Jarl, Jens Sj\"olund, Robert J. W. Frost, Anders Holst, Jonathan J. S. Scragg
Abstract: Self-driving labs (SDLs), employing automation and machine learning (ML) to accelerate experimental procedures, have enormous potential in the discovery of new materials. However, in thin film science, SDLs are mainly restricted to solution-based synthetic methods which are easier to automate but cannot access the broad chemical space of inorganic materials. This work presents an SDL based on magnetron co-sputtering. We are using combinatorial frameworks, obtaining accurate composition maps on multi-element, compositionally graded thin films. This normally requires time-consuming ex-situ analysis prone to systematic errors. We present a rapid and calibration-free in-situ, ML driven approach to produce composition maps for arbitrary source combinations and sputtering conditions. We develop a method to predict the composition distribution in a multi-element combinatorial thin film, using in-situ measurements from quartz-crystal microbalance sensors placed in a sputter chamber. For a given source, the sensor readings are learned as a function of the sputtering pressure and magnetron power, through active learning using Gaussian processes (GPs). The final GPs are combined with a geometric model of the deposition flux distribution in the chamber, which allows interpolation of the deposition rates from each source, at any position across the sample. We investigate several acquisition functions for the ML procedure. A fully Bayesian GP - BALM (Bayesian active learning MacKay) - achieved the best performance, learning the deposition rates for a single source in 10 experiments. Prediction accuracy for co-sputtering composition distributions was verified experimentally. Our framework dramatically increases throughput by avoiding the need for extensive characterisation or calibration, thus demonstrating the potential of ML-guided SDLs to accelerate materials exploration.
Authors: Siva Rajesh Kasa, Karan Gupta, Sumegh Roychowdhury, Ashutosh Kumar, Yaswanth Biruduraju, Santhosh Kumar Kasa, Nikhil Priyatam Pattisapu, Arindam Bhattacharya, Shailendra Agarwal, Vijay huddar
Abstract: The comparison between discriminative and generative classifiers has intrigued researchers since Efron's seminal analysis of logistic regression versus discriminant analysis. While early theoretical work established that generative classifiers exhibit lower sample complexity but higher asymptotic error in simple linear settings, these trade-offs remain unexplored in the transformer era. We present the first comprehensive evaluation of modern generative and discriminative architectures - Auto-regressive modeling, Masked Language Modeling, Discrete Diffusion, and Encoders for text classification. Our study reveals that the classical 'two regimes' phenomenon manifests distinctly across different architectures and training paradigms. Beyond accuracy, we analyze sample efficiency, calibration, noise robustness, and ordinality across diverse scenarios. Our findings offer practical guidance for selecting the most suitable modeling approach based on real-world constraints such as latency and data limitations.
Authors: Ray Zirui Zhang, Christopher E. Miles, Xiaohui Xie, John S. Lowengrub
Abstract: Uncertainty quantification in PDE inverse problems is essential in many applications. Scientific machine learning and AI enable data-driven learning of model components while preserving physical structure, and provide the scalability and adaptability needed for emerging imaging technologies and clinical insights. We develop a Bilevel Local Operator Learning framework for Bayesian inference in PDEs (B-BiLO). At the upper level, we sample parameters from the posterior via Hamiltonian Monte Carlo, while at the lower level we fine-tune a neural network via low-rank adaptation (LoRA) to approximate the solution operator locally. B-BiLO enables efficient gradient-based sampling without synthetic data or adjoint equations and avoids sampling in high-dimensional weight space, as in Bayesian neural networks, by optimizing weights deterministically. We analyze errors from approximate lower-level optimization and establish their impact on posterior accuracy. Numerical experiments across PDE models, including tumor growth, demonstrate that B-BiLO achieves accurate and efficient uncertainty quantification.
Authors: Saad Mokssit, Ouassim Karrakchou, Alejandro Mousist, Mounir Ghogho
Abstract: Early-exit neural networks reduce inference cost by enabling confident predictions at intermediate layers. However, joint training often leads to gradient interference, with deeper classifiers dominating optimization. We propose Confidence-Gated Training (CGT), a paradigm that conditionally propagates gradients from deeper exits only when preceding exits fail. This encourages shallow classifiers to act as primary decision points while reserving deeper layers for harder inputs. By aligning training with the inference-time policy, CGT mitigates overthinking, improves early-exit accuracy, and preserves efficiency. Experiments on the Indian Pines and Fashion-MNIST benchmarks show that CGT lowers average inference cost while improving overall accuracy, offering a practical solution for deploying deep models in resource-constrained environments.
Authors: Yuki Ichihara, Yuu Jinnai, Tetsuro Morimura, Mitsuki Sakamoto, Ryota Mitsuhashi, Eiji Uchibe
Abstract: Group Relative Policy Optimization (GRPO) has been shown to be an effective algorithm when an accurate reward model is available. However, such a highly reliable reward model is not available in many real-world tasks. In this paper, we particularly focus on multi-objective settings, in which we identify that GRPO is vulnerable to reward hacking, optimizing only one of the objectives at the cost of the others. To address this issue, we propose MO-GRPO, an extension of GRPO with a simple normalization method to reweight the reward functions automatically according to the variances of their values. We first show analytically that MO-GRPO ensures that all reward functions contribute evenly to the loss function while preserving the order of preferences, eliminating the need for manual tuning of the reward functions' scales. Then, we evaluate MO-GRPO experimentally in four domains: (i) the multi-armed bandits problem, (ii) simulated control task (Mo-Gymnasium), (iii) machine translation tasks on the WMT benchmark (En-Ja, En-Zh), and (iv) instruction following task. MO-GRPO achieves stable learning by evenly distributing correlations among the components of rewards, outperforming GRPO, showing MO-GRPO to be a promising algorithm for multi-objective reinforcement learning problems.
Authors: Bingshuai Liu, Ante Wang, Zijun Min, Liang Yao, Haibo Zhang, Yang Liu, Anxiang Zeng, Jinsong Su
Abstract: Large Language Models (LLMs) increasingly rely on reinforcement learning with verifiable rewards (RLVR) to elicit reliable chain-of-thought reasoning. However, the training process remains bottlenecked by the computationally expensive rollout stage. Existing acceleration methods-such as parallelization, objective- and data-driven modifications, and replay buffers-either incur diminishing returns, introduce bias, or overlook redundancy across iterations. We identify that rollouts from consecutive training epochs frequently share a large portion of overlapping segments, wasting computation. To address this, we propose SPEC-RL, a novel framework that integrates SPECulative decoding with the RL rollout process. SPEC-RL reuses prior trajectory segments as speculative prefixes and extends them via a draft-and-verify mechanism, avoiding redundant generation while ensuring policy consistency. Experiments on diverse math reasoning and generalization benchmarks, including AIME24, MATH-500, OlympiadBench, MMLU-STEM, and others, demonstrate that SPEC-RL reduces rollout time by 2-3x without compromising policy quality. As a purely rollout-stage enhancement, SPEC-RL integrates seamlessly with mainstream algorithms (e.g., PPO, GRPO, DAPO), offering a general and practical path to scale RLVR for large reasoning models. Our code is available at https://github.com/ShopeeLLM/Spec-RL
Authors: Ali Dadsetan, Frank Rudzicz
Abstract: We address the challenge of sample efficiency in differentially private fine-tuning of large language models (LLMs) using DP-SGD. While DP-SGD provides strong privacy guarantees, the added noise significantly increases the entropy of gradient matrices, disrupting their low-rank structure and slowing optimization. We propose a post-processing algorithm that leverages random matrix theory to denoise gradients, restore low-rank structure, and improve alignment with the original signal. Applied to DP-SGD fine-tuning of RoBERTa on GLUE tasks, our method improves sample efficiency compared to state-of-the-art approaches, substantially reducing training time when optimal performance is not required. This work demonstrates that matrix recovery techniques can enhance the utility of private language model training without compromising privacy guarantees.
Authors: Vicente Bosca, Tatum Rask, Sunia Tanweer, Andrew R. Tawfeek, Branden Stone
Abstract: This paper explores the topological signatures of ReLU neural network activation patterns. We consider feedforward neural networks with ReLU activation functions and analyze the polytope decomposition of the feature space induced by the network. Mainly, we investigate how the Fiedler partition of the dual graph and show that it appears to correlate with the decision boundary -- in the case of binary classification. Additionally, we compute the homology of the cellular decomposition -- in a regression task -- to draw similar patterns in behavior between the training loss and polyhedral cell-count, as the model is trained.
Authors: Jyler Menard, R. A. Mansbach
Abstract: Antimicrobial peptides (AMPs) are a promising class of therapeutics to treat bacterial infections. Discovering and designing such peptides is difficult because of the vast number of possible sequences of amino acids. Deep generative models, such as variational autoencoders, have shown value in peptide design due to their ability to model sequence space with a continuous-valued latent space. Although such models have already been used to great effect in biomolecular design, they still suffer from a lack of interpretability and rigorous quantification of latent space quality as a search space. We investigate (1) whether searching through a dimensionally-reduced variant of the latent design space may facilitate optimization, (2) how organizing latent spaces with physicochemical properties may improve the efficiency of optimizing antimicrobial activity, and (3) the interpretability of the spaces. We find that employing a dimensionally-reduced version of the latent space is more interpretable and can be advantageous, while we can organize the latent space with different physicochemical properties even at different percentages of available labels. This work lays crucial groundwork for biophysically-motivated peptide design procedures.
Authors: Tobias Fuchs, Nadja Klein
Abstract: Real-world data is frequently noisy and ambiguous. In crowdsourcing, for example, human annotators may assign conflicting class labels to the same instances. Partial-label learning (PLL) addresses this challenge by training classifiers when each instance is associated with a set of candidate labels, only one of which is correct. While early PLL methods approximate the true label posterior, they are often computationally intensive. Recent deep learning approaches improve scalability but rely on surrogate losses and heuristic label refinement. We introduce a novel probabilistic framework that directly approximates the posterior distribution over true labels using amortized variational inference. Our method employs neural networks to predict variational parameters from input data, enabling efficient inference. This approach combines the expressiveness of deep learning with the rigor of probabilistic modeling, while remaining architecture-agnostic. Theoretical analysis and extensive experiments on synthetic and real-world datasets demonstrate that our method achieves state-of-the-art performance in both accuracy and efficiency.
Authors: Han Wu, Jie Yin
Abstract: Few-shot knowledge graph relational learning seeks to perform reasoning over relations given only a limited number of training examples. While existing approaches largely adopt a meta-learning framework for enabling fast adaptation to new relations, they suffer from two key pitfalls. First, they learn relation meta-knowledge in isolation, failing to capture common relational patterns shared across tasks. Second, they struggle to effectively incorporate local, task-specific contexts crucial for rapid adaptation. To address these limitations, we propose MoEMeta, a novel meta-learning framework that disentangles globally shared knowledge from task-specific contexts to enable both effective model generalization and rapid adaptation. MoEMeta introduces two key innovations: (i) a mixture-of-experts (MoE) model that learns globally shared relational prototypes to enhance generalization, and (ii) a task-tailored adaptation mechanism that captures local contexts for fast task-specific adaptation. By balancing global generalization with local adaptability, MoEMeta significantly advances few-shot relational learning. Extensive experiments and analyses on three KG benchmarks show that MoEMeta consistently outperforms existing baselines, achieving state-of-the-art performance.
Authors: Abdelaziz Bounhar, Hadi Abdine, Evan Dufraisse, Ahmad Chamma, Amr Mohamed, Dani Bouch, Michalis Vazirgiannis, Guokan Shang
Abstract: Large language models (LLMs) trained for step-by-step reasoning often become excessively verbose, raising inference cost. Standard Reinforcement Learning with Verifiable Rewards (RLVR) pipelines filter out ``easy'' problems for training efficiency, leaving the model to train primarily on harder problems that require longer reasoning chains. This skews the output length distribution upward, resulting in a \textbf{model that conflates ``thinking longer'' with ``thinking better''}. In this work, we show that retaining and modestly up-weighting moderately easy problems acts as an implicit length regularizer. Exposing the model to solvable short-chain tasks constrains its output distribution and prevents runaway verbosity. The result is \textbf{\emph{emergent brevity for free}}: the model learns to solve harder problems without inflating the output length, \textbf{ despite the absence of any explicit length penalization}. RLVR experiments using this approach on \textit{Qwen3-4B-Thinking-2507} (with a 16k token limit) achieve baseline pass@1 AIME25 accuracy while generating solutions that are, on average, nearly twice as short. The code is available at \href{https://github.com/MBZUAI-Paris/Frugal-AI}{GitHub}, with datasets and models on \href{https://huggingface.co/collections/MBZUAI-Paris/k2-think-mini-68dcfa8b114686a4bd3dc2bc}{Hugging Face}.
URLs: https://github.com/MBZUAI-Paris/Frugal-AI, https://huggingface.co/collections/MBZUAI-Paris/k2-think-mini-68dcfa8b114686a4bd3dc2bc
Authors: Tiberiu Musat
Abstract: Grokking is a puzzling phenomenon in neural networks where full generalization occurs only after a substantial delay following the complete memorization of the training data. Previous research has linked this delayed generalization to representation learning driven by weight decay, but the precise underlying dynamics remain elusive. In this paper, we argue that post-memorization learning can be understood through the lens of constrained optimization: gradient descent effectively minimizes the weight norm on the zero-loss manifold. We formally prove this in the limit of infinitesimally small learning rates and weight decay coefficients. To further dissect this regime, we introduce an approximation that decouples the learning dynamics of a subset of parameters from the rest of the network. Applying this framework, we derive a closed-form expression for the post-memorization dynamics of the first layer in a two-layer network. Experiments confirm that simulating the training process using our predicted gradients reproduces both the delayed generalization and representation learning characteristic of grokking.
Authors: Tingyang Wei, Jiao Liu, Abhishek Gupta, Chin Chun Ooi, Puay Siew Tan, Yew-Soon Ong
Abstract: Many real-world applications require solving families of expensive multi-objective optimization problems~(EMOPs) under varying operational conditions. This can be formulated as parametric expensive multi-objective optimization problems (P-EMOPs) where each task parameter defines a distinct optimization instance. Current multi-objective Bayesian optimization methods have been widely used for finding finite sets of Pareto optimal solutions for individual tasks. However, P-EMOPs present a fundamental challenge: the continuous task parameter space can contain infinite distinct problems, each requiring separate expensive evaluations. This demands learning an inverse model that can directly predict optimized solutions for any task-preference query without expensive re-evaluation. This paper introduces a novel parametric multi-objective Bayesian optimizer that learns this inverse model by alternating between (1) generative solution sampling via conditional generative models and (2) acquisition-driven search leveraging inter-task synergies. This approach enables effective optimization across multiple tasks and finally achieves direct solution prediction for unseen parameterized EMOPs without additional expensive evaluations. We theoretically justify the faster convergence by leveraging inter-task synergies through task-aware Gaussian processes. Based on that, empirical studies in synthetic and real-world benchmarks further verify the effectiveness of the proposed parametric optimizer.
Authors: Xiang Ma, Taihua Chen, Pengcheng Wang, Xuemei Li, Caiming Zhang
Abstract: Time series forecasting is crucial for applications in various domains. Conventional methods often rely on global decomposition into trend, seasonal, and residual components, which become ineffective for real-world series dominated by local, complex, and highly dynamic patterns. Moreover, the high model complexity of such approaches limits their applicability in real-time or resource-constrained environments. In this work, we propose a novel \textbf{RE}liability-aware \textbf{C}odebook-\textbf{AS}sisted \textbf{T}ime series forecasting framework (\textbf{ReCast}) that enables lightweight and robust prediction by exploiting recurring local shapes. ReCast encodes local patterns into discrete embeddings through patch-wise quantization using a learnable codebook, thereby compactly capturing stable regular structures. To compensate for residual variations not preserved by quantization, ReCast employs a dual-path architecture comprising a quantization path for efficient modeling of regular structures and a residual path for reconstructing irregular fluctuations. A central contribution of ReCast is a reliability-aware codebook update strategy, which incrementally refines the codebook via weighted corrections. These correction weights are derived by fusing multiple reliability factors from complementary perspectives by a distributionally robust optimization (DRO) scheme, ensuring adaptability to non-stationarity and robustness to distribution shifts. Extensive experiments demonstrate that ReCast outperforms state-of-the-art (SOTA) models in accuracy, efficiency, and adaptability to distribution shifts.
Authors: Jiawei Shen, Jia Zhu, Hanghui Guo, Weijie Shi, Guoqing Ma, Yidan Liang, Jingjiang Liu, Hao Chen, Shimin Di
Abstract: Temporal Knowledge Graph Reasoning (TKGR) aims to complete missing factual elements along the timeline. Depending on the temporal position of the query, the task is categorized into interpolation and extrapolation. Existing interpolation methods typically embed temporal information into individual facts to complete missing historical knowledge, while extrapolation techniques often leverage sequence models over graph snapshots to identify recurring patterns for future event prediction. These methods face two critical challenges: limited contextual modeling in interpolation and cognitive generalization bias in extrapolation. To address these, we propose a unified method for TKGR, dubbed DynaGen. For interpolation, DynaGen dynamically constructs entity-centric subgraphs and processes them with a synergistic dual-branch GNN encoder to capture evolving structural context. For extrapolation, it applies a conditional diffusion process, which forces the model to learn underlying evolutionary principles rather than just superficial patterns, enhancing its ability to predict unseen future events. Extensive experiments on six benchmark datasets show DynaGen achieves state-of-the-art performance. On average, compared to the second-best models, DynaGen improves the Mean Reciprocal Rank (MRR) score by 2.61 points for interpolation and 1.45 points for extrapolation.
Authors: Anastasiia Alokhina, Pan Li
Abstract: Transformers exhibit a notable property of \emph{size generalization}, demonstrating an ability to extrapolate from smaller token sets to significantly longer ones. This behavior has been documented across diverse applications, including point clouds, graphs, and natural language. Despite its empirical success, this capability still lacks some rigorous theoretical characterizations. In this paper, we develop a theoretical framework to analyze this phenomenon for geometric data, which we represent as discrete samples from a continuous source (e.g., point clouds from manifolds, graphs from graphons). Our core contribution is a bound on the error between the Transformer's output for a discrete sample and its continuous-domain equivalent. We prove that for Transformers with stable positional encodings, this bound is determined by the sampling density and the intrinsic dimensionality of the data manifold. Experiments on graphs and point clouds of various sizes confirm the tightness of our theoretical bound.
Authors: Jakob De Moor, Hans Weytjens, Johannes De Smedt, Jochen De Weerdt
Abstract: Prescriptive Process Monitoring (PresPM) recommends interventions during business processes to optimize key performance indicators (KPIs). In realistic settings, interventions are rarely isolated: organizations need to align sequences of interventions to jointly steer the outcome of a case. Existing PresPM approaches fall short in this respect. Many focus on a single intervention decision, while others treat multiple interventions independently, ignoring how they interact over time. Methods that do address these dependencies depend either on simulation or data augmentation to approximate the process to train a Reinforcement Learning (RL) agent, which can create a reality gap and introduce bias. We introduce SCOPE, a PresPM approach that learns aligned sequential intervention recommendations. SCOPE employs backward induction to estimate the effect of each candidate intervention action, propagating its impact from the final decision point back to the first. By leveraging causal learners, our method can utilize observational data directly, unlike methods that require constructing process approximations for reinforcement learning. Experiments on both an existing synthetic dataset and a new semi-synthetic dataset show that SCOPE consistently outperforms state-of-the-art PresPM techniques in optimizing the KPI. The novel semi-synthetic setup, based on a real-life event log, is provided as a reusable benchmark for future work on sequential PresPM.
Authors: Xueming Yan, Boyan Xu, Yaochu Jin, Lixian Xiao, Wenlong Ye, Runyang Cai, Zeqi Zheng, Jingfa Liu, Aimin Yang, Yongduan Song
Abstract: Indonesian, spoken by over 200 million people, remains underserved in multimodal emotion recognition research despite its dominant presence on Southeast Asian social media platforms. We introduce IndoMER, the first multimodal emotion recognition benchmark for Indonesian, comprising 1,944 video segments from 203 speakers with temporally aligned text, audio, and visual annotations across seven emotion categories. The dataset exhibits realistic challenges including cross-modal inconsistency and long-tailed class distributions shaped by Indonesian cultural communication norms. To address these challenges, we propose OmniMER, a multimodal adaptation framework built upon Qwen2.5-Omni that enhances emotion recognition through three auxiliary modality-specific perception tasks: emotion keyword extraction for text, facial expression analysis for video, and prosody analysis for audio. These auxiliary tasks help the model identify emotion-relevant cues in each modality before fusion, reducing reliance on spurious correlations in low-resource settings. Experiments on IndoMER show that OmniMER achieves 0.582 Macro-F1 on sentiment classification and 0.454 on emotion recognition, outperforming the base model by 7.6 and 22.1 absolute points respectively. Cross-lingual evaluation on the Chinese CH-SIMS dataset further demonstrates the generalizability of the proposed framework. The dataset and code are publicly available. https://github.com/yanxm01/INDOMER
Authors: Jorge Sastre, Daniel Faronbi, Jos\'e Miguel Alonso, Peter Traver, Javier Ib\'a\~nez, Nuria Lloret
Abstract: The matrix exponential is a fundamental operator in scientific computing and system simulation, with applications ranging from control theory and quantum mechanics to modern generative machine learning. While Pad\'e approximants combined with scaling and squaring have long served as the standard, recent Taylor-based methods, which utilize polynomial evaluation schemes that surpass the classical Paterson--Stockmeyer technique, offer superior accuracy and reduced computational complexity. This paper presents an optimized Taylor-based algorithm for the matrix exponential, specifically designed for the high-throughput requirements of generative AI flows. We provide a rigorous error analysis and develop a dynamic selection strategy for the Taylor order and scaling factor to minimize computational effort under a prescribed error tolerance. Extensive numerical experiments demonstrate that our approach provides significant acceleration and maintains high numerical stability compared to existing state-of-the-art implementations. These results establish the proposed method as a highly efficient tool for large-scale generative modeling.
Authors: Qianyi Chen, Bo Li
Abstract: While conformal prediction provides robust marginal coverage guarantees, achieving reliable conditional coverage for specific inputs remains challenging. Although exact distribution-free conditional coverage is impossible with finite samples, recent work has focused on improving the conditional coverage of standard conformal procedures. Distinct from approaches that target relaxed notions of conditional coverage, we directly minimize the mean squared error of conditional coverage by refining the quantile regression components that underpin many conformal methods. Leveraging a Taylor expansion, we derive a sharp surrogate objective for quantile regression: a density-weighted pinball loss, where the weights are given by the conditional density of the conformity score evaluated at the true quantile. We propose a three-headed quantile network that estimates these weights via finite differences using auxiliary quantile levels at \(1-\alpha \pm \delta\), subsequently fine-tuning the central quantile by optimizing the weighted loss. We provide a theoretical analysis with exact non-asymptotic guarantees characterizing the resulting excess risk. Extensive experiments on diverse high-dimensional real-world datasets demonstrate remarkable improvements in conditional coverage performance.
Authors: Yu Li, Sizhe Tang, Rongqian Chen, Fei Xu Yu, Guangyu Jiang, Mahdi Imani, Nathaniel D. Bastian, Tian Lan
Abstract: Automated cyber defense (ACD) seeks to protect computer networks with minimal or no human intervention, reacting to intrusions by taking corrective actions such as isolating hosts, resetting services, deploying decoys, or updating access controls. However, existing approaches for ACD, such as deep reinforcement learning (RL), often face difficult exploration in complex networks with large decision/state spaces and thus require an expensive amount of samples. Inspired by the need to learn sample-efficient defense policies, we frame ACD in CAGE Challenge 4 (CAGE-4 / CC4) as a context-based partially observable Markov decision problem and propose a planning-centric defense policy based on Monte Carlo Tree Search (MCTS). It explicitly models the exploration-exploitation tradeoff in ACD and uses statistical sampling to guide exploration and decision making. We make novel use of graph neural networks (GNNs) to embed observations from the network as attributed graphs, to enable permutation-invariant reasoning over hosts and their relationships. To make our solution practical in complex search spaces, we guide MCTS with learned graph embeddings and priors over graph-edit actions, combining model-free generalization and policy distillation with look-ahead planning. We evaluate the resulting agent on CC4 scenarios involving diverse network structures and adversary behaviors, and show that our search-guided, graph-embedding-based planning improves defense reward and robustness relative to state-of-the-art RL baselines.
Authors: Linfeng Ye, Zhixiang Chi, Konstantinos N. Plataniotis, En-hui Yang
Abstract: In this paper, we propose a novel information theoretic surrogate loss; normalized conditional mutual information (NCMI); as a drop in alternative to the de facto cross-entropy (CE) for training deep neural network (DNN) based classifiers. We first observe that the model's NCMI is inversely proportional to its accuracy. Building on this insight, we introduce an alternating algorithm to efficiently minimize the NCMI. Across image recognition and whole-slide imaging (WSI) subtyping benchmarks, NCMI-trained models surpass state of the art losses by substantial margins at a computational cost comparable to that of CE. Notably, on ImageNet, NCMI yields a 2.77% top-1 accuracy improvement with ResNet-50 comparing to the CE; on CAMELYON-17, replacing CE with NCMI improves the macro-F1 by 8.6% over the strongest baseline. Gains are consistent across various architectures and batch sizes, suggesting that NCMI is a practical and competitive alternative to CE.
Authors: Longwen Wang, Xuan'er Wu, Xiaohui Hu, Yirui Liu, Yuankai Fan, Kaidong Yu, Qizhen Weng, Wei Xi, Xuelong Li
Abstract: Effective reward design is a central challenge in Reinforcement Learning (RL) for code generation. Mainstream pass/fail outcome rewards enforce functional correctness via executing unit tests, but the resulting sparsity limits potential performance gains. While recent work has explored external Reward Models (RM) to generate richer, continuous rewards, the learned RMs suffer from reward misalignment and prohibitive computational cost. In this paper, we introduce \textbf{VeRPO} (\textbf{V}erifiable D\textbf{e}nse \textbf{R}eward \textbf{P}olicy \textbf{O}ptimization), a novel RL framework for code generation that synthesizes \textit{robust and dense rewards fully grounded in verifiable execution feedback}. The core idea of VeRPO is constructing dense rewards from weighted partial success: by dynamically estimating the difficulty weight of each unit test based on the execution statistics during training, a dense reward is derived from the sum of weights of the passed unit tests. To solidify the consistency between partial success and end-to-end functional correctness, VeRPO further integrates the dense signal with global execution outcomes, establishing a robust and dense reward paradigm relying solely on verifiable execution feedback. Extensive experiments across diverse benchmarks and settings demonstrate that VeRPO consistently outperforms outcome-driven and RM-based baselines, achieving up to +8.83\% gain in pass@1 with negligible time cost (< 0.02\%) and zero GPU memory overhead.
Authors: Sifan Yang, Wenhao Yang, Wei Jiang, Lijun Zhang
Abstract: We investigate distributed online convex optimization with compressed communication, where $n$ learners connected by a network collaboratively minimize a sequence of global loss functions using only local information and compressed data from neighbors. Prior work has established regret bounds of $O(\max\{\omega^{-2}\rho^{-4}n^{1/2},\omega^{-4}\rho^{-8}\}n\sqrt{T})$ and $O(\max\{\omega^{-2}\rho^{-4}n^{1/2},\omega^{-4}\rho^{-8}\}n\ln{T})$ for convex and strongly convex functions, respectively, where $\omega\in(0,1]$ is the compression quality factor ($\omega=1$ means no compression) and $\rho<1$ is the spectral gap of the communication matrix. However, these regret bounds suffer from a quadratic or even quartic dependence on $\omega^{-1}$. Moreover, the super-linear dependence on $n$ is also undesirable. To overcome these limitations, we propose a novel algorithm that achieves improved regret bounds of $\tilde{O}(\omega^{-1/2}\rho^{-1}n\sqrt{T})$ and $\tilde{O}(\omega^{-1}\rho^{-2}n\ln{T})$ for convex and strongly convex functions, respectively. The primary idea is to design a two-level blocking update framework incorporating two novel ingredients: an online gossip strategy and an error compensation scheme, which collaborate to achieve a better consensus among learners. Furthermore, we establish the first lower bounds for this problem, justifying the optimality of our results with respect to both $\omega$ and $T$. Additionally, we consider the bandit feedback scenario, and extend our method with the classic gradient estimators to enhance existing regret bounds.
Authors: Yihan Zhang, Marco Mondelli, Ramji Venkataramanan
Abstract: In a mixed generalized linear model, the goal is to learn multiple signals from unlabeled observations: each sample comes from exactly one signal, but it is not known which one. We consider the prototypical problem of estimating two statistically independent signals in a mixed generalized linear model with Gaussian covariates. Spectral methods are a popular class of estimators which output the top two eigenvectors of a suitable data-dependent matrix. However, despite the wide applicability, their design is still obtained via heuristic considerations, and the number of samples $n$ needed to guarantee recovery is super-linear in the signal dimension $d$. In this paper, we develop exact asymptotics on spectral methods in the challenging proportional regime in which $n, d$ grow large and their ratio converges to a finite constant. This allows us optimize the design of the spectral method, and combine it with a simple linear estimator, to minimize the estimation error. Our characterization exploits a mix of tools from random matrices, free probability and the theory of approximate message passing algorithms. Numerical simulations for mixed linear regression and phase retrieval demonstrate the advantage enabled by our analysis over existing designs of spectral methods.
Authors: Adrian de Wynter, Xun Wang, Alex Sokolov, Qilong Gu, Si-Qing Chen
Abstract: We present an empirical evaluation of various outputs generated by nine of the most widely-available large language models (LLMs). Our analysis is done with off-the-shelf, readily-available tools. We find a correlation between percentage of memorized text, percentage of unique text, and overall output quality, when measured with respect to output pathologies such as counterfactual and logically-flawed statements, and general failures like not staying on topic. Overall, 80.0% of the outputs evaluated contained memorized data, but outputs containing the most memorized content were also more likely to be considered of high quality. We discuss and evaluate mitigation strategies, showing that, in the models evaluated, the rate of memorized text being output is reduced. We conclude with a discussion on potential implications around what it means to learn, to memorize, and to evaluate quality text.
Authors: Minqi Shao, Jianjun Zhao
Abstract: Quantum Neural Networks (QNNs) have achieved initial success in various tasks by integrating quantum computing and neural networks. However, growing concerns about their reliability and robustness highlight the need for systematic testing. Unfortunately, current testing methods for QNNs remain underdeveloped, with limited practical utility and insufficient empirical evaluation. As an initial effort, we design a set of superposition-targeted coverage criteria to evaluate QNN state exploration embedded in test suites. To characterize the effectiveness, scalability, and robustness of the criteria, we conduct a comprehensive empirical study using benchmark datasets and QNN architectures. We first evaluate their sensitivity to input diversity under multiple data settings, and analyze their correlation with the number of injected faults. We then assess their scalability to increasing circuit scales. The robustness is further studied under practical quantum constraints including insufficient measurement and quantum noise. The results demonstrate the effectiveness of quantifying test adequacy and the potential applicability to larger-scale circuits and realistic quantum execution, while also revealing some limitations. Finally, we provide insights and recommendations for future QNN testing.
Authors: Linhao Huang, Xue Jiang, Zhiqiang Wang, Wentao Mo, Xi Xiao, Yong-Jie Yin, Bo Han, Feng Zheng
Abstract: Video-based multimodal large language models (V-MLLMs) have shown vulnerability to adversarial examples in video-text multimodal tasks. However, the transferability of adversarial videos to unseen models - a common and practical real-world scenario - remains unexplored. In this paper, we pioneer an investigation into the transferability of adversarial video samples across V-MLLMs. We find that existing adversarial attack methods face significant limitations when applied in black-box settings for V-MLLMs, which we attribute to the following shortcomings: (1) lacking generalization in perturbing video features, (2) focusing only on sparse key-frames, and (3) failing to integrate multimodal information. To address these limitations and deepen the understanding of V-MLLM vulnerabilities in black-box scenarios, we introduce the Image-to-Video MLLM (I2V-MLLM) attack. In I2V-MLLM, we utilize an image-based multimodal large language model (I-MLLM) as a surrogate model to craft adversarial video samples. Multimodal interactions and spatiotemporal information are integrated to disrupt video representations within the latent space, improving adversarial transferability. Additionally, a perturbation propagation technique is introduced to handle different unknown frame sampling strategies. Experimental results demonstrate that our method can generate adversarial examples that exhibit strong transferability across different V-MLLMs on multiple video-text multimodal tasks. Compared to white-box attacks on these models, our black-box attacks (using BLIP-2 as a surrogate model) achieve competitive performance, with average attack success rate (AASR) of 57.98% on MSVD-QA and 58.26% on MSRVTT-QA for Zero-Shot VideoQA tasks, respectively.
Authors: Yiyou Sun, Georgia Zhou, Haoyue Bai, Hao Wang, Dacheng Li, Nouha Dziri, Dawn Song
Abstract: Recent supervised fine-tuning (SFT) approaches have significantly improved language models' performance on mathematical reasoning tasks, even when models are trained at a small scale. However, the specific capabilities enhanced through such fine-tuning remain poorly understood. In this paper, we conduct a detailed analysis of model performance on the AIME24 dataset to understand how reasoning capabilities evolve. We discover a ladder-like structure in problem difficulty, categorize questions into four tiers (Easy, Medium, Hard, and Extremely Hard (Exh)), and identify the specific requirements for advancing between tiers. We find that progression from Easy to Medium tier requires adopting an R1 reasoning style with minimal SFT (500-1K instances), while Hard-level questions suffer from frequent model's errors at each step of the reasoning chain, with accuracy plateauing at around 65% despite logarithmic scaling. Exh-level questions present a fundamentally different challenge; they require unconventional problem-solving skills that current models uniformly struggle with. Additional findings reveal that carefully curated small-scale datasets offer limited advantage-scaling dataset size proves far more effective. Our analysis provides a clearer roadmap for advancing language model capabilities in mathematical reasoning.
Authors: Shubham Vaishnav, Praveen Kumar Donta, Sindri Magn\'usson
Abstract: IoT networks often face conflicting routing goals such as maximizing packet delivery, minimizing delay, and conserving limited battery energy. These priorities can also change dynamically: for example, an emergency alert requires high reliability, while routine monitoring prioritizes energy efficiency to prolong network lifetime. Existing works, including many deep reinforcement learning approaches, are typically centralized and assume static objectives, making them slow to adapt when preferences shift. We propose a dynamic and fully distributed multi-objective Q-learning routing algorithm that learns multiple per-preference Q-tables in parallel and introduces a novel greedy interpolation policy to act near-optimally for unseen preferences without retraining or central coordination. A theoretical analysis further shows that the optimal value function is Lipschitz-continuous in the preference parameter, ensuring that the proposed greedy interpolation policy yields provably near-optimal behavior. Simulations show that our approach adapts in real time to shifting priorities and achieves up to 80-90\% lower energy consumption and more than 2-5x higher cumulative rewards and packet delivery compared to six baseline protocols, under dynamic and distributed settings. Sensitivity analysis across varying preference window lengths confirms that the proposed DPQ framework consistently achieves higher composite reward than all baseline methods, demonstrating robustness to changes in operating conditions.
Authors: Ali Rashidi Moghadam, Sayedmohammadreza Rastegari, Mehdi Tale Masouleh, Ahmad Kalhor
Abstract: Grasp pose detection in cluttered, real-world environments remains a significant challenge due to noisy and incomplete sensory data combined with complex object geometries. This paper introduces Grasp the Graph 2.0 (GtG 2.0) method, a lightweight yet highly effective hypothesis-and-test robotics grasping framework which leverages an ensemble of Graph Neural Networks for efficient geometric reasoning from point cloud data. Building on the success of GtG 1.0, which demonstrated the potential of Graph Neural Networks for grasp detection but was limited by assumptions of complete, noise-free point clouds and 4-Dof grasping, GtG 2.0 employs a conventional Grasp Pose Generator to efficiently produce 7-Dof grasp candidates. Candidates are assessed with an ensemble Graph Neural Network model which includes points within the gripper jaws (inside points) and surrounding contextual points (outside points). This improved representation boosts grasp detection performance over previous methods using the same generator. GtG 2.0 shows up to a 35% improvement in Average Precision on the GraspNet-1Billion benchmark compared to hypothesis-and-test and Graph Neural Network-based methods, ranking it among the top three frameworks. Experiments with a 3-Dof Delta Parallel robot and Kinect-v1 camera show a success rate of 91% and a clutter completion rate of 100%, demonstrating its flexibility and reliability.
Authors: Hafez Ghaemi, Eilif Muller, Shahab Bakhtiari
Abstract: Joint-embedding self-supervised learning (SSL) commonly relies on transformations such as data augmentation and masking to learn visual representations, a task achieved by enforcing invariance or equivariance with respect to these transformations applied to two views of an image. This dominant two-view paradigm in SSL often limits the flexibility of learned representations for downstream adaptation by creating performance trade-offs between high-level invariance-demanding tasks such as image classification and more fine-grained equivariance-related tasks. In this work, we propose \emph{seq-JEPA}, a world modeling framework that introduces architectural inductive biases into joint-embedding predictive architectures to resolve this trade-off. Without relying on dual equivariance predictors or loss terms, seq-JEPA simultaneously learns two architecturally separate representations for equivariance- and invariance-demanding tasks. To do so, our model processes short sequences of different views (observations) of inputs. Each encoded view is concatenated with an embedding of the relative transformation (action) that produces the next observation in the sequence. These view-action pairs are passed through a transformer encoder that outputs an aggregate representation. A predictor head then conditions this aggregate representation on the upcoming action to predict the representation of the next observation. Empirically, seq-JEPA demonstrates strong performance on both equivariance- and invariance-demanding downstream tasks without sacrificing one for the other. Furthermore, it excels at tasks that inherently require aggregating a sequence of observations, such as path integration across actions and predictive learning across eye movements.
Authors: Moritz Stargalla, Christoph Hertrich, Daniel Reichman
Abstract: An established measure of the expressive power of a given ReLU neural network is the number of linear regions into which it partitions the input space. There exist many different, non-equivalent definitions of what a linear region actually is. We systematically assess which papers use which definitions and discuss how they relate to each other. We then analyze the computational complexity of counting the number of such regions for the various definitions. Generally, this turns out to be an intractable problem. We prove NP- and #P-hardness results already for networks with one hidden layer and strong hardness of approximation results for two or more hidden layers. Finally, on the algorithmic side, we demonstrate that counting linear regions can at least be achieved in polynomial space for some common definitions.
Authors: Bradley Lamb, Saroj Upreti, Yunfei Wang, Daniel Struble, Chenhui Zhu, Guillaume Freychet, Xiaodan Gu, Boran Ma
Abstract: The morphology of block copolymers (BCPs) critically influences material properties and applications. This work introduces a machine learning (ML)-enabled, high-throughput framework for analyzing grazing incidence small-angle X-ray scattering (GISAXS) data and atomic force microscopy (AFM) images to characterize BCP thin film morphology. A convolutional neural network was trained to classify AFM images by morphology type, achieving 97% testing accuracy. Classified images were then analyzed to extract 2D grain size measurements from the samples in a high-throughput manner. ML models were developed to predict morphological features based on processing parameters such as solvent ratio, additive type, and additive ratio. GISAXS-based properties were predicted with strong performances ($R^2$ > 0.75), while AFM-based property predictions were less accurate ($R^2$ < 0.60), likely due to the localized nature of AFM measurements compared to the bulk information captured by GISAXS. Beyond model performance, interpretability was addressed using Shapley Additive exPlanations (SHAP). SHAP analysis revealed that the additive ratio had the largest impact on morphological predictions, where additive provides the BCP chains with increased volume to rearrange into thermodynamically favorable morphologies. This interpretability helps validate model predictions and offers insight into parameter importance. Altogether, the presented framework combining high-throughput characterization and interpretable ML offers an approach to exploring and optimizing BCP thin film morphology across a broad processing landscape.
Authors: Hieu Tran, Phuong-Anh Nguyen-Le, Huy Nghiem, Quang-Nhan Nguyen, Wei Ai, Marine Carpuat
Abstract: Machine translation (MT) systems universally degrade when faced with code-mixed text. This problem is more acute for low-resource languages that lack dedicated parallel corpora. This work directly addresses this gap for Vietnamese-English, a language context characterized by challenges including orthographic ambiguity and the frequent omission of diacritics in informal text. We introduce VietMix, the first expert-translated, naturally occurring parallel corpus of Vietnamese-English code-mixed text. We establish VietMix's utility by developing a data augmentation pipeline that leverages iterative fine-tuning and targeted filtering. Experiments show that models augmented with our data outperform strong back-translation baselines by up to +3.5 xCOMET points and improve zero-shot models by up to +11.9 points. Our work delivers a foundational resource for a challenging language pair and provides a validated, transferable framework for building and augmenting corpora in other low-resource settings.
Authors: Elias Najarro, Nicolas Bessone, Sebastian Risi
Abstract: Self-organizing systems demonstrate how simple local rules can generate complex stochastic patterns. Many natural systems rely on such dynamics, making self-organization central to understanding natural complexity. A fundamental challenge in modeling such systems is solving the inverse problem: finding the unknown causal parameters from macroscopic observations. This task becomes particularly difficult when observations have a strong stochastic component, yielding diverse yet equivalent patterns. Traditional inverse methods fail in this setting, as pixel-wise metrics cannot capture feature similarities between variable outcomes. In this work, we introduce a novel inverse modeling method specifically designed to handle stochasticity in the observable space, leveraging the capacity of visual embeddings to produce robust representations that capture perceptual invariances. By mapping the pattern representations onto an invariant embedding space, we can effectively recover unknown causal parameters without the need for handcrafted objective functions or heuristics. We evaluate the method on three self-organizing systems: a physical, a biological, and a social one; namely, a reaction-diffusion system, a model of embryonic development, and an agent-based model of social segregation. We show that the method reliably recovers parameters despite stochasticity in the pattern outcomes. We further apply the method to real biological patterns, highlighting its potential as a tool for both theorists and experimentalists to investigate the dynamics underlying complex stochastic pattern formation.
Authors: Rok Cestnik, Erik A. Martens
Abstract: We present a simple and scalable implementation of next-generation reservoir computing (NGRC) for modeling dynamical systems from time-series data. The method uses a pseudorandom nonlinear projection of time-delay embedded inputs, allowing the feature-space dimension to be chosen independently of the observation size and offering a flexible alternative to polynomial-based NGRC projections. We demonstrate the approach on benchmark tasks, including attractor reconstruction and bifurcation diagram estimation, using partial and noisy measurements. We further show that small amounts of measurement noise during training act as an effective regularizer, improving long-term autonomous stability compared to standard regression alone. Across all tests, the models remain stable over long rollouts and generalize beyond the training data. The framework offers explicit control of system state during prediction, and these properties make NGRC a natural candidate for applications such as surrogate modeling and digital-twin applications.
Authors: Tom Burgert, Oliver Stoll, Paolo Rota, Beg\"um Demir
Abstract: The hypothesis that Convolutional Neural Networks (CNNs) are inherently texture-biased has shaped much of the discourse on feature use in deep learning. We revisit this hypothesis by examining limitations in the cue-conflict experiment by Geirhos et al. To address these limitations, we propose a domain-agnostic framework that quantifies feature reliance through systematic suppression of shape, texture, and color cues, avoiding the confounds of forced-choice conflicts. By evaluating humans and neural networks under controlled suppression conditions, we find that CNNs are not inherently texture-biased but predominantly rely on local shape features. Nonetheless, this reliance can be substantially mitigated through modern training strategies or architectures (ConvNeXt, ViTs). We further extend the analysis across computer vision, medical imaging, and remote sensing, revealing that reliance patterns differ systematically: computer vision models prioritize shape, medical imaging models emphasize color, and remote sensing models exhibit a stronger reliance on texture. Code is available at https://github.com/tomburgert/feature-reliance.
Authors: Chao Ying, Jun Jin, Haotian Zhang, Qinglong Tian, Yanyuan Ma, Yixuan Li, Jiwei Zhao
Abstract: We study an unsupervised domain adaptation problem where the source domain consists of subpopulations defined by the binary label $Y$ and a binary background (or environment) $A$. We focus on a challenging setting in which one such subpopulation in the source domain is unobservable. Naively ignoring this unobserved group can result in biased estimates and degraded predictive performance. Despite this structured missingness, we show that the prediction in the target domain can still be recovered. Specifically, we rigorously derive both background-specific and overall prediction models for the target domain. For practical implementation, we propose the distribution matching method to estimate the subpopulation proportions. We provide theoretical guarantees for the asymptotic behavior of our estimator, and establish an upper bound on the prediction error. Experiments on both synthetic and real-world datasets show that our method outperforms the naive benchmark that does not account for this unobservable source subpopulation.
Authors: Amin Karimi Monsefi, Nikhil Bhendawade, Manuel Rafael Ciosici, Dominic Culver, Yizhe Zhang, Irina Belousova
Abstract: Autoregressive language models (ARMs) deliver strong likelihoods, but are inherently serial: they generate one token per forward pass, which limits throughput and inflates latency for long sequences. Diffusion Language Models (DLMs) parallelize across positions and thus appear promising for language generation, yet standard discrete diffusion typically needs hundreds to thousands of model evaluations to reach high quality, trading serial depth for iterative breadth. We introduce FS-DFM, Few-Step Discrete Flow-Matching. A discrete flow-matching model designed for speed without sacrificing quality. The core idea is simple: make the number of sampling steps an explicit parameter and train the model to be consistent across step budgets, so one big move lands where many small moves would. We pair this with a reliable update rule that moves probability in the right direction without overshooting, and with strong teacher guidance distilled from long-run trajectories. Together, these choices make few-step sampling stable, accurate, and easy to control. On language modeling benchmarks, FS-DFM with 8 sampling steps achieves perplexity parity with a 1,024-step discrete-flow baseline for generating 1,024 tokens using a similar-size model, delivering up to 128 times faster sampling and corresponding latency/throughput gains.
Authors: Anna Hallin
Abstract: The rise of foundation models -- large, pretrained machine learning models that can be finetuned to a variety of tasks -- has revolutionized the fields of natural language processing and computer vision. In high-energy physics, the question of whether these models can be implemented directly in physics research, or even built from scratch, tailored for particle physics data, has generated an increasing amount of attention. This review, which is the first on the topic of foundation models in high-energy physics, summarizes and discusses the research that has been published in the field so far.
Authors: Yoann Poupart
Abstract: Interpretability of Deep Neural Networks (DNNs) is a growing field driven by the study of vision and language models. Yet, some use cases, like image captioning, or domains like Deep Reinforcement Learning (DRL), require complex modelling, with multiple inputs and outputs or use composable and separated networks. As a consequence, they rarely fit natively into the API of popular interpretability frameworks. We thus present TDHook, an open-source, lightweight, generic interpretability framework based on $\texttt{tensordict}$ and applicable to any $\texttt{torch}$ model. It focuses on handling complex composed models which can be trained for Computer Vision, Natural Language Processing, Reinforcement Learning or any other domain. This library features ready-to-use methods for attribution, probing and a flexible get-set API for interventions, and is aiming to bridge the gap between these method classes to make modern interpretability pipelines more accessible. TDHook is designed with minimal dependencies, requiring roughly half as much disk space as $\texttt{transformer_lens}$, and, in our controlled benchmark, achieves up to a $\times$2 speed-up over $\texttt{captum}$ when running integrated gradients for multi-target pipelines on both CPU and GPU. In addition, to value our work, we showcase concrete use cases of our library with composed interpretability pipelines in Computer Vision (CV) and Natural Language Processing (NLP), as well as with complex models in DRL.
Authors: Runyang You, Yongqi Li, Meng Liu, Wenjie Wang, Liqiang Nie, Wenjie Li
Abstract: Parallel test-time scaling (TTS) is a pivotal approach for enhancing large language models (LLMs), typically by sampling multiple token-based chains-of-thought in parallel and aggregating outcomes through voting or search. Recent advances in latent reasoning, where intermediate reasoning unfolds in continuous vector spaces, offer a more efficient alternative to explicit Chain-of-Thought, yet whether such latent models can similarly benefit from parallel TTS remains open, mainly due to the absence of sampling mechanisms in continuous space, and the lack of probabilistic signals for advanced trajectory aggregation. This work enables parallel TTS for latent reasoning models by addressing the above issues. For sampling, we introduce two uncertainty-inspired stochastic strategies: Monte Carlo Dropout and Additive Gaussian Noise. For aggregation, we design a Latent Reward Model (LatentRM) trained with step-wise contrastive objective to score and guide latent reasoning. Extensive experiments and visualization analyses show that both sampling strategies scale effectively with compute and exhibit distinct exploration dynamics, while LatentRM enables effective trajectory selection. Together, our explorations open a new direction for scalable inference in continuous spaces. Code and checkpoints released at https://github.com/ModalityDance/LatentTTS
Authors: Mete Harun Akcay, Buse Gul Atli, Siddharth Prakash Rao, Alexandros Bakas
Abstract: As the volume of stored data continues to grow, identifying and protecting sensitive information within large repositories becomes increasingly challenging, especially when shared with multiple users with different roles and permissions. This work presents a system architecture for trusted data sharing with policy-driven access control, enabling selective protection of sensitive regions while maintaining scalability. The proposed architecture integrates four core modules that combine automated detection of sensitive regions, post-correction, key management, and access control. Sensitive regions are secured using a hybrid scheme that employs symmetric encryption for efficiency and Attribute-Based Encryption for policy enforcement. The system supports efficient key distribution and isolates key storage to strengthen overall security. To demonstrate its applicability, we evaluate the system on visual datasets, where Privacy-Sensitive Objects in images are automatically detected, reassessed, and selectively encrypted prior to sharing in a data repository. Experimental results show that our system provides effective PSO detection, increases macro-averaged F1 score (5%) and mean Average Precision (10%), and maintains an average policy-enforced decryption time of less than 1 second per image. These results demonstrate the effectiveness, efficiency and scalability of our proposed solution for fine-grained access control.
Authors: Babaniyi Olaniyi, Ina Kalisa, Ana Fern\'andez del R\'io, Jean Marie Vianney Hakizayezu, Enric Jan\'e, Eniola Olaleye, Juan Francisco Garamendi, Ivan Nazarov, Aditya Rastogi, Mateo Diaz-Quiroz, \'Africa Peri\'a\~nez, Regis Hitimana
Abstract: As part of Rwanda's transition toward universal health coverage, the national Community-Based Health Insurance (CBHI) scheme is moving from retrospective fee-for-service reimbursements to prospective capitation payments for public primary healthcare providers. This work outlines a data-driven approach to designing, calibrating, and monitoring the capitation model using individual-level claims data from the Intelligent Health Benefits System (IHBS). We introduce a transparent, interpretable formula for allocating payments to Health Centers and their affiliated Health Posts. The formula is based on catchment population, service utilization patterns, and patient inflows, with parameters estimated via regression models calibrated on national claims data. Repeated validation exercises show the payment scheme closely aligns with historical spending while promoting fairness and adaptability across diverse facilities. In addition to payment design, the same dataset enables actionable behavioral insights. We highlight the use case of monitoring antibiotic prescribing patterns, particularly in pediatric care, to flag potential overuse and guideline deviations. Together, these capabilities lay the groundwork for a learning health financing system: one that connects digital infrastructure, resource allocation, and service quality to support continuous improvement and evidence-informed policy reform.
Authors: Ivan Zakazov, Berke Argin, Oussama Gabouj, Kamel Charaf, Alexander Sharipov, Alexi Semiz, Lorenzo Drudi, Nicolas Baldwin, Robert West
Abstract: Motivated by the high costs of using black-box Large Language Models (LLMs), we introduce a novel prompt compression paradigm, under which we use smaller LLMs to compress inputs for the larger ones. We present the first comprehensive LLM-as-a-compressor benchmark spanning 25 open- and closed-source models, which reveals significant disparity in models' compression ability in terms of (i) preserving semantically important information (ii) following the user-provided compression rate (CR). We further improve the performance of gpt-4.1-mini, the best overall vanilla compressor, with Textgrad-based compression meta-prompt optimization. We also identify the most promising open-source vanilla LLM - Qwen3-4B - and post-train it with a combination of supervised fine-tuning (SFT) and Group Relative Policy Optimization (GRPO), pursuing the dual objective of CR adherence and maximizing the downstream task performance. We call the resulting model Cmprsr and demonstrate its superiority over both extractive and vanilla abstractive compression across the entire range of compression rates on lengthy inputs from MeetingBank and LongBench as well as short prompts from GSM8k. The latter highlights Cmprsr's generalizability across varying input lengths and domains. Moreover, Cmprsr closely follows the requested compression rate, offering fine control over the cost-quality trade-off.
Authors: Yuexin Xiang, Yuchen Lei, Yuanzhe Zhang, Qin Wang, Tsz Hon Yuen, Jiangshan Yu
Abstract: Stablecoins such as USDT and USDC aspire to peg stability by coupling issuance controls with reserve attestations. In practice, however, the transparency is split across two worlds: verifiable on-chain traces and off-chain disclosures locked in unstructured text that are unconnected. We introduce a large language model (LLM)-based automated framework that bridges these two dimensions by aligning on-chain issuance data with off-chain disclosure statements. First, we propose an integrative framework using LLMs to capture and analyze on- and off-chain data through document parsing and semantic alignment, extracting key financial indicators from issuer attestations and mapping them to corresponding on-chain metrics. Second, we integrate multi-chain issuance records and disclosure documents within a model context protocol (MCP) framework that standardizes LLMs access to both quantitative market data and qualitative disclosure narratives. This framework enables unified retrieval and contextual alignment across heterogeneous stablecoin information sources and facilitates consistent analysis. Third, we demonstrate the capability of LLMs to operate across heterogeneous data modalities in blockchain analytics, quantifying discrepancies between reported and observed circulation and examining their implications for cross-chain transparency and price dynamics. Our findings reveal systematic gaps between disclosed and verifiable data, showing that LLM-assisted analysis enhances cross-modal transparency and supports automated, data-driven auditing in decentralized finance (DeFi).
Authors: Dashti A. Ali, Aras T. Asaad, Jacob J. Peoples, Mohammad Hamghalam, Alex Robins, Mane Piliposyan, Richard K. G. Do, Natalie Gangai, Yun S. Chun, Ahmad Bashir Barekzai, Jayasree Chakraborty, Hala Khasawneh, Camila Vilela, Natally Horvat, Jo\~ao Miranda, Alice C. Wei, Amber L. Simpson
Abstract: The development of machine learning (ML) models based on computed tomography (CT) imaging modality has been a major focus of recent research in the medical imaging domain. Incorporating robust feature engineering approach can highly improve the performance of these models. Topological data analysis (TDA), a recent development based on the mathematical field of algebraic topology, mainly focuses on the data from a topological perspective, extracting deeper insight and higher dimensional structures from the data. Persistent homology (PH), a fundamental tool in the area of TDA, can extract topological features such as connected components, cycles and voids from the data. A popular approach to construct PH from 3D CT images is to utilize the 3D cubical complex filtration, a method adapted for grid-structured data. However, this approach may not always yield the best performance and can suffer from computational complexity with higher resolution CT images. This study introduces a novel patch-based PH construction approach tailored for volumetric medical imaging data, in particular CT modality. A wide range of experiments has been conducted on several datasets of 3D CT images to comprehensively analyze the performance of the proposed method with various parameters and benchmark it against the 3D cubical complex algorithm. Our results highlight the dominance of the patch-based TDA approach in terms of both classification performance and time-efficiency. The proposed approach outperformed the cubical complex method, achieving average improvement of 10.38%, 6.94%, 2.06%, 11.58%, and 8.51% in accuracy, AUC, sensitivity, specificity, and F1 score, respectively, across all datasets. Finally, we provide a convenient python package, Patch-TDA, to facilitate the utilization of the proposed approach.
Authors: Akshaj Prashanth Rao, Advait Singh, Saumya Kumaar Saksena, Dhruv Kumar
Abstract: Prompt injection and jailbreaking attacks pose persistent security challenges to large language model (LLM)-based systems. We present PromptScreen, an efficient and systematically evaluated defense architecture that mitigates these threats through a lightweight, multi-stage pipeline. Its core component is a semantic filter based on text normalization, TF-IDF representations, and a Linear SVM classifier. Despite its simplicity, this module achieves 93.4% accuracy and 96.5% specificity on held-out data, substantially reducing attack throughput while incurring negligible computational overhead. Building on this efficient foundation, the full pipeline integrates complementary detection and mitigation mechanisms that operate at successive stages, providing strong robustness with minimal latency. In comparative experiments, our SVM-based configuration improves overall accuracy from 35.1% to 93.4% while reducing average time-to-completion from approximately 450 s to 47 s, yielding over 10 times lower latency than ShieldGemma. These results demonstrate that the proposed design simultaneously advances defensive precision and efficiency, addressing a core limitation of current model-based moderators. Evaluation across a curated corpus of over 30,000 labeled prompts, including benign, jailbreak, and application-layer injections, confirms that staged, resource-efficient defenses can robustly secure modern LLM-driven applications.
Authors: Basile Terver, Tsung-Yen Yang, Jean Ponce, Adrien Bardes, Yann LeCun
Abstract: A long-standing challenge in AI is to develop agents capable of solving a wide range of physical tasks and generalizing to new, unseen tasks and environments. A popular recent approach involves training a world model from state-action trajectories and subsequently use it with a planning algorithm to solve new tasks. Planning is commonly performed in the input space, but a recent family of methods has introduced planning algorithms that optimize in the learned representation space of the world model, with the promise that abstracting irrelevant details yields more efficient planning. In this work, we characterize models from this family as JEPA-WMs and investigate the technical choices that make algorithms from this class work. We propose a comprehensive study of several key components with the objective of finding the optimal approach within the family. We conducted experiments using both simulated environments and real-world robotic data, and studied how the model architecture, the training objective, and the planning algorithm affect planning success. We combine our findings to propose a model that outperforms two established baselines, DINO-WM and V-JEPA-2-AC, in both navigation and manipulation tasks. Code, data and checkpoints are available at https://github.com/facebookresearch/jepa-wms.
Authors: Hao-Tsung Yang, Ssu-Yuan Lo, Kuan-Lun Chen, Ching-Kai Wang
Abstract: We address the Diverse Traveling Salesman Problem (D-TSP), a bi-criteria optimization challenge that seeks a set of $k$ distinct TSP tours. The objective requires every selected tour to have a length at most $c|T^*|$ (where $|T^*|$ is the optimal tour length) while minimizing the average Jaccard similarity across all tour pairs. This formulation is crucial for applications requiring both high solution quality and fault tolerance, such as logistics planning, robotics pathfinding or strategic patrolling. Current methods are limited: traditional heuristics, such as the Niching Memetic Algorithm (NMA) or bi-criteria optimization, incur high computational complexity $O(n^3)$, while modern neural approaches (e.g., RF-MA3S) achieve limited diversity quality and rely on complex, external mechanisms. To overcome these limitations, we propose a novel hybrid framework that decomposes D-TSP into two efficient steps. First, we utilize a simple Graph Pointer Network (GPN), augmented with an approximated sequence entropy loss, to efficiently sample a large, diverse pool of high-quality tours. This simple modification effectively controls the quality-diversity trade-off without complex external mechanisms. Second, we apply a greedy algorithm that yields a 2-approximation for the dispersion problem to select the final $k$ maximally diverse tours from the generated pool. Our results demonstrate state-of-the-art performance. On the Berlin instance, our model achieves an average Jaccard index of $0.015$, significantly outperforming NMA ($0.081$) and RF-MA3S. By leveraging GPU acceleration, our GPN structure achieves a near-linear empirical runtime growth of $O(n)$. While maintaining solution diversity comparable to complex bi-criteria algorithms, our approach is over 360 times faster on large-scale instances (783 cities), delivering high-quality TSP solutions with unprecedented efficiency and simplicity.
Authors: Keith Frankston, Benjamin Howard
Abstract: We introduce a recursive AlphaZero-style Monte--Carlo tree search algorithm, "RMCTS". The advantage of RMCTS over AlphaZero's MCTS-UCB is speed. In RMCTS, the search tree is explored in a breadth-first manner, so that network inferences naturally occur in large batches. This significantly reduces the GPU latency cost. We find that RMCTS is often more than 40 times faster than MCTS-UCB when searching a single root state, and about 3 times faster when searching a large batch of root states. The recursion in RMCTS is based on computing optimized posterior policies at each game state in the search tree, starting from the leaves and working back up to the root. Here we use the posterior policy explored in "Monte--Carlo tree search as regularized policy optimization" (Grill, et al.) Their posterior policy is the unique policy which maximizes the expected reward given estimated action rewards minus a penalty for diverging from the prior policy. The tree explored by RMCTS is not defined in an adaptive manner, as it is in MCTS-UCB. Instead, the RMCTS tree is defined by following prior network policies at each node. This is a disadvantage, but the speedup advantage is more significant, and in practice we find that RMCTS-trained networks match the quality of MCTS-UCB-trained networks in roughly one-third of the training time. We include timing and quality comparisons of RMCTS vs. MCTS-UCB for three games: Connect-4, Dots-and-Boxes, and Othello.
Authors: Alois Duston, Tan Bui-Thanh
Abstract: We study variance reduction for score estimation and diffusion-based sampling in settings where the clean (target) score is available or can be approximated. Starting from the Target Score Identity (TSI), which expresses the noisy marginal score as a conditional expectation of the target score under the forward diffusion, we develop: (i) a plug-and-play nonparametric self-normalized importance sampling estimator compatible with standard reverse-time solvers, (ii) a variance-minimizing \emph{state- and time-dependent} blending rule between Tweedie-type and TSI estimators together with an anti-correlation analysis, (iii) a data-only extension based on locally fitted proxy scores, and (iv) a likelihood-tilting extension to Bayesian inverse problems. We also propose a \emph{Critic--Gate} distillation scheme that amortizes the state-dependent blending coefficient into a neural gate. Experiments on synthetic targets and PDE-governed inverse problems demonstrate improved sample quality for a fixed simulation budget.
Authors: Dennis Holzmann, Sven Wachsmuth
Abstract: We present a novel approach for hand-object action recognition that leverages 2D point tracks as an additional motion cue. While most existing methods rely on RGB appearance, human pose estimation, or their combination, our work demonstrates that tracking randomly sampled image points across video frames can substantially improve recognition accuracy. Unlike prior approaches, we do not detect hands, objects, or interaction regions. Instead, we employ CoTracker to follow a set of randomly initialized points through each video and use the resulting trajectories, together with the corresponding image frames, as input to a Transformer-based recognition model. Surprisingly, our method achieves notable gains even when only the initial frame and the point tracks are provided, without incorporating the full video sequence. Experimental results confirm that integrating 2D point tracks consistently enhances performance compared to the same model trained without motion information, highlighting their potential as a lightweight yet effective representation for hand-object action understanding.
Authors: Santiago Acevedo, Alessandro Laio, Marco Baroni
Abstract: We study how syntactic and semantic information is encoded in inner layer representations of Large Language Models (LLMs), focusing on the very large DeepSeek-V3. We find that, by averaging hidden-representation vectors of sentences sharing syntactic structure or meaning, we obtain vectors that capture a significant proportion of the syntactic and semantic information contained in the representations. In particular, subtracting these syntactic and semantic ``centroids'' from sentence vectors strongly affects their similarity with syntactically and semantically matched sentences, respectively, suggesting that syntax and semantics are, at least partially, linearly encoded. We also find that the cross-layer encoding profiles of syntax and semantics are different, and that the two signals can to some extent be decoupled, suggesting differential encoding of these two types of linguistic information in LLM representations.