Authors: Xueyi Wang, Elisabeth Wilhelm
Abstract: Sleep quality significantly impacts well-being. Therefore, healthcare providers and individuals need accessible and reliable forecasting tools for preventive interventions. This paper introduces an interpretable, individualized two-stage adaptive spatial-temporal model for predicting sleep quality scores. Our proposed framework combines multi-scale convolutional layers to model spatial interactions across multiple input variables, recurrent layers and attention mechanisms to capture long-term temporal dependencies, and a two-stage domain adaptation strategy to enhance generalization. The first adaptation stage is applied during training to mitigate overfitting on the training set. In the second stage, a source-free test-time adaptation mechanism is employed to adapt the model to new users without requiring labels. We conducted various experiments with five input window sizes (3, 5, 7, 9, and 11 days) and five prediction window sizes (1, 3, 5, 7, and 9 days). Our model consistently outperformed time series forecasting baseline approaches, including Long Short-Term Memory (LSTM), Informer, PatchTST, and TimesNet. The best performance was achieved with a three-day input window and a one-day prediction window, yielding a root mean square error (RMSE) of 0.216. Furthermore, the model demonstrated good predictive performance even for longer forecasting horizons (e.g, with a 0.257 RMSE for a three-day prediction window), highlighting its practical utility for real-world applications. We also conducted an explainability analysis to examine how different features influence sleep quality. These findings proved that the proposed framework offers a robust, adaptive, and explainable solution for personalized sleep forecasting using sparse data from commercial wearable devices.
Authors: Yu Song, Zhigang Hua, Yan Xie, Jingzhe Liu, Bo Long, Hui Liu
Abstract: Self-supervised learning (SSL) has shown great promise in graph representation learning. However, most existing graph SSL methods are developed and evaluated under a single-dataset setting, leaving their cross-dataset transferability largely unexplored and limiting their ability to leverage knowledge transfer and large-scale pretraining, factors that are critical for developing generalized intelligence beyond fitting training data. To address this gap and advance foundation model research for graphs, we present GSTBench, the first systematic benchmark for evaluating the transferability of graph SSL methods. We conduct large-scale pretraining on ogbn-papers100M and evaluate five representative SSL methods across a diverse set of target graphs. Our standardized experimental setup decouples confounding factors such as model architecture, dataset characteristics, and adaptation protocols, enabling rigorous comparisons focused solely on pretraining objectives. Surprisingly, we observe that most graph SSL methods struggle to generalize, with some performing worse than random initialization. In contrast, GraphMAE, a masked autoencoder approach, consistently improves transfer performance. We analyze the underlying factors that drive these differences and offer insights to guide future research on transferable graph SSL, laying a solid foundation for the "pretrain-then-transfer" paradigm in graph learning. Our code is available at https://github.com/SongYYYY/GSTBench.
Authors: Lingyu Zhang, Pengfei Xu, Guobin Wu, Jian Liang, Ruiyang Dong, Yunhai Wang, Xuan Song
Abstract: Traffic demand prediction plays a critical role in intelligent transportation systems. Existing traffic prediction models primarily rely on temporal traffic data, with limited efforts incorporating human knowledge and experience for urban traffic demand forecasting. However, in real-world scenarios, traffic knowledge and experience derived from human daily life significantly influence precise traffic prediction. Such knowledge and experiences can guide the model in uncovering latent patterns within traffic data, thereby enhancing the accuracy and robustness of predictions. To this end, this paper proposes integrating structured temporal traffic data with textual data representing human knowledge and experience, resulting in a novel knowledge-guided cross-modal feature representation learning (KGCM) model for traffic demand prediction. Based on regional transportation characteristics, we construct a prior knowledge dataset using a large language model combined with manual authoring and revision, covering both regional and global knowledge and experiences. The KGCM model then learns multimodal data features through designed local and global adaptive graph networks, as well as a cross-modal feature fusion mechanism. A proposed reasoning-based dynamic update strategy enables dynamic optimization of the graph model's parameters, achieving optimal performance. Experiments on multiple traffic datasets demonstrate that our model accurately predicts future traffic demand and outperforms existing state-of-the-art (SOTA) models.
Authors: Zehua Li
Abstract: This paper presents a configuration-first framework for evaluating cross-backend compatibility in deep learning systems deployed on CPU, GPU, and compiled runtimes. The framework decouples experiments from code using YAML, supports both library and repository models, and employs a three-tier verification protocol covering tensor-level closeness, activation alignment, and task-level metrics. Through 672 checks across multiple models and tolerance settings, we observe that 72.0% of runs pass, with most discrepancies occurring under stricter thresholds. Our results show that detection models and compiled backends are particularly prone to drift, often due to nondeterministic post-processing. We further demonstrate that deterministic adapters and selective fallbacks can substantially improve agreement without significant performance loss. To our knowledge, this is the first unified framework that systematically quantifies and mitigates cross-backend drift in deep learning, providing a reproducible methodology for dependable deployment across heterogeneous runtimes.
Authors: Wenxiong Li, Hanyu Liao, Suiyin Chen
Abstract: In reliability engineering, conventional surrogate models encounter the "curse of dimensionality" as the number of random variables increases. While the active learning Kriging surrogate approaches with high-dimensional model representation (HDMR) enable effective approximation of high-dimensional functions and are widely applied to optimization problems, there are rare studies specifically focused on reliability analysis, which prioritizes prediction accuracy in critical regions over uniform accuracy across the entire domain. This study develops an active learning surrogate model method based on the Kriging-HDMR modeling for reliability analysis. The proposed approach facilitates the approximation of high-dimensional limit state functions through a composite representation constructed from multiple low-dimensional sub-surrogate models. The architecture of the surrogate modeling framework comprises three distinct stages: developing single-variable sub-surrogate models for all random variables, identifying the requirements for coupling-variable sub-surrogate models, and constructing the coupling-variable sub-surrogate models. Optimization mathematical models for selection of design of experiment samples are formulated based on each stage's characteristics, with objectives incorporating uncertainty variance, predicted mean, sample location and inter-sample distances. A candidate sample pool-free approach is adopted to achieve the selection of informative samples. Numerical experiments demonstrate that the proposed method achieves high computational efficiency while maintaining strong predictive accuracy in solving high-dimensional reliability problems.
Authors: Zirui Li, Bin Yang, Meng Wang
Abstract: Arrival time prediction (ATP) of public transport vehicles is essential in improving passenger experience and supporting traffic management. Deep learning has demonstrated outstanding performance in ATP due to its ability to model non-linear and temporal dynamics. In the multi-step ATP, non-stationary data will degrade the model performance due to the variation in variables' joint distribution along the temporal direction. Previous studies mainly applied normalization to eliminate the non-stationarity in time series, thereby achieving better predictability. However, the normalization may obscure useful characteristics inherent in non-stationarity, which is known as the over-stationarization. In this work, to trade off predictability and non-stationarity, a new approach for multi-step ATP, named non-stationary ATP ( NSATP), is proposed. The method consists of two stages: series stationarization and non-stationarity effect recovery. The first stage aims at improving the predictability. As for the latter, NSATP extends a state-of-the-art method from one-dimensional to two dimensional based models to capture the hidden periodicity in time series and designs a compensation module of over-stationarization by learning scaling and shifting factors from raw data. 125 days' public transport operational data of Dresden is collected for validation. Experimental results show that compared to baseline methods, the proposed NSATP can reduce RMSE, MAE, and MAPE by 2.37%, 1.22%, and 2.26% for trams and by 1.72%, 0.60%, and 1.17% for buses, respectively.
Authors: Jiajun Chai, Guojun Yin, Zekun Xu, Chuhuai Yue, Yi Jia, Siyu Xia, Xiaohan Wang, Jiwen Jiang, Xiaoguang Li, Chengqi Dong, Hang He, Wei Lin
Abstract: Large language models excel at basic reasoning but struggle with tasks that require interaction with external tools. We present RLFactory, a plug-and-play reinforcement learning post-training framework for multi-round tool use. RLFactory tackles (i) tool-call stability and adaptability amid tool heterogeneity and interface issues via an asyncio-based asynchronous caller and a decoupled tool/training architecture, and (ii) diverse evaluation needs via a reward layer supporting rule-based, model-judgment, and tool-verification signals. It reconstructs the MDP by introducing observation markers from tool feedback, closing the loop among model, tools, and environment, and implements a generate-parse-invoke-update workflow for dynamic policy optimization. On Search-R1 with Qwen3-4B, RLFactory achieves a 0.486 test score on the Natural Questions (NQ) dataset, surpassing larger models trained with similar techniques (e.g., Qwen2.5-7B-Instruct-GRPO at 0.473), and increases training throughput by 6.8x. RLFactory provides a low-barrier, highly adaptable framework for strengthening multi-round tool use of LLMs in real-world scenarios. Code: https://github.com/Simple-Efficient/RL-Factory.
Authors: Xiaomeng Hu, Fei Huang, Chenhan Yuan, Junyang Lin, Tsung-Yi Ho
Abstract: As large language models (LLMs) are increasingly deployed in real-world applications, ensuring the safety of their outputs during decoding has become a critical challenge. However, existing decoding-time interventions, such as Contrastive Decoding, often force a severe trade-off between safety and response quality. In this work, we propose CARE, a novel framework for decoding-time safety alignment that integrates three key components: (1) a guard model for real-time safety monitoring, enabling detection of potentially unsafe content; (2) a rollback mechanism with a token buffer to correct unsafe outputs efficiently at an earlier stage without disrupting the user experience; and (3) a novel introspection-based intervention strategy, where the model generates self-reflective critiques of its previous outputs and incorporates these reflections into the context to guide subsequent decoding steps. The framework achieves a superior safety-quality trade-off by using its guard model for precise interventions, its rollback mechanism for timely corrections, and our novel introspection method for effective self-correction. Experimental results demonstrate that our framework achieves a superior balance of safety, quality, and efficiency, attaining a low harmful response rate and minimal disruption to the user experience while maintaining high response quality.
Authors: Lishan Yang, Nam Kha Nguygen, Po Hu, Wei Emma Zhang, Yanjun Shu, Mong Yuan Sim, Weitong Chen
Abstract: Foundation models have demonstrated remarkable performance across a wide range of tasks, yet their large parameter sizes pose challenges for practical deployment, especially in decentralized environments. Parameter-efficient fine-tuning (PEFT), such as Low-Rank Adaptation (LoRA), reduces local computing and memory overhead, making it attractive for federated learning. However, existing federated LoRA methods typically assume uniform rank configurations and unimodal inputs, overlooking two key real-world challenges: (1) heterogeneous client resources have different LoRA ranks, and (2) multimodal data settings with potentially missing modalities. In this work, we propose FediLoRA, a simple yet effective framework for federated multimodal fine-tuning under heterogeneous LoRA ranks and missing modalities. FediLoRA introduces a dimension-wise aggregation strategy that reweights LoRA updates without information dilution during aggregation. It also includes a lightweight layer-wise model editing method that selectively incorporates global parameters to repair local components which improves both client and global model performances. Experimental results on three multimodal benchmark datasets demonstrate that FediLoRA achieves superior performance over competitive baselines in both global and personalized settings, particularly in the presence of modality incompleteness.
Authors: Sima Najafzadehkhoei, George Vega Yon, Bernardo Modenesi, Derek S. Meyer
Abstract: Calibrating agent-based epidemic models is computationally demanding. We present a supervised machine learning calibrator that learns the inverse mapping from epidemic time series to SIR parameters. A three-layer bidirectional LSTM ingests 60-day incidence together with population size and recovery rate, and outputs transmission probability, contact rate, and R0. Training uses a composite loss with an epidemiology-motivated consistency penalty that encourages R0 \* recovery rate to equal transmission probability \* contact rate. In a 1000-scenario simulation study, we compare the calibrator with Approximate Bayesian Computation (likelihood-free MCMC). The method achieves lower error across all targets (MAE: R0 0.0616 vs 0.275; transmission 0.0715 vs 0.128; contact 1.02 vs 4.24), produces tighter predictive intervals with near nominal coverage, and reduces wall clock time from 77.4 s to 2.35 s per calibration. Although contact rate and transmission probability are partially nonidentifiable, the approach reproduces epidemic curves more faithfully than ABC, enabling fast and practical calibration. We evaluate it on SIR agent based epidemics generated with epiworldR and provide an implementation in R.
Authors: Xinquan Wu, Xuefeng Yan, Mingqiang Wei, Donghai Guan
Abstract: The Flexible Job-shop Scheduling Problem (FJSP) is a classical combinatorial optimization problem that has a wide-range of applications in the real world. In order to generate fast and accurate scheduling solutions for FJSP, various deep reinforcement learning (DRL) scheduling methods have been developed. However, these methods are mainly focused on the design of DRL scheduling Agent, overlooking the modeling of DRL environment. This paper presents a simple chronological DRL environment for FJSP based on discrete event simulation and an end-to-end DRL scheduling model is proposed based on the proximal policy optimization (PPO). Furthermore, a short novel state representation of FJSP is proposed based on two state variables in the scheduling environment and a novel comprehensible reward function is designed based on the scheduling area of machines. Experimental results on public benchmark instances show that the performance of simple priority dispatching rules (PDR) is improved in our scheduling environment and our DRL scheduling model obtains competing performance compared with OR-Tools, meta-heuristic, DRL and PDR scheduling methods.
Authors: Eduardo Lobo Lustoda Cabral, Paulo Pirozelli, Larissa Driemeier
Abstract: The increasing size of large neural network models, specifically language models and foundational image models, poses deployment challenges, prompting efforts to reduce memory requirements and enhance computational efficiency. These efforts are critical to ensure practical deployment and effective utilization of these models across various applications. In this work, a novel type of neural network layers and models is developed that uses only single-bit parameters. In this novel type of models all parameters of all layers, including kernel weights and biases, only have values equal to zero or one. This novel type of models uses layers named as binary normalized layer. These binary normalized layers can be of any type, such as fully connected, convolutional, attention, etc., and they consist of slight variations of the corresponding conventional layers. To show the effectiveness of the binary normalized layers, two different models are configured to solve a multiclass image classification problem and a language decoder to predict the next token of a sequence. The model to solve the image classification has convolutional and fully connected layers, and the language model is composed of transformer blocks with multi-head attention. The results show that models with binary normalized layers present almost the same results obtained by equivalent models with real 32-bit parameters. The binary normalized layers allow to develop models that use 32 times less memory than current models and have equivalent performance. Besides, the binary normalized layers can be easily implemented on current computers using 1-bit arrays, and do not require the development of dedicated electronic hardware. This novel type of layers opens a new era for large neural network models with reduced memory requirements that can be deployed using simple and cheap hardware, such as mobile devices or only cpus.
Authors: Vishal Rishi
Abstract: Slow feature analysis (SFA), as a method for learning slowly varying features in classification and signal analysis, has attracted increasing attention in recent years. Recent probabilistic extensions to SFA learn effective representations for classification tasks. Notably, the Probabilistic Adaptive Slow Feature Analysis models the slow features as states in an ARMA process and estimate the model from the observations. However, there is a need to develop efficient methods to infer the states (slow features) from the observations and the model. In this paper, a recursive extension to the linear PASFA has been proposed. The proposed algorithm performs MMSE estimation of states evolving according to an ARMA process, given the observations and the model. Although current methods tackle this problem using Kalman filters after transforming the ARMA process into a state space model, the original states (or slow features) that form useful representations cannot be easily recovered. The proposed technique is evaluated on a synthetic dataset to demonstrate its correctness.
Authors: Kaizheng Wang
Abstract: The Bayesian paradigm offers principled tools for sequential decision-making under uncertainty, but its reliance on a probabilistic model for all parameters can hinder the incorporation of complex structural constraints. We introduce a minimalist Bayesian framework that places a prior only on the component of interest, such as the location of the optimum. Nuisance parameters are eliminated via profile likelihood, which naturally handles constraints. As a direct instantiation, we develop a MINimalist Thompson Sampling (MINTS) algorithm. Our framework accommodates structured problems, including continuum-armed Lipschitz bandits and dynamic pricing. It also provides a probabilistic lens on classical convex optimization algorithms such as the center of gravity and ellipsoid methods. We further analyze MINTS for multi-armed bandits and establish near-optimal regret guarantees.
Authors: Federico Cerutti
Abstract: This paper presents a methodological approach to financial time series analysis by combining causal discovery and uncertainty-aware forecasting. As a case study, we focus on four key U.S. macroeconomic indicators -- GDP, economic growth, inflation, and unemployment -- and we apply the LPCMCI framework with Gaussian Process Distance Correlation (GPDC) to uncover dynamic causal relationships in quarterly data from 1970 to 2021. Our results reveal a robust unidirectional causal link from economic growth to GDP and highlight the limited connectivity of inflation, suggesting the influence of latent factors. Unemployment exhibits strong autoregressive dependence, motivating its use as a case study for probabilistic forecasting. Leveraging the Chronos framework, a large language model trained for time series, we perform zero-shot predictions on unemployment. This approach delivers accurate forecasts one and two quarters ahead, without requiring task-specific training. Crucially, the model's uncertainty-aware predictions yield 90\% confidence intervals, enabling effective anomaly detection through statistically principled deviation analysis. This study demonstrates the value of combining causal structure learning with probabilistic language models to inform economic policy and enhance forecasting robustness.
Authors: Serra Aksoy
Abstract: Artificial intelligence deployment for automated photovoltaic (PV) monitoring faces interpretability barriers that limit adoption in energy infrastructure applications. While deep learning achieves high accuracy in thermal fault detection, validation that model decisions align with thermal physics principles remains lacking, creating deployment hesitancy where understanding model reasoning is critical. This study provides a systematic comparison of convolutional neural networks (ResNet-18, EfficientNet-B0) and vision transformers (ViT-Tiny, Swin-Tiny) for thermal PV fault detection, using XRAI saliency analysis to assess alignment with thermal physics principles. This represents the first systematic comparison of CNNs and vision transformers for thermal PV fault detection with physics-validated interpretability. Evaluation on 20,000 infrared images spanning normal operation and 11 fault categories shows that Swin Transformer achieves the highest performance (94% binary accuracy; 73% multiclass accuracy) compared to CNN approaches. XRAI analysis reveals that models learn physically meaningful features, such as localized hotspots for cell defects, linear thermal paths for diode failures, and thermal boundaries for vegetation shading, consistent with expected thermal signatures. However, performance varies significantly across fault types: electrical faults achieve strong detection (F1-scores >0.90) while environmental factors like soiling remain challenging (F1-scores 0.20-0.33), indicating limitations imposed by thermal imaging resolution. The thermal physics-guided interpretability approach provides methodology for validating AI decision-making in energy monitoring applications, addressing deployment barriers in renewable energy infrastructure.
Authors: Sergey Pozdnyakov, Philippe Schwaller
Abstract: High-dimensional linear mappings, or linear layers, dominate both the parameter count and the computational cost of most modern deep-learning models. We introduce a general drop-in replacement, lookup multivariate Kolmogorov-Arnold Networks (lmKANs), which deliver a substantially better trade-off between capacity and inference cost. Our construction expresses a general high-dimensional mapping through trainable low-dimensional multivariate functions. These functions can carry dozens or hundreds of trainable parameters each, and yet it takes only a few multiplications to compute them because they are implemented as spline lookup tables. Empirically, lmKANs reduce inference FLOPs by up to 6.0x while matching the flexibility of MLPs in general high-dimensional function approximation. In another feedforward fully connected benchmark, on the tabular-like dataset of randomly displaced methane configurations, lmKANs enable more than 10x higher H100 throughput at equal accuracy. Within frameworks of Convolutional Neural Networks, lmKAN-based CNNs cut inference FLOPs at matched accuracy by 1.6-2.1x and by 1.7x on the CIFAR-10 and ImageNet-1k datasets, respectively. Our code, including dedicated CUDA kernels, is available online at https://github.com/schwallergroup/lmkan.
Authors: Ziheng Chen, Xiao-Jun Wu, Nicu Sebe
Abstract: Normalization layers are crucial for deep learning, but their Euclidean formulations are inadequate for data on manifolds. On the other hand, many Riemannian manifolds in machine learning admit gyro-structures, enabling principled extensions of Euclidean neural networks to non-Euclidean domains. Inspired by this, we introduce GyroBN, a principled Riemannian batch normalization framework for gyrogroups. We establish two necessary conditions, namely \emph{pseudo-reduction} and \emph{gyroisometric gyrations}, that guarantee GyroBN with theoretical control over sample statistics, and show that these conditions hold for all known gyrogroups in machine learning. Our framework also incorporates several existing Riemannian normalization methods as special cases. We further instantiate GyroBN on seven representative geometries, including the Grassmannian, five constant curvature spaces, and the correlation manifold, and derive novel gyro and Riemannian structures to enable these instantiations. Experiments across these geometries demonstrate the effectiveness of GyroBN. The code is available at https://github.com/GitZH-Chen/GyroBN.git.
Authors: Adrian Hayler, Xingyue Huang, \.Ismail \.Ilkan Ceylan, Michael Bronstein, Ben Finkelshtein
Abstract: Graph foundation models (GFMs) have recently emerged as a promising paradigm for achieving broad generalization across various graph data. However, existing GFMs are often trained on datasets that were shown to poorly represent real-world graphs, limiting their generalization performance. In contrast, tabular foundation models (TFMs) not only excel at classical tabular prediction tasks but have also shown strong applicability in other domains such as time series forecasting, natural language processing, and computer vision. Motivated by this, we take an alternative view to the standard perspective of GFMs and reformulate node classification as a tabular problem. Each node can be represented as a row with feature, structure, and label information as columns, enabling TFMs to directly perform zero-shot node classification via in-context learning. In this work, we introduce TabGFM, a graph foundation model framework that first converts a graph into a table via feature and structural encoders, applies multiple TFMs to diversely subsampled tables, and then aggregates their outputs through ensemble selection. Through experiments on 28 real-world datasets, TabGFM achieves consistent improvements over task-specific GNNs and state-of-the-art GFMs, highlighting the potential of tabular reformulation for scalable and generalizable graph learning.
Authors: Anatoly A. Krasnovsky
Abstract: Mechanistic interpretability has identified functional subgraphs within large language models (LLMs), known as Transformer Circuits (TCs), that appear to implement specific algorithms. Yet we lack a formal, single-pass way to quantify when an active circuit is behaving coherently and thus likely trustworthy. Building on prior systems-theoretic proposals, we specialize a sheaf/cohomology and causal emergence perspective to TCs and introduce the Effective-Information Consistency Score (EICS). EICS combines (i) a normalized sheaf inconsistency computed from local Jacobians and activations, with (ii) a Gaussian EI proxy for circuit-level causal emergence derived from the same forward state. The construction is white-box, single-pass, and makes units explicit so that the score is dimensionless. We further provide practical guidance on score interpretation, computational overhead (with fast and exact modes), and a toy sanity-check analysis. Empirical validation on LLM tasks is deferred.
Authors: Andy Xu, Rohan Desai, Larry Wang, Gabriel Hope, Ethan Ritz
Abstract: Discovering novel materials is critical for technological advancements such as solar cells, batteries, and carbon capture. However, the development of new materials is constrained by a slow and expensive trial-and-error process. To accelerate this pipeline, we introduce PLaID++, a Large Language Model (LLM) fine-tuned for stable and property-guided crystal generation. We fine-tune Qwen-2.5 7B to generate crystal structures using a novel Wyckoff-based text representation. We show that generation can be effectively guided with a reinforcement learning technique based on Direct Preference Optimization (DPO), with sampled structures categorized by their stability, novelty, and space group. By encoding symmetry constraints directly into text and guiding model outputs towards desirable chemical space, PLaID++ generates structures that are thermodynamically stable, unique, and novel at a $\sim$50\% greater rate than prior methods and conditionally generates structures with desired space group properties. Our experiments highlight the effectiveness of iterative DPO, achieving $\sim$115\% and $\sim$50\% improvements in unconditional and space group conditioned generation, respectively, compared to fine-tuning alone. Our work demonstrates the potential of adapting post-training techniques from natural language processing to materials design, paving the way for targeted and efficient discovery of novel materials.
Authors: Yiyue Chen, Usman Akram, Chianing Wang, Haris Vikalo
Abstract: Motivated by the high resource costs and privacy concerns associated with centralized machine learning, federated learning (FL) has emerged as an efficient alternative that enables clients to collaboratively train a global model while keeping their data local. However, in real-world deployments, client data distributions often evolve over time and differ significantly across clients, introducing heterogeneity that degrades the performance of standard FL algorithms. In this work, we introduce Fed-REACT, a federated learning framework designed for heterogeneous and evolving client data. Fed-REACT combines representation learning with evolutionary clustering in a two-stage process: (1) in the first stage, each client learns a local model to extracts feature representations from its data; (2) in the second stage, the server dynamically groups clients into clusters based on these representations and coordinates cluster-wise training of task-specific models for downstream objectives such as classification or regression. We provide a theoretical analysis of the representation learning stage, and empirically demonstrate that Fed-REACT achieves superior accuracy and robustness on real-world datasets.
Authors: Adrien Couetoux, Thomas Devenyns, Lise Diagne, David Champagne, Pierre-Yves Mousset, Chris Anagnostopoulos
Abstract: In pharmaceutical R&D, predicting the efficacy of a pharmaceutical in treating a particular disease prior to clinical testing or any real-world use has been challenging. In this paper, we propose a flexible and modular machine learning-based approach for predicting the efficacy of an untested pharmaceutical for treating a disease. We train a machine learning model using sets of pharmaceutical-pathway weight impact scores and patient data, which can include patient characteristics and observed clinical outcomes. The resulting model then analyses weighted impact scores of an untested pharmaceutical across human biological molecule-protein pathways to generate a predicted efficacy value. We demonstrate how the method works on a real-world dataset with patient treatments and outcomes, with two different weight impact score algorithms We include methods for evaluating the generalisation performance on unseen treatments, and to characterise conditions under which the approach can be expected to be most predictive. We discuss specific ways in which our approach can be iterated on, making it an initial framework to support future work on predicting the effect of untested drugs, leveraging RWD clinical data and drug embeddings.
Authors: Abhimanyu Bellam, Jung-Eun Kim
Abstract: Post Training Quantization (PTQ) is widely adopted due to its high compression capacity and speed with minimal impact on accuracy. However, we observed that disparate impacts are exacerbated by quantization, especially for minority groups. Our analysis explains that in the course of quantization there is a chain of factors attributed to a disparate impact across groups during forward and backward passes. We explore how the changes in weights and activations induced by quantization cause cascaded impacts in the network, resulting in logits with lower variance, increased loss, and compromised group accuracies. We extend our study to verify the influence of these impacts on group gradient norms and eigenvalues of the Hessian matrix, providing insights into the state of the network from an optimization point of view. To mitigate these effects, we propose integrating mixed precision Quantization Aware Training (QAT) with dataset sampling methods and weighted loss functions, therefore providing fair deployment of quantized neural networks.
Authors: Pranav Pawar, Dhwaj Jain, Varun Gupta, Kaustav Dedhia, Dashrath Kale, Sudhir Dhekane
Abstract: This paper presents a practical investigation into fine-tuning model parameters for mathematical reasoning tasks through experimenting with various configurations including randomness control, reasoning depth, and sampling strategies, careful tuning demonstrates substantial improvements in efficiency as well as performance. A holistically optimized framework is introduced for five state-of-the-art models on mathematical reasoning tasks, exhibiting significant performance boosts while maintaining solution correctness. Through systematic parameter optimization across Qwen2.5-72B, Llama-3.1-70B, DeepSeek-V3, Mixtral-8x22B, and Yi-Lightning, consistent efficiency gains are demonstrated with 100% optimization success rate. The methodology achieves an average 29.4% reduction in computational cost and 23.9% improvement in inference speed across all tested models. This framework systematically searches parameter spaces including temperature (0.1-0.5), reasoning steps (4-12), planning periods (1-4), and nucleus sampling (0.85-0.98), determining optimal configurations through testing on mathematical reasoning benchmarks. Critical findings show that lower temperature regimes (0.1-0.4) and reduced reasoning steps (4-6) consistently enhance efficiency without compromising accuracy. DeepSeek-V3 achieves the highest accuracy at 98%, while Mixtral-8x22B delivers the most cost-effective performance at 361.5 tokens per accurate response. Key contributions include: (1) the first comprehensive optimization study for five diverse SOTA models in mathematical reasoning, (2) a standardized production-oriented parameter optimization framework, (3) discovery of universal optimization trends applicable across model architectures, and (4) production-ready configurations with extensive performance characterization.
Authors: Shalev Manor, Mohammad Kohandel
Abstract: Solving inverse problems with Physics-Informed Neural Networks (PINNs) is computationally expensive for multi-query scenarios, as each new set of observed data requires a new, expensive training procedure. We present Inverse-Parameter Basis PINNs (IP-Basis PINNs), a meta-learning framework that extends the foundational work of Desai et al. (2022) to enable rapid and efficient inference for inverse problems. Our method employs an offline-online decomposition: a deep network is first trained offline to produce a rich set of basis functions that span the solution space of a parametric differential equation. For each new inverse problem online, this network is frozen, and solutions and parameters are inferred by training only a lightweight linear output layer against observed data. Key innovations that make our approach effective for inverse problems include: (1) a novel online loss formulation for simultaneous solution reconstruction and parameter identification, (2) a significant reduction in computational overhead via forward-mode automatic differentiation for PDE loss evaluation, and (3) a non-trivial validation and early-stopping mechanism for robust offline training. We demonstrate the efficacy of IP-Basis PINNs on three diverse benchmarks, including an extension to universal PINNs for unknown functional terms-showing consistent performance across constant and functional parameter estimation, a significant speedup per query over standard PINNs, and robust operation with scarce and noisy data.
Authors: Evgeny Alves Limarenko, Anastasiia Alexandrovna Studenikina
Abstract: In multi-task learning (MTL), gradient conflict poses a significant challenge. Effective methods for addressing this problem, including PCGrad, CAGrad, and GradNorm, in their original implementations are computationally demanding, which significantly limits their application in modern large models and transformers. We propose Gradient Conductor (GCond), a method that builds upon PCGrad principles by combining them with gradient accumulation and an adaptive arbitration mechanism. We evaluated GCond on self-supervised learning tasks using MobileNetV3-Small and ConvNeXt architectures on the ImageNet 1K dataset and a combined head and neck CT scan dataset, comparing the proposed method against baseline linear combinations and state-of-the-art gradient conflict resolution methods. The stochastic mode of GCond achieved a two-fold computational speedup while maintaining optimization quality, and demonstrated superior performance across all evaluated metrics, achieving lower L1 and SSIM losses compared to other methods on both datasets. GCond exhibited high scalability, being successfully applied to both compact models (MobileNetV3-Small) and large architectures (ConvNeXt-tiny and ConvNeXt-Base). It also showed compatibility with modern optimizers such as AdamW and Lion/LARS. Therefore, GCond offers a scalable and efficient solution to the problem of gradient conflicts in multi-task learning.
Authors: Luke McLennan, Yi Wang, Ryan Farell, Minh Nguyen, Chandrajit Bajaj
Abstract: We introduce a robust framework for learning various generalized Hamiltonian dynamics from noisy, sparse phase-space data and in an unsupervised manner based on variational Bayesian inference. Although conservative, dissipative, and port-Hamiltonian systems might share the same initial total energy of a closed system, it is challenging for a single Hamiltonian network model to capture the distinctive and varying motion dynamics and physics of a phase space, from sampled observational phase space trajectories. To address this complicated Hamiltonian manifold learning challenge, we extend sparse symplectic, random Fourier Gaussian processes learning with predictive successive numerical estimations of the Hamiltonian landscape, using a generalized form of state and conjugate momentum Hamiltonian dynamics, appropriate to different classes of conservative, dissipative and port-Hamiltonian physical systems. In addition to the kernelized evidence lower bound (ELBO) loss for data fidelity, we incorporate stability and conservation constraints as additional hyper-parameter balanced loss terms to regularize the model's multi-gradients, enforcing physics correctness for improved prediction accuracy with bounded uncertainty.
Authors: Jeff Shen, Lindsay Smith
Abstract: We present cryptogram solving as an ideal testbed for studying neural network generalization in combinatorially complex domains. In this task, models must decrypt text encoded with substitution ciphers, choosing from 26! possible mappings without explicit access to the cipher. We develop ALICE (an Architecture for Learning Interpretable Cryptogram dEcipherment): a simple encoder-only Transformer that sets a new state-of-the-art for both accuracy and speed on this decryption problem. Surprisingly, ALICE generalizes to unseen ciphers after training on only ${\sim}1500$ unique ciphers, a minute fraction ($3.7 \times 10^{-24}$) of the possible cipher space. To enhance interpretability, we introduce a novel bijective decoding head that explicitly models permutations via the Gumbel-Sinkhorn method, enabling direct extraction of learned cipher mappings. Through early exit analysis, we reveal how ALICE progressively refines its predictions in a way that appears to mirror common human strategies for this task: early layers employ frequency-based heuristics, middle layers form word structures, and final layers correct individual characters. Our architectural innovations and analysis methods extend beyond cryptograms to any domain with bijective mappings and combinatorial structure, offering new insights into neural network generalization and interpretability.
Authors: Alyssa Unell, Noel C. F. Codella, Sam Preston, Peniel Argaw, Wen-wai Yim, Zelalem Gero, Cliff Wong, Rajesh Jena, Eric Horvitz, Amanda K. Hall, Ruican Rachel Zhong, Jiachen Li, Shrey Jain, Mu Wei, Matthew Lungren, Hoifung Poon
Abstract: The National Comprehensive Cancer Network (NCCN) provides evidence-based guidelines for cancer treatment. Translating complex patient presentations into guideline-compliant treatment recommendations is time-intensive, requires specialized expertise, and is prone to error. Advances in large language model (LLM) capabilities promise to reduce the time required to generate treatment recommendations and improve accuracy. We present an LLM agent-based approach to automatically generate guideline-concordant treatment trajectories for patients with non-small cell lung cancer (NSCLC). Our contributions are threefold. First, we construct a novel longitudinal dataset of 121 cases of NSCLC patients that includes clinical encounters, diagnostic results, and medical histories, each expertly annotated with the corresponding NCCN guideline trajectories by board-certified oncologists. Second, we demonstrate that existing LLMs possess domain-specific knowledge that enables high-quality proxy benchmark generation for both model development and evaluation, achieving strong correlation (Spearman coefficient r=0.88, RMSE = 0.08) with expert-annotated benchmarks. Third, we develop a hybrid approach combining expensive human annotations with model consistency information to create both the agent framework that predicts the relevant guidelines for a patient, as well as a meta-classifier that verifies prediction accuracy with calibrated confidence scores for treatment recommendations (AUROC=0.800), a critical capability for communicating the accuracy of outputs, custom-tailoring tradeoffs in performance, and supporting regulatory compliance. This work establishes a framework for clinically viable LLM-based guideline adherence systems that balance accuracy, interpretability, and regulatory requirements while reducing annotation costs, providing a scalable pathway toward automated clinical decision support.
Authors: Li-Chin Chen, Ji-Tian Sheu, Yuh-Jue Chuang
Abstract: Demographic attributes are universally present in electronic health records and serve as vital predictors in clinical risk stratification and treatment decisions. Despite their significance, these attributes are often relegated to auxiliary roles in model design, with limited attention has been given to learning their representations. This study proposes a General Demographic Pre-trained (GDP) model as a foundational representation framework tailored to age and gender. The model is pre-trained and evaluated using datasets with diverse diseases and population compositions from different geographic regions. The GDP architecture explores combinations of ordering strategies and encoding methods to transform tabular demographic inputs into latent embeddings. Experimental results demonstrate that sequential ordering substantially improves model performance in discrimination, calibration, and the corresponding information gain at each decision tree split, particularly in diseases where age and gender contribute significantly to risk stratification. Even in datasets where demographic attributes hold relatively low predictive value, GDP enhances the representational importance, increasing their influence in downstream gradient boosting models. The findings suggest that foundational models for tabular demographic attributes can generalize across tasks and populations, offering a promising direction for improving predictive performance in healthcare applications.
Authors: Yuxuan Bai, Yuxuan Sun, Tan Chen, Wei Chen, Sheng Zhou, Zhisheng Niu
Abstract: Federated edge learning (FEEL) enables collaborative model training across distributed clients over wireless networks without exposing raw data. While most existing studies assume static datasets, in real-world scenarios clients may continuously collect data with time-varying and non-independent and identically distributed (non-i.i.d.) characteristics. A critical challenge is how to adapt models in a timely yet efficient manner to such evolving data. In this paper, we propose FedTeddi, a temporal-drift-and-divergence-aware scheduling algorithm that facilitates fast convergence of FEEL under dynamic data evolution and communication resource limits. We first quantify the temporal dynamics and non-i.i.d. characteristics of data using temporal drift and collective divergence, respectively, and represent them as the Earth Mover's Distance (EMD) of class distributions for classification tasks. We then propose a novel optimization objective and develop a joint scheduling and bandwidth allocation algorithm, enabling the FEEL system to learn from new data quickly without forgetting previous knowledge. Experimental results show that our algorithm achieves higher test accuracy and faster convergence compared to benchmark methods, improving the rate of convergence by 58.4% on CIFAR-10 and 49.2% on CIFAR-100 compared to random scheduling.
Authors: Qihu Xie, Yuan Li, Yi Kang
Abstract: Implicit neural representations have recently been extended to represent convolutional neural network weights via neural representation for neural networks, offering promising parameter compression benefits. However, standard multi-layer perceptrons used in neural representation for neural networks exhibit a pronounced spectral bias, hampering their ability to reconstruct high-frequency details effectively. In this paper, we propose SBS, a parameter-efficient enhancement to neural representation for neural networks that suppresses spectral bias using two techniques: (1) a unidirectional ordering-based smoothing that improves kernel smoothness in the output space, and (2) unidirectional ordering-based smoothing aware random fourier features that adaptively modulate the frequency bandwidth of input encodings based on layer-wise parameter count. Extensive evaluations on various ResNet models with datasets CIFAR-10, CIFAR-100, and ImageNet, demonstrate that SBS achieves significantly better reconstruction accuracy with less parameters compared to SOTA.
Authors: Qasim Zia, Avais Jan, Zafar Iqbal, Muhammad Mumtaz Ali, Mukarram Ali, Murray Patterson
Abstract: Cardiac arrest is one of the biggest global health problems, and early identification and management are key to enhancing the patient's prognosis. In this paper, we propose a novel framework that combines an EfficientNet-based deep learning model with a digital twin system to improve the early detection and analysis of cardiac arrest. We use compound scaling and EfficientNet to learn the features of cardiovascular images. In parallel, the digital twin creates a realistic and individualized cardiovascular system model of the patient based on data received from the Internet of Things (IoT) devices attached to the patient, which can help in the constant assessment of the patient and the impact of possible treatment plans. As shown by our experiments, the proposed system is highly accurate in its prediction abilities and, at the same time, efficient. Combining highly advanced techniques such as deep learning and digital twin (DT) technology presents the possibility of using an active and individual approach to predicting cardiac disease.
Authors: Gyuyeon Na, Minjung Park, Hyeonjeong Cha, Soyoun Kim, Sunyoung Moon, Sua Lee, Jaeyoung Choi, Hyemin Lee, Sangmi Chai
Abstract: Blockchain transaction networks are complex, with evolving temporal patterns and inter-node relationships. To detect illicit activities, we propose a hybrid GCN-GRU model that captures both structural and sequential features. Using real Bitcoin transaction data (2020-2024), our model achieved 0.9470 Accuracy and 0.9807 AUC-ROC, outperforming all baselines.
Authors: Arslan Majal, Aamir Hussain Chughtai, Muhammad Tahir
Abstract: We present a learning-based outlier-robust filter for a general setup where the measurement noise can be correlated. Since it is an enhanced version of EM-based outlier robust filter (EMORF), we call it as EMORF-II. As it is equipped with an additional powerful feature to learn the outlier characteristics during inference along with outlier-detection, EMORF-II has improved outlier-mitigation capability. Numerical experiments confirm performance gains as compared to the state-of-the-art methods in terms of accuracy with an increased computational overhead. However, thankfully the computational complexity order remains at par with other practical methods making it a useful choice for diverse applications.
Authors: Long Li, Jiaran Hao, Jason Klein Liu, Zhijian Zhou, Xiaoyu Tan, Wei Chu, Zhe Wang, Shirui Pan, Chao Qu, Yuan Qi
Abstract: A central paradox in fine-tuning Large Language Models (LLMs) with Reinforcement Learning with Verifiable Reward (RLVR) is the frequent degradation of multi-attempt performance (Pass@k) despite improvements in single-attempt accuracy (Pass@1). This is often accompanied by catastrophic forgetting, where models lose previously acquired skills. While various methods have been proposed, the choice and function of the divergence term have been surprisingly unexamined as a proactive solution. We argue that standard RLVR objectives -- both those using the mode-seeking reverse KL-divergence and those forgoing a divergence term entirely -- lack a crucial mechanism for knowledge retention. The reverse-KL actively accelerates this decay by narrowing the policy, while its absence provides no safeguard against the model drifting from its diverse knowledge base. We propose a fundamental shift in perspective: using the divergence term itself as the solution. Our framework, Diversity-Preserving Hybrid RL (DPH-RL), leverages mass-covering f-divergences (like forward-KL and JS-divergence) to function as a rehearsal mechanism. By continuously referencing the initial policy, this approach forces the model to maintain broad solution coverage. Extensive experiments on math and SQL generation demonstrate that DPH-RL not only resolves the Pass@k degradation but improves both Pass@1 and Pass@k in- and out-of-domain. Additionally, DPH-RL is more training-efficient because it computes f-divergence using generator functions, requiring only sampling from the initial policy and no online reference model. Our work highlights a crucial, overlooked axis for improving RLVR, demonstrating that the proper selection of a divergence measure is a powerful tool for building more general and diverse reasoning models.
Authors: Antoine Ledent, Petr Kasalick\'y, Rodrigo Alves, Hady W. Lauw
Abstract: We introduce a new convolutional AutoEncoder architecture for user modelling and recommendation tasks with several improvements over the state of the art. Firstly, our model has the flexibility to learn a set of associations and combinations between different interaction types in a way that carries over to each user and item. Secondly, our model is able to learn jointly from both the explicit ratings and the implicit information in the sampling pattern (which we refer to as `implicit feedback'). It can also make separate predictions for the probability of consuming content and the likelihood of granting it a high rating if observed. This not only allows the model to make predictions for both the implicit and explicit feedback, but also increases the informativeness of the predictions: in particular, our model can identify items which users would not have been likely to consume naturally, but would be likely to enjoy if exposed to them. Finally, we provide several generalization bounds for our model, which to the best of our knowledge, are among the first generalization bounds for auto-encoders in a Recommender Systems setting; we also show that optimizing our loss function guarantees the recovery of the exact sampling distribution over interactions up to a small error in total variation. In experiments on several real-life datasets, we achieve state-of-the-art performance on both the implicit and explicit feedback prediction tasks despite relying on a single model for both, and benefiting from additional interpretability in the form of individual predictions for the probabilities of each possible rating.
Authors: Adithya Ramachandran, Thorkil Flensmark B. Neergaard, Tom\'as Arias-Vergara, Andreas Maier, Siming Bayer
Abstract: Advancements in smart metering technologies have significantly improved the ability to monitor and manage water utilities. In the context of increasing uncertainty due to climate change, securing water resources and supply has emerged as an urgent global issue with extensive socioeconomic ramifications. Hourly consumption data from end-users have yielded substantial insights for projecting demand across regions characterized by diverse consumption patterns. Nevertheless, the prediction of water demand remains challenging due to influencing non-deterministic factors, such as meteorological conditions. This work introduces a novel method for short-term water demand forecasting for District Metered Areas (DMAs) which encompass commercial, agricultural, and residential consumers. Unsupervised contrastive learning is applied to categorize end-users according to distinct consumption behaviors present within a DMA. Subsequently, the distinct consumption behaviors are utilized as features in the ensuing demand forecasting task using wavelet-transformed convolutional networks that incorporate a cross-attention mechanism combining both historical data and the derived representations. The proposed approach is evaluated on real-world DMAs over a six-month period, demonstrating improved forecasting performance in terms of MAPE across different DMAs, with a maximum improvement of 4.9%. Additionally, it identifies consumers whose behavior is shaped by socioeconomic factors, enhancing prior knowledge about the deterministic patterns that influence demand.
Authors: Jad Yehya, Mansour Benbakoura, C\'edric Allain, Beno\^it Malezieux, Matthieu Kowalski, Thomas Moreau
Abstract: Identifying recurring patterns and rare events in large-scale signals is a fundamental challenge in fields such as astronomy, physical simulations, and biomedical science. Convolutional Dictionary Learning (CDL) offers a powerful framework for modeling local structures in signals, but its use for detecting rare or anomalous events remains largely unexplored. In particular, CDL faces two key challenges in this setting: high computational cost and sensitivity to artifacts and outliers. In this paper, we introduce RoseCDL, a scalable and robust CDL algorithm designed for unsupervised rare event detection in long signals. RoseCDL combines stochastic windowing for efficient training on large datasets with inline outlier detection to enhance robustness and isolate anomalous patterns. This reframes CDL as a practical tool for event discovery and characterization in real-world signals, extending its role beyond traditional tasks like compression or denoising.
Authors: Zhiyuan He, Xufang Luo, Yike Zhang, Yuqing Yang, Lili Qiu
Abstract: We propose $\Delta L$ Normalization, a simple yet effective loss aggregation method tailored to the characteristic of dynamic generation lengths in Reinforcement Learning with Verifiable Rewards (RLVR). Recently, RLVR has demonstrated strong potential in improving the reasoning capabilities of large language models (LLMs), but a major challenge lies in the large variability of response lengths during training, which leads to high gradient variance and unstable optimization. Although previous methods such as GRPO, DAPO, and Dr. GRPO introduce different loss normalization terms to address this issue, they either produce biased estimates or still suffer from high gradient variance. By analyzing the effect of varying lengths on policy loss both theoretically and empirically, we reformulate the problem as finding a minimum-variance unbiased estimator. Our proposed $\Delta L$ Normalization not only provides an unbiased estimate of the true policy loss but also minimizes gradient variance in theory. Extensive experiments show that it consistently achieves superior results across different model sizes, maximum lengths, and tasks. Our code will be made public at https://github.com/zerolllin/Delta-L-Normalization.
Authors: Zakeria Sharif Ali
Abstract: This paper introduces the Univariate Gaussian Mixture Model Neural Network (uGMM-NN), a novel neural architecture that embeds probabilistic reasoning directly into the computational units of deep networks. Unlike traditional neurons, which apply weighted sums followed by fixed nonlinearities, each uGMM-NN node parameterizes its activations as a univariate Gaussian mixture, with learnable means, variances, and mixing coefficients. This design enables richer representations by capturing multimodality and uncertainty at the level of individual neurons, while retaining the scalability of standard feedforward networks. We demonstrate that uGMM-NN can achieve competitive discriminative performance compared to conventional multilayer perceptrons, while additionally offering a probabilistic interpretation of activations. The proposed framework provides a foundation for integrating uncertainty-aware components into modern neural architectures, opening new directions for both discriminative and generative modeling.
Authors: Liya Gaynutdinova, Martin Do\v{s}k\'a\v{r}, Ond\v{r}ej Roko\v{s}, Ivana Pultarov\'a
Abstract: Physics-informed neural networks (PINNs) have shown promise in solving partial differential equations (PDEs) relevant to multiscale modeling, but they often fail when applied to materials with discontinuous coefficients, such as media with piecewise constant properties. This paper introduces a dual formulation for the PINN framework to improve the reliability of the homogenization of periodic thermo-conductive composites, for both strong and variational (weak) formulations. The dual approach facilitates the derivation of guaranteed upper and lower error bounds, enabling more robust detection of PINN failure. We compare standard PINNs applied to smoothed material approximations with variational PINNs (VPINNs) using both spectral and neural network-based test functions. Our results indicate that while strong-form PINNs may outperform VPINNs in controlled settings, they are sensitive to material discontinuities and may fail without clear diagnostics. In contrast, VPINNs accommodate piecewise constant material parameters directly but require careful selection of test functions to avoid instability. Dual formulation serves as a reliable indicator of convergence quality, and its integration into PINN frameworks enhances their applicability to homogenization problems in micromechanics.
Authors: Mehdi Bejani, Marco Mauri, Daniele Acconcia, Simone Todaro, Stefano Mariani
Abstract: This paper presents an innovative Transformer-based deep learning strategy for optimizing the placement of sensors aiming at structural health monitoring of semiconductor probe cards. Failures in probe cards, including substrate cracks and loosened screws, would critically affect semiconductor manufacturing yield and reliability. Some failure modes could be detected by equipping a probe card with adequate sensors. Frequency response functions from simulated failure scenarios are adopted within a finite element model of a probe card. A comprehensive dataset, enriched by physics-informed scenario expansion and physics-aware statistical data augmentation, is exploited to train a hybrid Convolutional Neural Network and Transformer model. The model achieves high accuracy (99.83%) in classifying the probe card health states (baseline, loose screw, crack) and an excellent crack detection recall (99.73%). Model robustness is confirmed through a rigorous framework of 3 repetitions of 10-fold stratified cross-validation. The attention mechanism also pinpoints critical sensor locations: an analysis of the attention weights offers actionable insights for designing efficient, cost-effective monitoring systems by optimizing sensor configurations. This research highlights the capability of attention-based deep learning to advance proactive maintenance, enhancing operational reliability and yield in semiconductor manufacturing.
Authors: Zhoujun Cheng, Richard Fan, Shibo Hao, Taylor W. Killian, Haonan Li, Suqi Sun, Hector Ren, Alexander Moreno, Daqian Zhang, Tianjun Zhong, Yuxin Xiong, Yuanzhe Hu, Yutao Xie, Xudong Han, Yuqi Wang, Varad Pimpalkhute, Yonghao Zhuang, Aaryamonvikram Singh, Xuezhi Liang, Anze Xie, Jianshu She, Desai Fan, Chengqian Gao, Liqun Ma, Mikhail Yurochkin, John Maggs, Xuezhe Ma, Guowei He, Zhiting Hu, Zhengzhong Liu, Eric P. Xing
Abstract: K2-Think is a reasoning system that achieves state-of-the-art performance with a 32B parameter model, matching or surpassing much larger models like GPT-OSS 120B and DeepSeek v3.1. Built on the Qwen2.5 base model, our system shows that smaller models can compete at the highest levels by combining advanced post-training and test-time computation techniques. The approach is based on six key technical pillars: Long Chain-of-thought Supervised Finetuning, Reinforcement Learning with Verifiable Rewards (RLVR), Agentic planning prior to reasoning, Test-time Scaling, Speculative Decoding, and Inference-optimized Hardware, all using publicly available open-source datasets. K2-Think excels in mathematical reasoning, achieving state-of-the-art scores on public benchmarks for open-source models, while also performing strongly in other areas such as Code and Science. Our results confirm that a more parameter-efficient model like K2-Think 32B can compete with state-of-the-art systems through an integrated post-training recipe that includes long chain-of-thought training and strategic inference-time enhancements, making open-source reasoning systems more accessible and affordable. K2-Think is freely available at k2think.ai, offering best-in-class inference speeds of over 2,000 tokens per second per request via the Cerebras Wafer-Scale Engine.
Authors: Ali Nawaz, Amir Ahmad, Shehroz S. Khan
Abstract: Class imbalance poses a significant challenge to supervised classification, particularly in critical domains like medical diagnostics and anomaly detection where minority class instances are rare. While numerous studies have explored rebalancing techniques to address this issue, less attention has been given to evaluating the performance of binary classifiers under imbalance when no such techniques are applied. Therefore, the goal of this study is to assess the performance of binary classifiers "as-is", without performing any explicit rebalancing. Specifically, we systematically evaluate the robustness of a diverse set of binary classifiers across both real-world and synthetic datasets, under progressively reduced minority class sizes, using one-shot and few-shot scenarios as baselines. Our approach also explores varying data complexities through synthetic decision boundary generation to simulate real-world conditions. In addition to standard classifiers, we include experiments using undersampling, oversampling strategies, and one-class classification (OCC) methods to examine their behavior under severe imbalance. The results confirm that classification becomes more difficult as data complexity increases and the minority class size decreases. While traditional classifiers deteriorate under extreme imbalance, advanced models like TabPFN and boosting-based ensembles retain relatively higher performance and better generalization compared to traditional classifiers. Visual interpretability and evaluation metrics further validate these findings. Our work offers valuable guidance on model selection for imbalanced learning, providing insights into classifier robustness without dependence on explicit rebalancing techniques.
Authors: Lachlan Simpson, Kyle Millar, Adriel Cheng, Cheng-Chew Lim, Hong Gunn Chew
Abstract: Integrated Gradients (IG) is a common explainability technique to address the black-box problem of neural networks. Integrated gradients assumes continuous data. Graphs are discrete structures making IG ill-suited to graphs. In this work, we introduce graph-based integrated gradients (GB-IG); an extension of IG to graphs. We demonstrate on four synthetic datasets that GB-IG accurately identifies crucial structural components of the graph used in classification tasks. We further demonstrate on three prevalent real-world graph datasets that GB-IG outperforms IG in highlighting important features for node classification tasks.
Authors: Pierre Lambert, Edouard Couplet, Michel Verleysen, John Aldo Lee
Abstract: Neighbour embeddings (NE) allow the representation of high dimensional datasets into lower dimensional spaces and are often used in data visualisation. In practice, accelerated approximations are employed to handle very large datasets. Accelerating NE is challenging, and two main directions have been explored: very coarse approximations based on negative sampling (as in UMAP) achieve high effective speed but may lack quality in the extracted structures; less coarse approximations, as used in FIt-SNE or BH-t-SNE, offer better structure preservation at the cost of speed, while also restricting the target dimensionality to 2 or 3, limiting NE to visualisation. In some variants, the precision of these costlier accelerations also enables finer-grained control on the extracted structures through dedicated hyperparameters. This paper proposes to bridge the gab between both approaches by introducing a novel way to accelerate NE, requiring a small number of computations per iteration while maintaining good fine-grained structure preservation and flexibility through hyperparameter tuning, without limiting the dimensionality of the embedding space. The method was designed for interactive exploration of data; as such, it abandons the traditional two-phased approach of other NE methods, allowing instantaneous visual feedback when changing hyperparameters, even when these control processes happening on the high-dimensional side of the computations. Experiments using a publicly available, GPU accelerated GUI integration of the method show promising results in terms of speed, flexibility in the structures getting extracted, and show potential uses in broader machine learning contexts with minimal algorithmic modifications. Central to this algorithm is a novel approach to iterative approximate nearest neighbour search, which shows promising results compared to nearest neighbour descent.
Authors: Shusen Ma, Tianhao Zhang, Qijiu Xia, Yun-Bo Zhao
Abstract: Multivariate time series forecasting (MTSF) often faces challenges from missing variables, which hinder conventional spatial-temporal graph neural networks in modeling inter-variable correlations. While GinAR addresses variable missing using attention-based imputation and adaptive graph learning for the first time, it lacks interpretability and fails to capture more latent temporal patterns due to its simple recursive units (RUs). To overcome these limitations, we propose the Interpretable Bidirectional-modeling Network (IBN), integrating Uncertainty-Aware Interpolation (UAI) and Gaussian kernel-based Graph Convolution (GGCN). IBN estimates the uncertainty of reconstructed values using MC Dropout and applies an uncertainty-weighted strategy to mitigate high-risk reconstructions. GGCN explicitly models spatial correlations among variables, while a bidirectional RU enhances temporal dependency modeling. Extensive experiments show that IBN achieves state-of-the-art forecasting performance under various missing-rate scenarios, providing a more reliable and interpretable framework for MTSF with missing variables. Code is available at: https://github.com/zhangth1211/NICLab-IBN.
Authors: Songkai Ma, Zhaorui Zhang, Sheng Di, Benben Liu, Xiaodong Yu, Xiaoyi Lu, Dan Wang
Abstract: With the widespread application of Mixture of Experts (MoE) reasoning models in the field of LLM learning, efficiently serving MoE models under limited GPU memory constraints has emerged as a significant challenge. Offloading the non-activated experts to main memory has been identified as an efficient approach to address such a problem, while it brings the challenges of transferring the expert between the GPU memory and main memory. We need to explore an efficient approach to compress the expert and analyze how the compression error affects the inference performance. To bridge this gap, we propose employing error-bounded lossy compression algorithms (such as SZ3 and CuSZp) to compress non-activated experts, thereby reducing data transfer overhead during MoE inference. We conduct extensive experiments across various benchmarks and present a comprehensive analysis of how compression-induced errors in different experts affect overall inference accuracy. The results indicate that experts in the shallow layers, which are primarily responsible for the attention mechanism and the transformation of input tokens into vector representations, exhibit minimal degradation in inference accuracy when subjected to bounded errors. In contrast, errors in the middle-layer experts, which are central to model reasoning, significantly impair inference accuracy. Interestingly, introducing bounded errors in the deep-layer experts, which are mainly responsible for instruction following and output integration, can sometimes lead to improvements in inference accuracy.
Authors: Jonathan Teagan
Abstract: This study applies a range of forecasting techniques,including ARIMA, Prophet, Long Short Term Memory networks (LSTM), Temporal Convolutional Networks (TCN), and XGBoost, to model and predict Russian equipment losses during the ongoing war in Ukraine. Drawing on daily and monthly open-source intelligence (OSINT) data from WarSpotting, we aim to assess trends in attrition, evaluate model performance, and estimate future loss patterns through the end of 2025. Our findings show that deep learning models, particularly TCN and LSTM, produce stable and consistent forecasts, especially under conditions of high temporal granularity. By comparing different model architectures and input structures, this study highlights the importance of ensemble forecasting in conflict modeling, and the value of publicly available OSINT data in quantifying material degradation over time.
Authors: Mohammad Zana Majidi, Sajjad Karimi, Teng Wang, Robert Kluger, Reginald Souleyrette
Abstract: Predicting injuries and fatalities in traffic crashes plays a critical role in enhancing road safety, improving emergency response, and guiding public health interventions. This study investigates the added value of unstructured crash narratives (written by police officers at the scene) when combined with structured crash data to predict injury severity. Two widely used Natural Language Processing (NLP) techniques, Term Frequency-Inverse Document Frequency (TF-IDF) and Word2Vec, were employed to extract semantic meaning from the narratives, and their effectiveness was compared. To address the challenge of class imbalance, a K-Nearest Neighbors-based oversampling method was applied to the training data prior to modeling. The dataset consists of crash records from Kentucky spanning 2019 to 2023. To account for roadway heterogeneity, three road classification schemes were used: (1) eight detailed functional classes (e.g., Urban Two-Lane, Rural Interstate, Urban Multilane Divided), (2) four broader paired categories (e.g., Urban vs. Rural, Freeway vs. Non-Freeway), and (3) a unified dataset without classification. A total of 102 machine learning models were developed by combining structured features and narrative-based features using the two NLP techniques alongside three ensemble algorithms: XGBoost, Random Forest, and AdaBoost. Results demonstrate that models incorporating narrative data consistently outperform those relying solely on structured data. Among all combinations, TF-IDF coupled with XGBoost yielded the most accurate predictions in most subgroups. The findings highlight the power of integrating textual and structured crash information to enhance person-level injury prediction. This work offers a practical and adaptable framework for transportation safety professionals to improve crash severity modeling, guide policy decisions, and design more effective countermeasures.
Authors: Antoine de Mathelin, Christopher Tosh, Wesley Tansey
Abstract: Personalizing combination therapies in oncology requires navigating an immense space of possible drug and dose combinations, a task that remains largely infeasible through exhaustive experimentation. Recent developments in patient-derived models have enabled high-throughput ex vivo screening, but the number of feasible experiments is limited. Further, a tight therapeutic window makes gathering molecular profiling information (e.g. RNA-seq) impractical as a means of guiding drug response prediction. This leads to a challenging cold-start problem: how do we select the most informative combinations to test early, when no prior information about the patient is available? We propose a strategy that leverages a pretrained deep learning model built on historical drug response data. The model provides both embeddings for drug combinations and dose-level importance scores, enabling a principled selection of initial experiments. We combine clustering of drug embeddings to ensure functional diversity with a dose-weighting mechanism that prioritizes doses based on their historical informativeness. Retrospective simulations on large-scale drug combination datasets show that our method substantially improves initial screening efficiency compared to baselines, offering a viable path for more effective early-phase decision-making in personalized combination drug screens.
Authors: Yajun Yu, Steve Jiang, Robert Timmerman, Hao Peng
Abstract: Personalized ultra-fractionated stereotactic adaptive radiotherapy (PULSAR) is a novel treatment that delivers radiation in pulses of protracted intervals. Accurate prediction of gross tumor volume (GTV) changes through regression models has substantial prognostic value. This study aims to develop a multi-omics based support vector regression (SVR) model for predicting GTV change. A retrospective cohort of 39 patients with 69 brain metastases was analyzed, based on radiomics (MRI images) and dosiomics (dose maps) features. Delta features were computed to capture relative changes between two time points. A feature selection pipeline using least absolute shrinkage and selection operator (Lasso) algorithm with weight- or frequency-based ranking criterion was implemented. SVR models with various kernels were evaluated using the coefficient of determination (R2) and relative root mean square error (RRMSE). Five-fold cross-validation with 10 repeats was employed to mitigate the limitation of small data size. Multi-omics models that integrate radiomics, dosiomics, and their delta counterparts outperform individual-omics models. Delta-radiomic features play a critical role in enhancing prediction accuracy relative to features at single time points. The top-performing model achieves an R2 of 0.743 and an RRMSE of 0.022. The proposed multi-omics SVR model shows promising performance in predicting continuous change of GTV. It provides a more quantitative and personalized approach to assist patient selection and treatment adjustment in PULSAR.
Authors: Katherine Berry, Liang Cheng
Abstract: Graph Neural Networks (GNNs) have gained traction in the complex domain of drug discovery because of their ability to process graph-structured data such as drug molecule models. This approach has resulted in a myriad of methods and models in published literature across several categories of drug discovery research. This paper covers the research categories comprehensively with recent papers, namely molecular property prediction, including drug-target binding affinity prediction, drug-drug interaction study, microbiome interaction prediction, drug repositioning, retrosynthesis, and new drug design, and provides guidance for future work on GNNs for drug discovery.
Authors: Philipp Lepold, Jonas Leichtle, Tobias R\"oddiger, Michael Beigl
Abstract: Automatic sleep staging typically relies on gold-standard EEG setups, which are accurate but obtrusive and impractical for everyday use outside sleep laboratories. This limits applicability in real-world settings, such as home environments, where continuous, long-term monitoring is needed. Detecting sleep onset is particularly relevant, enabling consumer applications (e.g. automatically pausing media playback when the user falls asleep). Recent research has shown correlations between in-ear EEG and full-scalp EEG for various phenomena, suggesting wearable, in-ear devices could allow unobtrusive sleep monitoring. We investigated the feasibility of using single-channel in-ear electrophysiological (ExG) signals for automatic sleep staging in a wearable device by conducting a sleep study with 11~participants (mean age: 24), using a custom earpiece with a dry eartip electrode (D\"atwyler SoftPulse) as a measurement electrode in one ear and a reference in the other. Ground truth sleep stages were obtained from an Apple Watch Ultra, validated for sleep staging. Our system achieved 90.5% accuracy for binary sleep detection (Awake vs. Asleep) and 65.1% accuracy for four-class staging (Awake, REM, Core, Deep) using leave-one-subject-out validation. These findings demonstrate the potential of in-ear electrodes as a low-effort, comfortable approach to sleep monitoring, with applications such as stopping podcasts when users fall asleep.
Authors: Qing-xin Meng, Xia Lei, Jian-wei Liu
Abstract: This paper investigates the problem of Online Convex-Concave Optimization, which extends Online Convex Optimization to two-player time-varying convex-concave games. The goal is to minimize the dynamic duality gap (D-DGap), a critical performance measure that evaluates players' strategies against arbitrary comparator sequences. Existing algorithms fail to deliver optimal performance, particularly in stationary or predictable environments. To address this, we propose a novel modular algorithm with three core components: an Adaptive Module that dynamically adjusts to varying levels of non-stationarity, a Multi-Predictor Aggregator that identifies the best predictor among multiple candidates, and an Integration Module that effectively combines their strengths. Our algorithm achieves a minimax optimal D-DGap upper bound, up to a logarithmic factor, while also ensuring prediction error-driven D-DGap bounds. The modular design allows for the seamless replacement of components that regulate adaptability to dynamic environments, as well as the incorporation of components that integrate ``side knowledge'' from multiple predictors. Empirical results further demonstrate the effectiveness and adaptability of the proposed method.
Authors: Hamid Ahmad, Heiko Paulheim, Rita T. Sousa
Abstract: Knowledge graphs and ontologies represent entities and their relationships in a structured way, having gained significance in the development of modern AI applications. Integrating these semantic resources with machine learning models often relies on knowledge graph embedding models to transform graph data into numerical representations. Therefore, pre-trained models for popular knowledge graphs and ontologies are increasingly valuable, as they spare the need to retrain models for different tasks using the same data, thereby helping to democratize AI development and enabling sustainable computing. In this paper, we present Bio-KGvec2go, an extension of the KGvec2go Web API, designed to generate and serve knowledge graph embeddings for widely used biomedical ontologies. Given the dynamic nature of these ontologies, Bio-KGvec2go also supports regular updates aligned with ontology version releases. By offering up-to-date embeddings with minimal computational effort required from users, Bio-KGvec2go facilitates efficient and timely biomedical research.
Authors: Arun Verma, Zhaoxuan Wu, Zijian Zhou, Xiaoqiang Lin, Zhiliang Chen, Rachael Hwee Ling Sim, Rui Qiao, Jingtan Wang, Nhung Bui, Xinyuan Niu, Wenyang Hu, Gregory Kang Ruey Lau, Zi-Yu Khoo, Zitong Zhao, Xinyi Xu, Apivich Hemachandra, See-Kiong Ng, Bryan Kian Hsiang Low
Abstract: Large Language Models (LLMs) are large-scale pretrained models that have achieved remarkable success across diverse domains. These successes have been driven by unprecedented complexity and scale in both data and computations. However, due to the high costs of training such models, brute-force trial-and-error approaches to improve LLMs are not feasible. Inspired by the success of inverse problems in uncovering fundamental scientific laws, this position paper advocates that inverse problems can also efficiently uncover scaling laws that guide the building of LLMs to achieve the desirable performance with significantly better cost-effectiveness.
Authors: Yuan Pu, Yazhe Niu, Jia Tang, Junyu Xiong, Shuai Hu, Hongsheng Li
Abstract: In heterogeneous multi-task learning, tasks not only exhibit diverse observation and action spaces but also vary substantially in intrinsic difficulty. While conventional multi-task world models like UniZero excel in single-task settings, we find that when handling large-scale heterogeneous environments, gradient conflicts and the loss of model plasticity often constrain their sample and computational efficiency. In this work, we address these challenges from two perspectives: the single learning iteration and the overall learning process. First, we investigate the impact of key design spaces on extending UniZero to multi-task planning. We find that a Mixture-of-Experts (MoE) architecture provides the most substantial performance gains by mitigating gradient conflicts, leading to our proposed model, \textit{ScaleZero}. Second, to dynamically balance the computational load across the learning process, we introduce an online, LoRA-based \textit{dynamic parameter scaling} (DPS) strategy. This strategy progressively integrates LoRA adapters in response to task-specific progress, enabling adaptive knowledge retention and parameter expansion. Empirical evaluations on standard benchmarks such as Atari, DMControl (DMC), and Jericho demonstrate that ScaleZero, relying exclusively on online reinforcement learning with one model, attains performance on par with specialized single-task baselines. Furthermore, when augmented with our dynamic parameter scaling strategy, our method achieves competitive performance while requiring only 80\% of the single-task environment interaction steps. These findings underscore the potential of ScaleZero for effective large-scale multi-task learning. Our code is available at \textcolor{magenta}{https://github.com/opendilab/LightZero}.
Authors: Kasra Borazjani, Naji Khosravan, Rajeev Sahay, Bita Akram, Seyyedali Hosseinalipour
Abstract: Multi-modal multi-task (M3T) foundation models (FMs) have recently shown transformative potential in artificial intelligence, with emerging applications in education. However, their deployment in real-world educational settings is hindered by privacy regulations, data silos, and limited domain-specific data availability. We introduce M3T Federated Foundation Models (FedFMs) for education: a paradigm that integrates federated learning (FL) with M3T FMs to enable collaborative, privacy-preserving training across decentralized institutions while accommodating diverse modalities and tasks. Subsequently, this position paper aims to unveil M3T FedFMs as a promising yet underexplored approach to the education community, explore its potentials, and reveal its related future research directions. We outline how M3T FedFMs can advance three critical pillars of next-generation intelligent education systems: (i) privacy preservation, by keeping sensitive multi-modal student and institutional data local; (ii) personalization, through modular architectures enabling tailored models for students, instructors, and institutions; and (iii) equity and inclusivity, by facilitating participation from underrepresented and resource-constrained entities. We finally identify various open research challenges, including studying of (i) inter-institution heterogeneous privacy regulations, (ii) the non-uniformity of data modalities' characteristics, (iii) the unlearning approaches for M3T FedFMs, (iv) the continual learning frameworks for M3T FedFMs, and (v) M3T FedFM model interpretability, which must be collectively addressed for practical deployment.
Authors: Oliver Daniels, Stuart Armstrong, Alexandre Maranh\~ao, Mahirah Fairuz Rahman, Benjamin M. Marlin, Rebecca Gorman
Abstract: Deep neural networks are notoriously sensitive to spurious correlations - where a model learns a shortcut that fails out-of-distribution. Existing work on spurious correlations has often focused on incomplete correlations,leveraging access to labeled instances that break the correlation. But in cases where the spurious correlations are complete, the correct generalization is fundamentally \textit{underspecified}. To resolve this underspecification, we propose learning a set of concepts that are consistent with training data but make distinct predictions on a subset of novel unlabeled inputs. Using a self-training approach that encourages \textit{confident} and \textit{selective} disagreement, our method ACE matches or outperforms existing methods on a suite of complete-spurious correlation benchmarks, while remaining robust to incomplete spurious correlations. ACE is also more configurable than prior approaches, allowing for straight-forward encoding of prior knowledge and principled unsupervised model selection. In an early application to language-model alignment, we find that ACE achieves competitive performance on the measurement tampering detection benchmark \textit{without} access to untrusted measurements. While still subject to important limitations, ACE represents significant progress towards overcoming underspecification.
Authors: Yilun Kuang, Noah Amsel, Sanae Lotfi, Shikai Qiu, Andres Potapczynski, Andrew Gordon Wilson
Abstract: The core component of attention is the scoring function, which transforms the inputs into low-dimensional queries and keys and takes the dot product of each pair. While the low-dimensional projection improves efficiency, it causes information loss for certain tasks that have intrinsically high-dimensional inputs. Additionally, attention uses the same scoring function for all input pairs, without imposing a distance-dependent compute bias for neighboring tokens in the sequence. In this work, we address these shortcomings by proposing new scoring functions based on computationally efficient structured matrices with high ranks, including Block Tensor-Train (BTT) and Multi-Level Low Rank (MLR) matrices. On in-context regression tasks with high-dimensional inputs, our proposed scoring functions outperform standard attention for any fixed compute budget. On language modeling, a task that exhibits locality patterns, our MLR-based attention method achieves improved scaling laws compared to both standard attention and variants of sliding window attention. Additionally, we show that both BTT and MLR fall under a broader family of efficient structured matrices capable of encoding either full-rank or distance-dependent compute biases, thereby addressing significant shortcomings of standard attention. Finally, we show that MLR attention has promising results for long-range time-series forecasting.
Authors: Yuxing Liu, Yuze Ge, Rui Pan, An Kang, Tong Zhang
Abstract: Learning rate warmup is a popular and practical technique in training large-scale deep neural networks. Despite the huge success in practice, the theoretical advantages of this strategy of gradually increasing the learning rate at the beginning of the training process have not been fully understood. To resolve this gap between theory and practice, we first propose a novel family of generalized smoothness assumptions, and validate its applicability both theoretically and empirically. Under the novel smoothness assumption, we study the convergence properties of gradient descent (GD) in both deterministic and stochastic settings. It is shown that learning rate warmup consistently accelerates GD, and GD with warmup can converge at most $\Theta(T)$ times faster than with a non-increasing learning rate schedule in some specific cases, providing insights into the benefits of this strategy from an optimization theory perspective.
Authors: David Egea, Barproda Halder, Sanghamitra Dutta
Abstract: Automated detection of vulnerabilities in source code is an essential cybersecurity challenge, underpinning trust in digital systems and services. Graph Neural Networks (GNNs) have emerged as a promising approach as they can learn structural and logical code relationships in a data-driven manner. However, their performance is severely constrained by training data imbalances and label noise. GNNs often learn 'spurious' correlations from superficial code similarities, producing detectors that fail to generalize well to unseen real-world data. In this work, we propose a unified framework for robust and interpretable vulnerability detection, called VISION, to mitigate spurious correlations by systematically augmenting a counterfactual training dataset. Counterfactuals are samples with minimal semantic modifications but opposite labels. Our framework includes: (i) generating counterfactuals by prompting a Large Language Model (LLM); (ii) targeted GNN training on paired code examples with opposite labels; and (iii) graph-based interpretability to identify the crucial code statements relevant for vulnerability predictions while ignoring spurious ones. We find that VISION reduces spurious learning and enables more robust, generalizable detection, improving overall accuracy (from 51.8% to 97.8%), pairwise contrast accuracy (from 4.5% to 95.8%), and worst-group accuracy (from 0.7% to 85.5%) on the Common Weakness Enumeration (CWE)-20 vulnerability. We further demonstrate gains using proposed metrics: intra-class attribution variance, inter-class attribution distance, and node score dependency. We also release CWE-20-CFA, a benchmark of 27,556 functions (real and counterfactual) from the high-impact CWE-20 category. Finally, VISION advances transparent and trustworthy AI-based cybersecurity systems through interactive visualization for human-in-the-loop analysis.
Authors: Jiahuan Yu (University of Illinois Urbana-Champaign), Aryan Taneja (University of Illinois Urbana-Champaign), Junfeng Lin (Tsinghua University), Minjia Zhang (University of Illinois Urbana-Champaign)
Abstract: Modern Large Language Model (LLM) serving systems increasingly support interactive applications, like real-time chat assistants, code generation tools, and agentic workflows. However, the soaring energy cost of LLM inference presents a growing challenge for sustainable and cost-effective deployment. This paper introduces VoltanaLLM, a system for SLO-aware, energy-efficient LLM serving, built from a control theory perspective. VoltanaLLM co-designs frequency scaling and request routing in emerging prefill/decode disaggregated architectures, leveraging their decoupled execution to enable fine-grained phase-specific control. It consists of a feedback-driven frequency controller that dynamically adapts GPU frequency for prefill and decode phases, and a state-space router that explores routing decisions across frequency-scaled instances to minimize energy under latency constraints. We implement VoltanaLLM in SGLang and evaluate its performance over multiple state-of-the-art LLMs and real-world datasets. The results demonstrate that VoltanaLLM achieves up to 36.3% energy savings while maintaining near-perfect SLO attainment rate, paving the way for sustainable and intelligent LLM serving.
Authors: Yaoying Fu
Abstract: Given a continuous finitely piecewise linear function $f:\mathbb{R}^{n_0} \to \mathbb{R}$ and a fixed architecture $(n_0,\ldots,n_k;1)$ of feedforward ReLU neural networks, the exact function realization problem is to determine when some network with the given architecture realizes $f$. To develop a systematic way to answer these questions, we establish a connection between toric geometry and ReLU neural networks. This approach enables us to utilize numerous structures and tools from algebraic geometry to study ReLU neural networks. Starting with an unbiased ReLU neural network with rational weights, we define the ReLU fan, the ReLU toric variety, and the ReLU Cartier divisor associated with the network. This work also reveals the connection between the tropical geometry and the toric geometry of ReLU neural networks. As an application of the toric geometry framework, we prove a necessary and sufficient criterion of functions realizable by unbiased shallow ReLU neural networks by computing intersection numbers of the ReLU Cartier divisor and torus-invariant curves.
Authors: Neal G. Ravindra, Arijit Sehanobish
Abstract: High-quality, medically validated labels exist for clinical actigraphy data but not for ubiquitous consumer wearables like the Apple Watch. Manually labeling wearables data is expensive and doesn't scale. This paper offers a novel framework that transfers valuable labels from a source domain (e.g., actigraphy) to a target domain (e.g., Apple Watch) without requiring paired data. Instead of working with raw time-series signals, we project both domains into a shared latent embedding space using time-series foundation models (TSFMs) and develop a new framework to align the cross-device representations. Our method, Adversarial Alignment of TSFM Embeddings forces the distributions of source and target embeddings to align within this space, facilitating label transfer across device type.
Authors: Yingsheng Wang, Shuo Lu, Jian Liang, Aihua Zheng, Ran He
Abstract: Out-of-distribution (OOD) detection helps models identify data outside the training categories, crucial for security applications. While feature-based post-hoc methods address this by evaluating data differences in the feature space without changing network parameters, they often require access to training data, which may not be suitable for some data privacy scenarios. This may not be suitable in scenarios where data privacy protection is a concern. In this paper, we propose a simple yet effective post-hoc method, termed Classifier-based Feature Reconstruction (ClaFR), from the perspective of subspace projection. It first performs an orthogonal decomposition of the classifier's weights to extract the class-known subspace, then maps the original data features into this subspace to obtain new data representations. Subsequently, the OOD score is determined by calculating the feature reconstruction error of the data within the subspace. Compared to existing OOD detection algorithms, our method does not require access to training data while achieving leading performance on multiple OOD benchmarks. Our code is released at https://github.com/Aie0923/ClaFR.
Authors: Bryan Rodas, Natalie Montesino, Jakob Ambsdorf, David Klindt, Randall Balestriero
Abstract: Continued pretraining offers a promising solution for adapting foundation models to a new target domain. However, in specialized domains, available datasets are often very small, limiting the applicability of SSL methods developed for large-scale pretraining and making hyperparameter search infeasible. In addition, pretrained models are usually released as backbone-weights only, lacking important information to continue pretraining. We propose to bridge this gap with DIET-CP, a simple continued pretraining strategy, where any strong foundation model can be steered towards the new data distribution of interest. DIET-CP relies on a very simple objective, requires no labels, and introduces no more hyperparameters than supervised finetuning. It is stable across data modalities and backbone choices, while providing a significant performance boost for state-of-the-art models such as DINOv3 using only 1000 images.
Authors: Jimmy Joseph
Abstract: In this paper, we introduce the Protocol Genome, a self-supervised learning system that learns correlations from DICOM headers and achieves AUROC 0.901 (vs 0.847 baseline) and ECE 0.036 (vs 0.058) on fully held-out external validation. Our method also improves calibration and robustness across modalities (CT, MRI, CXR) and vendors. Clinical imaging is funneled through PACS/DICOM, where procedure choices (scanner make/model, sequence, kernel, kVp, TR/TE, and slice thickness) have consequences for contrast, noise, and artifact. These latent confounders impede the generalization of image-only networks across sites. We consider structured DICOM headers as a label and learn protocol-aware but clinically robust image representations. Protocol Genome obtains tokenized embeddings of de-identified header fields and models them along with image features using: (1) protocol-image contrastive learning, (2) masked protocol prediction, and (3) protocol-protocol translation. With 1.26M studies (7 health systems, 31 scanners, 3 vendors; CT, MR, CR/DR), we experiment on: (A) chest CT triage for PE, (B) brain MRI glioma grading, and (C) chest radiograph cardiomegaly detection. Relative to strong SSL baselines (SimCLR, MAE) as well as ImageNet transfer, Protocol Genome (+0.046: PE, +0.058: glioma, +0.041: cardiomegaly) is associated with higher external AUROC; 25-37% calibration improvements are obtained (p < 0.01, DeLong tests). While the gains may be task-dependent, they are preserved with 10-20% of labeled data. From a clinical point of view, the technique reduces false positives at protocol borders and is applicable in a PACS (DICOM C-FIND/C-MOVE, DICOMweb QIDO/WADO). We publish a model card and deployment guide, complete with both de-identification and bias audits.
Authors: Liviu Nicolae Firc\u{a}, Antonio B\u{a}rb\u{a}lau, Dan Oneata, Elena Burceanu
Abstract: Can models generalize attribute knowledge across semantically and perceptually dissimilar categories? While prior work has addressed attribute prediction within narrow taxonomic or visually similar domains, it remains unclear whether current models can abstract attributes and apply them to conceptually distant categories. This work presents the first explicit evaluation for the robustness of the attribute prediction task under such conditions, testing whether models can correctly infer shared attributes between unrelated object types: e.g., identifying that the attribute "has four legs" is common to both "dogs" and "chairs". To enable this evaluation, we introduce train-test split strategies that progressively reduce correlation between training and test sets, based on: LLM-driven semantic grouping, embedding similarity thresholding, embedding-based clustering, and supercategory-based partitioning using ground-truth labels. Results show a sharp drop in performance as the correlation between training and test categories decreases, indicating strong sensitivity to split design. Among the evaluated methods, clustering yields the most effective trade-off, reducing hidden correlations while preserving learnability. These findings offer new insights into the limitations of current representations and inform future benchmark construction for attribute reasoning.
Authors: Youjie Li, Cheng Wan, Zhiqi Lin, Hongyu Zhu, Jiacheng Yang, Ziang Song, Xinyi Di, Jiawei Wu, Huiyao Shu, Wenlei Bao, Yanghua Peng, Haibin Lin, Li-Wen Chang
Abstract: Large Language Models (LLMs) have scaled rapidly in size and complexity, requiring increasingly intricate parallelism for distributed training, such as 3D parallelism. This sophistication motivates a shift toward simpler, more debuggable programming paradigm like Single Program Multiple Data (SPMD). However, SPMD in eager execution introduces two key challenges: ensuring consistency with single-device execution and achieving high performance at scale. In this paper, we introduce veScale, an eager-mode training system that fully embraces SPMD paradigm to democratize distributed tensor programming. veScale addresses the prevalent issue of inconsistent results in systems like PyTorch by introducing a novel algorithm of distributed Random Number Generation (RNG) compatible with arbitrary sharded operators. veScale also significantly boosts training performance by reducing PyTorch primitive's overhead and improving communication efficiency. Evaluations show that veScale delivers up to 2.2x speedup over the state-of-the-art training systems, like TorchTitan, and cuts code complexity by 78.4%, while preserving single-device-equivalent results.
Authors: Kapil Madan
Abstract: This paper introduces ArGen (Auto-Regulation of Generative AI systems), a framework for aligning Large Language Models (LLMs) with complex sets of configurable, machine-readable rules spanning ethical principles, operational safety protocols, and regulatory compliance standards. Moving beyond just preference-based alignment, ArGen is designed to ensure LLMs adhere to these multifaceted policies through a novel synthesis of principle-based automated reward scoring, Group Relative Policy Optimisation (GRPO), and an Open Policy Agent (OPA) inspired governance layer. This approach provides the technical foundation for achieving and demonstrating compliance with diverse and nuanced governance requirements. To showcase the framework's capability to operationalize a deeply nuanced and culturally-specific value system, we present an in-depth case study: the development of a medical AI assistant guided by principles from Dharmic ethics (such as Ahimsa and Dharma), as derived from texts like the Bhagavad Gita. This challenging application demonstrates ArGen's adaptability, achieving a 70.9% improvement in domain-scope adherence over the baseline. Through our open-source repository, we show that ArGen's methodology offers a path to 'Governable Al' systems that are technically proficient, ethically robust, and verifiably compliant for safe deployment in diverse global contexts.
Authors: Muhammad Arif Hakimi Zamrai, Kamaludin Mohd Yusof
Abstract: In response to the prevalent concern of TCP SYN flood attacks within the context of Software-Defined Internet of Vehicles (SD-IoV), this study addresses the significant challenge of network security in rapidly evolving vehicular communication systems. This research focuses on optimizing a Random Forest Classifier model to achieve maximum accuracy and minimal detection time, thereby enhancing vehicular network security. The methodology involves preprocessing a dataset containing SYN attack instances, employing feature scaling and label encoding techniques, and applying Stratified K-Fold cross-validation to target key metrics such as accuracy, precision, recall, and F1-score. This research achieved an average value of 0.999998 for all metrics with a SYN DoS attack detection time of 0.24 seconds. Results show that the fine-tuned Random Forest model, configured with 20 estimators and a depth of 10, effectively differentiates between normal and malicious traffic with high accuracy and minimal detection time, which is crucial for SD-IoV networks. This approach marks a significant advancement and introduces a state-of-the-art algorithm in detecting SYN flood attacks, combining high accuracy with minimal detection time. It contributes to vehicular network security by providing a robust solution against TCP SYN flood attacks while maintaining network efficiency and reliability.
Authors: Andrew Kiruluta, Priscilla Burity
Abstract: We introduce Spectral NSR, a fully spectral neuro-symbolic reasoning framework that embeds logical rules as spectral templates and performs inference directly in the graph spectral domain. By leveraging graph signal processing (GSP) and frequency-selective filters grounded in the Laplacian eigenstructure of knowledge graphs, the architecture unifies the interpretability of symbolic reasoning with the scalability and adaptability of spectral learning. Beyond the core formulation, we incorporate a comprehensive set of extensions, including dynamic graph and basis learning, rational and diffusion filters for sharper spectral selectivity, mixture-of-spectral-experts for modular specialization, proof-guided training with spectral curricula, and uncertainty quantification for calibrated confidence. Additional enhancements such as large language model coupling, co-spectral transfer alignment, adversarial robustness, efficient GPU kernels, generalized Laplacians, and causal interventions further expand the versatility of the framework. Empirical evaluation on state-of-the-art reasoning benchmarks such as ProofWriter and CLUTRR demonstrates that Spectral NSR achieves superior accuracy, faster inference, improved robustness to adversarial perturbations, and higher interpretability compared to leading baselines including transformers, message-passing neural networks, and neuro-symbolic logic programming systems. Spectral attribution and proof-band agreement analyses confirm that model decisions align closely with symbolic proof structures, while transfer experiments validate effective domain adaptation through co-spectral alignment. These results establish Spectral NSR as a scalable and principled foundation for the next generation of reasoning systems, offering transparency, robustness, and generalization beyond conventional approaches.
Authors: Jun Gao, Jie Ding
Abstract: Many data applications involve counting queries, where a client specifies a feasible range of variables and a database returns the corresponding item counts. A program that produces the counts of different queries often risks leaking sensitive individual-level information. A popular approach to enhance data privacy is to return a noisy version of the actual count. It is typically achieved by adding independent noise to each query and then control the total privacy budget within a period. This approach may be limited in the number of queries and output accuracy in practice. Also, the returned counts do not maintain the total order for nested queries, an important feature in many applications. This work presents the design and analysis of a new method, sigma-counting, that addresses these challenges. Sigma-counting uses the notion of sigma-algebra to construct privacy-preserving counting queries. We show that the proposed concepts and methods can significantly improve output accuracy while maintaining a desired privacy level in the presence of massive queries to the same data. We also discuss how the technique can be applied to address large and time-varying datasets.
Authors: Mu Nan, Taohui Xiao, Ruoyou Wu, Shoujun Yu, Ye Li, Hairong Zheng, Shanshan Wang
Abstract: Diffusion MRI (dMRI) angular super-resolution (ASR) aims to reconstruct high-angular-resolution (HAR) signals from limited low-angular-resolution (LAR) data without prolonging scan time. However, existing methods are limited in recovering fine-grained angular details or preserving high fidelity due to inadequate modeling of q-space geometry and insufficient incorporation of physical constraints. In this paper, we introduce a Physics-Guided Diffusion Transformer (PGDiT) designed to explore physical priors throughout both training and inference stages. During training, a Q-space Geometry-Aware Module (QGAM) with b-vector modulation and random angular masking facilitates direction-aware representation learning, enabling the network to generate directionally consistent reconstructions with fine angular details from sparse and noisy data. In inference, a two-stage Spherical Harmonics-Guided Posterior Sampling (SHPS) enforces alignment with the acquired data, followed by heat-diffusion-based SH regularization to ensure physically plausible reconstructions. This coarse-to-fine refinement strategy mitigates oversmoothing and artifacts commonly observed in purely data-driven or generative models. Extensive experiments on general ASR tasks and two downstream applications, Diffusion Tensor Imaging (DTI) and Neurite Orientation Dispersion and Density Imaging (NODDI), demonstrate that PGDiT outperforms existing deep learning models in detail recovery and data fidelity. Our approach presents a novel generative ASR framework that offers high-fidelity HAR dMRI reconstructions, with potential applications in neuroscience and clinical research.
Authors: Yadi Cao, Futian Zhang, Wesley Liu, Tom Neiser, Orso Meneghini, Lawson Fuller, Sterling Smith, Raffi Nazikian, Brian Sammuli, Rose Yu
Abstract: The Trapped Gyro-Landau Fluid (TGLF) model provides fast, accurate predictions of turbulent transport in tokamaks, but whole device simulations requiring thousands of evaluations remain computationally expensive. Neural network (NN) surrogates offer accelerated inference with fully differentiable approximations that enable gradient-based coupling but typically require large training datasets to capture transport flux variations across plasma conditions, creating significant training burden and limiting applicability to expensive gyrokinetic simulations. We propose \textbf{TGLF-SINN (Spectra-Informed Neural Network)} with three key innovations: (1) principled feature engineering that reduces target prediction range, simplifying the learning task; (2) physics-guided regularization of transport spectra to improve generalization under sparse data; and (3) Bayesian Active Learning (BAL) to strategically select training samples based on model uncertainty, reducing data requirements while maintaining accuracy. Our approach achieves superior performance with significantly less training data. In offline settings, TGLF-SINN reduces logarithmic root mean squared error (LRMSE) by 12. 4\% compared to the current baseline \base. Using only 25\% of the complete dataset with BAL, we achieve LRMSE only 0.0165 higher than \base~and 0.0248 higher than our offline model (0.0583). In downstream flux matching applications, our NN surrogate provides 45x speedup over TGLF while maintaining comparable accuracy, demonstrating potential for training efficient surrogates for higher-fidelity models where data acquisition is costly and sparse.
Authors: Jisung Hwang, Jaihoon Kim, Minhyuk Sung
Abstract: We propose a novel regularization loss that enforces standard Gaussianity, encouraging samples to align with a standard Gaussian distribution. This facilitates a range of downstream tasks involving optimization in the latent space of text-to-image models. We treat elements of a high-dimensional sample as one-dimensional standard Gaussian variables and define a composite loss that combines moment-based regularization in the spatial domain with power spectrum-based regularization in the spectral domain. Since the expected values of moments and power spectrum distributions are analytically known, the loss promotes conformity to these properties. To ensure permutation invariance, the losses are applied to randomly permuted inputs. Notably, existing Gaussianity-based regularizations fall within our unified framework: some correspond to moment losses of specific orders, while the previous covariance-matching loss is equivalent to our spectral loss but incurs higher time complexity due to its spatial-domain computation. We showcase the application of our regularization in generative modeling for test-time reward alignment with a text-to-image model, specifically to enhance aesthetics and text alignment. Our regularization outperforms previous Gaussianity regularization, effectively prevents reward hacking and accelerates convergence.
Authors: Neeshu Rathi, Sanjeev Kumar
Abstract: The development of noise-resilient quantum machine learning (QML) algorithms is critical in the noisy intermediate-scale quantum (NISQ) era. In this work, we propose a quantum bagging framework that uses QMeans clustering as the base learner to reduce prediction variance and enhance robustness to label noise. Unlike bagging frameworks built on supervised learners, our method leverages the unsupervised nature of QMeans, combined with quantum bootstrapping via QRAM-based sampling and bagging aggregation through majority voting. Through extensive simulations on both noisy classification and regression tasks, we demonstrate that the proposed quantum bagging algorithm performs comparably to its classical counterpart using KMeans while exhibiting greater resilience to label corruption than supervised bagging methods. This highlights the potential of unsupervised quantum bagging in learning from unreliable data.
Authors: Diego Fajardo-Rojas, Levente Baljer, Jordina Aviles Verdera, Megan Hall, Daniel Cromb, Mary A. Rutherford, Lisa Story, Emma C. Robinson, Jana Hutter
Abstract: Preterm birth is a major cause of mortality and lifelong morbidity in childhood. Its complex and multifactorial origins limit the effectiveness of current clinical predictors and impede optimal care. In this study, a dual-branch deep learning architecture (PUUMA) was developed to predict gestational age (GA) at birth using T2* fetal MRI data from 295 pregnancies, encompassing a heterogeneous and imbalanced population. The model integrates both global whole-uterus and local placental features. Its performance was benchmarked against linear regression using cervical length measurements obtained by experienced clinicians from anatomical MRI and other Deep Learning architectures. The GA at birth predictions were assessed using mean absolute error. Accuracy, sensitivity, and specificity were used to assess preterm classification. Both the fully automated MRI-based pipeline and the cervical length regression achieved comparable mean absolute errors (3 weeks) and good sensitivity (0.67) for detecting preterm birth, despite pronounced class imbalance in the dataset. These results provide a proof of concept for automated prediction of GA at birth from functional MRI, and underscore the value of whole-uterus functional imaging in identifying at-risk pregnancies. Additionally, we demonstrate that manual, high-definition cervical length measurements derived from MRI, not currently routine in clinical practice, offer valuable predictive information. Future work will focus on expanding the cohort size and incorporating additional organ-specific imaging to improve generalisability and predictive performance.
Authors: Kamyar Barakati, Utkarsh Pratiush, Sheryl L. Sanchez, Aditya Raghavan, Delia J. Milliron, Mahshid Ahmadi, Philip D. Rack, Sergei V. Kalinin
Abstract: Image segmentation is a critical task in microscopy, essential for accurately analyzing and interpreting complex visual data. This task can be performed using custom models trained on domain-specific datasets, transfer learning from pre-trained models, or foundational models that offer broad applicability. However, foundational models often present a considerable number of non-transparent tuning parameters that require extensive manual optimization, limiting their usability for real-time streaming data analysis. Here, we introduce a reward function-based optimization to fine-tune foundational models and illustrate this approach for SAM (Segment Anything Model) framework by Meta. The reward functions can be constructed to represent the physics of the imaged system, including particle size distributions, geometries, and other criteria. By integrating a reward-driven optimization framework, we enhance SAM's adaptability and performance, leading to an optimized variant, SAM$^{*}$, that better aligns with the requirements of diverse segmentation tasks and particularly allows for real-time streaming data segmentation. We demonstrate the effectiveness of this approach in microscopy imaging, where precise segmentation is crucial for analyzing cellular structures, material interfaces, and nanoscale features.
Authors: Pietro Bartoli, Tommaso Bondini, Christian Veronesi, Andrea Giudici, Niccol\`o Antonello, Franco Zappa
Abstract: Keyword spotting (KWS) is a key enabling technology for hands-free interaction in embedded and IoT devices, where stringent memory and energy constraints challenge the deployment of AI-enabeld devices. In this work, we systematically evaluate and compare several state-of-the-art lightweight neural network architectures, including DS-CNN, LiCoNet, and TENet, alongside our proposed Typman-KWS (TKWS) architecture built upon MobileNet, specifically designed for efficient KWS on microcontroller units (MCUs). Unlike prior studies focused solely on model inference, our analysis encompasses the entire processing pipeline, from Mel-Frequency Cepstral Coefficient (MFCC) feature extraction to neural inference, and is benchmarked across three STM32 platforms (N6, H7, and U5). Our results show that TKWS with three residual blocks achieves up to 92.4% F1-score with only 14.4k parameters, reducing memory footprint without compromising the accuracy. Moreover, the N6 MCU with integrated neural acceleration achieves the best energy-delay product (EDP), enabling efficient, low-latency operation even with high-resolution features. Our findings highlight the model accuracy alone does not determine real-world effectiveness; rather, optimal keyword spotting deployments require careful consideration of feature extraction parameters and hardware-specific optimization.
Authors: Edgar Dobriban
Abstract: Generative Artificial Intelligence is emerging as an important technology, promising to be transformative in many areas. At the same time, generative AI techniques are based on sampling from probabilistic models, and by default, they come with no guarantees about correctness, safety, fairness, or other properties. Statistical methods offer a promising potential approach to improve the reliability of generative AI techniques. In addition, statistical methods are also promising for improving the quality and efficiency of AI evaluation, as well as for designing interventions and experiments in AI. In this paper, we review some of the existing work on these topics, explaining both the general statistical techniques used, as well as their applications to generative AI. We also discuss limitations and potential future directions.
Authors: Tom\'as Gonz\'alez, Mateo Dulce-Rubio, Aaditya Ramdas, M\'onica Ribero
Abstract: We propose a practical sequential test for auditing differential privacy guarantees of black-box mechanisms. The test processes streams of mechanisms' outputs providing anytime-valid inference while controlling Type I error, overcoming the fixed sample size limitation of previous batch auditing methods. Experiments show this test detects violations with sample sizes that are orders of magnitude smaller than existing methods, reducing this number from 50K to a few hundred examples, across diverse realistic mechanisms. Notably, it identifies DP-SGD privacy violations in \textit{under} one training run, unlike prior methods needing full model training.
Authors: Mert Ketenci, Vincent Jeanselme, Harry Reyes Nieva, Shalmali Joshi, No\'emie Elhadad
Abstract: Survival analysis is a fundamental tool for modeling time-to-event outcomes in healthcare. Recent advances have introduced flexible neural network approaches for improved predictive performance. However, most of these models do not provide interpretable insights into the association between exposures and the modeled outcomes, a critical requirement for decision-making in clinical practice. To address this limitation, we propose Additive Deep Hazard Analysis Mixtures (ADHAM), an interpretable additive survival model. ADHAM assumes a conditional latent structure that defines subgroups, each characterized by a combination of covariate-specific hazard functions. To select the number of subgroups, we introduce a post-training refinement that reduces the number of equivalent latent subgroups by merging similar groups. We perform comprehensive studies to demonstrate ADHAM's interpretability at the population, subgroup, and individual levels. Extensive experiments on real-world datasets show that ADHAM provides novel insights into the association between exposures and outcomes. Further, ADHAM remains on par with existing state-of-the-art survival baselines in terms of predictive performance, offering a scalable and interpretable approach to time-to-event prediction in healthcare.
Authors: Yuqi Zhou, Zhanhong Cheng, Lingqian Hu, Yuheng Bu, Shenhao Wang
Abstract: Nested logit (NL) has been commonly used for discrete choice analysis, including a wide range of applications such as travel mode choice, automobile ownership, or location decisions. However, the classical NL models are restricted by their limited representation capability and handcrafted utility specification. While researchers introduced deep neural networks (DNNs) to tackle such challenges, the existing DNNs cannot explicitly capture inter-alternative correlations in the discrete choice context. To address the challenges, this study proposes a novel concept - alternative graph - to represent the relationships among travel mode alternatives. Using a nested alternative graph, this study further designs a nested-utility graph neural network (NestGNN) as a generalization of the classical NL model in the neural network family. Theoretically, NestGNNs generalize the classical NL models and existing DNNs in terms of model representation, while retaining the crucial two-layer substitution patterns of the NL models: proportional substitution within a nest but non-proportional substitution beyond a nest. Empirically, we find that the NestGNNs significantly outperform the benchmark models, particularly the corresponding NL models by 9.2\%. As shown by elasticity tables and substitution visualization, NestGNNs retain the two-layer substitution patterns as the NL model, and yet presents more flexibility in its model design space. Overall, our study demonstrates the power of NestGNN in prediction, interpretation, and its flexibility of generalizing the classical NL model for analyzing travel mode choice.
Authors: Kutub Uddin, Muhammad Umar Farooq, Awais Khan, Khalid Mahmood Malik
Abstract: The widespread use of generative AI has shown remarkable success in producing highly realistic deepfakes, posing a serious threat to various voice biometric applications, including speaker verification, voice biometrics, audio conferencing, and criminal investigations. To counteract this, several state-of-the-art (SoTA) audio deepfake detection (ADD) methods have been proposed to identify generative AI signatures to distinguish between real and deepfake audio. However, the effectiveness of these methods is severely undermined by anti-forensic (AF) attacks that conceal generative signatures. These AF attacks span a wide range of techniques, including statistical modifications (e.g., pitch shifting, filtering, noise addition, and quantization) and optimization-based attacks (e.g., FGSM, PGD, C \& W, and DeepFool). In this paper, we investigate the SoTA ADD methods and provide a comparative analysis to highlight their effectiveness in exposing deepfake signatures, as well as their vulnerabilities under adversarial conditions. We conducted an extensive evaluation of ADD methods on five deepfake benchmark datasets using two categories: raw and spectrogram-based approaches. This comparative analysis enables a deeper understanding of the strengths and limitations of SoTA ADD methods against diverse AF attacks. It does not only highlight vulnerabilities of ADD methods, but also informs the design of more robust and generalized detectors for real-world voice biometrics. It will further guide future research in developing adaptive defense strategies that can effectively counter evolving AF techniques.
Authors: Fernando Spadea, Oshani Seneviratne
Abstract: We present a lightweight neuro-symbolic framework to mitigate over-personalization in LLM-based recommender systems by adapting user-side Knowledge Graphs (KGs) at inference time. Instead of retraining models or relying on opaque heuristics, our method restructures a user's Personalized Knowledge Graph (PKG) to suppress feature co-occurrence patterns that reinforce Personalized Information Environments (PIEs), i.e., algorithmically induced filter bubbles that constrain content diversity. These adapted PKGs are used to construct structured prompts that steer the language model toward more diverse, Out-PIE recommendations while preserving topical relevance. We introduce a family of symbolic adaptation strategies, including soft reweighting, hard inversion, and targeted removal of biased triples, and a client-side learning algorithm that optimizes their application per user. Experiments on a recipe recommendation benchmark show that personalized PKG adaptations significantly increase content novelty while maintaining recommendation quality, outperforming global adaptation and naive prompt-based methods.
Authors: Haike Xu, Tong Chen
Abstract: The widely used retrieve-and-rerank pipeline faces two critical limitations: they are constrained by the initial retrieval quality of the top-k documents, and the growing computational demands of LLM-based rerankers restrict the number of documents that can be effectively processed. We introduce Reranker-Guided-Search (RGS), a novel approach that bypasses these limitations by directly retrieving documents according to reranker preferences rather than following the traditional sequential reranking method. Our method uses a greedy search on proximity graphs generated by approximate nearest neighbor algorithms, strategically prioritizing promising documents for reranking based on document similarity. Experimental results demonstrate substantial performance improvements across multiple benchmarks: 3.5 points on BRIGHT, 2.9 on FollowIR, and 5.1 on M-BEIR, all within a constrained reranker budget of 100 documents. Our analysis suggests that, given a fixed pair of embedding and reranker models, strategically selecting documents to rerank can significantly improve retrieval accuracy under limited reranker budget.
Authors: Erencem Ozbey, Dimitrios I. Diochnos
Abstract: Working with annotated data is the cornerstone of supervised learning. Nevertheless, providing labels to instances is a task that requires significant human effort. Several critical real-world applications make things more complicated because no matter how many labels may have been identified in a task of interest, it could be the case that examples corresponding to novel classes may appear in the future. Not unsurprisingly, prior work in this, so-called, `open-world' context has focused a lot on semi-supervised approaches. Focusing on image classification, somehow paradoxically, we propose a fully unsupervised approach to the problem of determining the novel categories in a particular dataset. Our approach relies on estimating the number of clusters using Vision Transformers, which utilize attention mechanisms to generate vector embeddings. Furthermore, we incorporate manifold learning techniques to refine these embeddings by exploiting the intrinsic geometry of the data, thereby enhancing the overall image clustering performance. Overall, we establish new State-of-the-Art results on single-modal clustering and Novel Class Discovery on CIFAR-10, CIFAR-100, ImageNet-100, and Tiny ImageNet. We do so, both when the number of clusters is known or unknown ahead of time. The code is available at: https://github.com/DROWCULA/DROWCULA.
Authors: Yann Alibert, Jeanne Davoult, Sara Marques
Abstract: Numerical calculations of planetary system formation are very demanding in terms of computing power. These synthetic planetary systems can however provide access to correlations, as predicted in a given numerical framework, between the properties of planets in the same system. Such correlations can, in return, be used in order to guide and prioritize observational campaigns aiming at discovering some types of planets, as Earth-like planets. Our goal is to develop a generative model which is capable of capturing correlations and statistical relationships between planets in the same system. Such a model, trained on the Bern model, offers the possibility to generate large number of synthetic planetary systems with little computational cost, that can be used, for example, to guide observational campaigns. Our generative model is based on the transformer architecture which is well-known to efficiently capture correlations in sequences and is at the basis of all modern Large Language Models. To assess the validity of the generative model, we perform visual and statistical comparisons, as well as a machine learning driven tests. Finally, as a use case example, we consider the TOI-469 system, in which we aim at predicting the possible properties of planets c and d, based on the properties of planet b (the first that has been detected). We show using different comparison methods that the properties of systems generated by our model are very similar to the ones of the systems computed directly by the Bern model. We also show in the case of the TOI-469 system, that using the generative model allows to predict the properties of planets not yet observed, based on the properties of the already observed planet. We provide our model to the community on our website www.ai4exoplanets.com.
Authors: Luigi Troiano, Francesco Gissi, Vincenzo Benedetto, Genny Tortora
Abstract: Gradient-based neural network training traditionally enforces symmetry between forward and backward propagation, requiring activation functions to be differentiable (or sub-differentiable) and strictly monotonic in certain regions to prevent flat gradient areas. This symmetry, linking forward activations closely to backward gradients, significantly restricts the selection of activation functions, particularly excluding those with substantial flat or non-differentiable regions. In this paper, we challenge this assumption through mathematical analysis, demonstrating that precise gradient magnitudes derived from activation functions are largely redundant, provided the gradient direction is preserved. Empirical experiments conducted on foundational architectures - such as Multi-Layer Perceptrons (MLPs), Convolutional Neural Networks (CNNs), and Binary Neural Networks (BNNs) - confirm that relaxing forward-backward symmetry and substituting traditional gradients with simpler or stochastic alternatives does not impair learning and may even enhance training stability and efficiency. We explicitly demonstrate that neural networks with flat or non-differentiable activation functions, such as the Heaviside step function, can be effectively trained, thereby expanding design flexibility and computational efficiency. Further empirical validation with more complex architectures remains a valuable direction for future research.
Authors: Xin Wang, Ting Dang, Xinyu Zhang, Vassilis Kostakos, Michael J. Witbrock, Hong Jia
Abstract: Mobile and wearable healthcare monitoring play a vital role in facilitating timely interventions, managing chronic health conditions, and ultimately improving individuals' quality of life. Previous studies on large language models (LLMs) have highlighted their impressive generalization abilities and effectiveness in healthcare prediction tasks. However, most LLM-based healthcare solutions are cloud-based, which raises significant privacy concerns and results in increased memory usage and latency. To address these challenges, there is growing interest in compact models, Small Language Models (SLMs), which are lightweight and designed to run locally and efficiently on mobile and wearable devices. Nevertheless, how well these models perform in healthcare prediction remains largely unexplored. We systematically evaluated SLMs on health prediction tasks using zero-shot, few-shot, and instruction fine-tuning approaches, and deployed the best performing fine-tuned SLMs on mobile devices to evaluate their real-world efficiency and predictive performance in practical healthcare scenarios. Our results show that SLMs can achieve performance comparable to LLMs while offering substantial gains in efficiency and privacy. However, challenges remain, particularly in handling class imbalance and few-shot scenarios. These findings highlight SLMs, though imperfect in their current form, as a promising solution for next-generation, privacy-preserving healthcare monitoring.
Authors: Aida Kostikova, Ole P\"utz, Steffen Eger, Olga Sabelfeld, Benjamin Paassen
Abstract: Migration has been a core topic in German political debate, from millions of expellees post World War II over labor migration to refugee movements in the recent past. Studying political speech regarding such wide-ranging phenomena in depth traditionally required extensive manual annotations, limiting the scope of analysis to small subsets of the data. Large language models (LLMs) have the potential to partially automate even complex annotation tasks. We provide an extensive evaluation of a multiple LLMs in annotating (anti-)solidarity subtypes in German parliamentary debates compared to a large set of thousands of human reference annotations (gathered over a year). We evaluate the influence of model size, prompting differences, fine-tuning, historical versus contemporary data; and we investigate systematic errors. Beyond methodological evaluation, we also interpret the resulting annotations from a social science lense, gaining deeper insight into (anti-)solidarity trends towards migrants in the German post-World War II period and recent past. Our data reveals a high degree of migrant-directed solidarity in the postwar period, as well as a strong trend towards anti-solidarity in the German parliament since 2015, motivating further research. These findings highlight the promise of LLMs for political text analysis and the importance of migration debates in Germany, where demographic decline and labor shortages coexist with rising polarization.
Authors: M. Hadi Sepanj, Benyamin Ghojogh, Paul Fieguth
Abstract: Self-supervised learning (SSL) has emerged as a powerful paradigm for representation learning by optimizing geometric objectives--such as invariance to augmentations, variance preservation, and feature decorrelation--without requiring labels. However, most existing methods operate in Euclidean space, limiting their ability to capture nonlinear dependencies and geometric structures. In this work, we propose Kernel VICReg, a novel self-supervised learning framework that lifts the VICReg objective into a Reproducing Kernel Hilbert Space (RKHS). By kernelizing each term of the loss-variance, invariance, and covariance--we obtain a general formulation that operates on double-centered kernel matrices and Hilbert-Schmidt norms, enabling nonlinear feature learning without explicit mappings. We demonstrate that Kernel VICReg not only avoids representational collapse but also improves performance on tasks with complex or small-scale data. Empirical evaluations across MNIST, CIFAR-10, STL-10, TinyImageNet, and ImageNet100 show consistent gains over Euclidean VICReg, with particularly strong improvements on datasets where nonlinear structures are prominent. UMAP visualizations further confirm that kernel-based embeddings exhibit better isometry and class separation. Our results suggest that kernelizing SSL objectives is a promising direction for bridging classical kernel methods with modern representation learning.
Authors: Ji Xie, Trevor Darrell, Luke Zettlemoyer, XuDong Wang
Abstract: Unified multimodal models (UMMs) unify visual understanding and generation within a single architecture. However, conventional training relies on image-text pairs (or sequences) whose captions are typically sparse and miss fine-grained visual details--even when they use hundreds of words to describe a simple image. We introduce Reconstruction Alignment (RecA), a resource-efficient post-training method that leverages visual understanding encoder embeddings as dense "text prompts," providing rich supervision without captions. Concretely, RecA conditions a UMM on its own visual understanding embeddings and optimizes it to reconstruct the input image with a self-supervised reconstruction loss, thereby realigning understanding and generation. Despite its simplicity, RecA is broadly applicable: across autoregressive, masked-autoregressive, and diffusion-based UMMs, it consistently improves generation and editing fidelity. With only 27 GPU-hours, post-training with RecA substantially improves image generation performance on GenEval (0.73$\rightarrow$0.90) and DPGBench (80.93$\rightarrow$88.15), while also boosting editing benchmarks (ImgEdit 3.38$\rightarrow$3.75, GEdit 6.94$\rightarrow$7.25). Notably, RecA surpasses much larger open-source models and applies broadly across diverse UMM architectures, establishing it as an efficient and general post-training alignment strategy for UMMs
Authors: Jared Rieck, Julia Wrobel, Joshua L. Gowin, Yue Wang, Martin Paulus, Ryan Peterson
Abstract: Recent advances in neuroimaging analysis have enabled accurate decoding of mental state from brain activation patterns during functional magnetic resonance imaging scans. A commonly applied tool for this purpose is principal components regression regularized with the least absolute shrinkage and selection operator (LASSO PCR), a type of multi-voxel pattern analysis (MVPA). This model presumes that all components are equally likely to harbor relevant information, when in fact the task-related signal may be concentrated in specific components. In such cases, the model will fail to select the optimal set of principal components that maximizes the total signal relevant to the cognitive process under study. Here, we present modifications to LASSO PCR that allow for a regularization penalty tied directly to the index of the principal component, reflecting a prior belief that task-relevant signal is more likely to be concentrated in components explaining greater variance. Additionally, we propose a novel hybrid method, Joint Sparsity-Ranked LASSO (JSRL), which integrates component-level and voxel-level activity under an information parity framework and imposes ranked sparsity to guide component selection. We apply the models to brain activation during risk taking, monetary incentive, and emotion regulation tasks. Results demonstrate that incorporating sparsity ranking into LASSO PCR produces models with enhanced classification performance, with JSRL achieving up to 51.7\% improvement in cross-validated deviance $R^2$ and 7.3\% improvement in cross-validated AUC. Furthermore, sparsity-ranked models perform as well as or better than standard LASSO PCR approaches across all classification tasks and allocate predictive weight to brain regions consistent with their established functional roles, offering a robust alternative for MVPA.
Authors: Zhuoqing Song, Peng Sun, Huizhuo Yuan, Quanquan Gu
Abstract: In standard causal attention, each token's query, key, and value (QKV) are static and encode only preceding context. We introduce CAuSal aTtention with Lookahead kEys (CASTLE), an attention mechanism that continually updates each token's keys as the context unfolds. We term these updated keys lookahead keys because they belong to earlier positions yet integrate information from tokens that appear later relative to those positions, while strictly preserving the autoregressive property. Although the mechanism appears sequential, we derive a mathematical equivalence that avoids explicitly materializing lookahead keys at each position and enables efficient parallel training. On language modeling benchmarks, CASTLE consistently outperforms standard causal attention across model scales, reducing validation perplexity and improving performance on a range of downstream tasks.
Authors: Chi-Yang Hsu, Alexander Braylan, Yiheng Su, Omar Alonso, Matthew Lease
Abstract: We motivate and share a new benchmark for instance-level performance prediction of long-form generation tasks having multi-faceted, fine-grained quality metrics. Our task-, model- and metric-agnostic formulation predicts continuous evaluation metric scores given only black-box model inputs and outputs. Beyond predicting point estimates of metric scores, the benchmark also requires inferring prediction intervals to quantify uncertainty around point estimates. Evaluation spans 11 long-form datasets/tasks with multiple LLMs, baselines, and metrics per task. We show that scores can be effectively predicted across long-form generation tasks using as few as 16 training examples. Overall, we introduce a novel and useful task, a valuable benchmark to drive progress, and baselines ready for practical adoption today.
Authors: Cunxi Yu, Rongjian Liang, Chia-Tung Ho, Haoxing Ren
Abstract: Large language models (LLMs) have recently shown strong coding abilities, enabling not only static code generation but also iterative code self-evolving through agentic frameworks. Recently, AlphaEvolve \cite{novikov2025alphaevolve} demonstrated that LLM-based coding agents can autonomously improve algorithms and surpass human experts, with scopes limited to isolated kernels spanning hundreds of lines of code. Inspired by AlphaEvolve, we present SATLUTION, the first framework to extend LLM-based code evolution to the full repository scale, encompassing hundreds of files and tens of thousands of lines of C/C++ code. Targeting Boolean Satisfiability (SAT), the canonical NP-complete problem and a cornerstone of both theory and applications. SATLUTION orchestrates LLM agents to directly evolve solver repositories under strict correctness guarantees and distributed runtime feedback, while simultaneously self-evolving its own evolution policies and rules. Starting from SAT Competition 2024 codebases and benchmark, SATLUTION evolved solvers that decisively outperformed the human-designed winners of the SAT Competition 2025, and also surpassed both 2024 and 2025 champions on the 2024 benchmarks.
Authors: Sankalp Tattwadarshi Swain, Anshika Krishnatray, Dhruv Kumar, Jagat Sesh Challa
Abstract: Existing evaluation studies on linguistic competence of large language models (LLM agents) have focused primarily on vocabulary learning, morphological rule induction, syntactic generalization, pragmatic inference, and cross-linguistic transfer. However, none assess whether LLM agents can acquire a language through pattern recognition and interactive feedback, a central feature of human language acquisition. We propose a novel experimental framework in which an LLM agent is evaluated on its ability to acquire and use a newly constructed language (Tinkatongue) in conversation with a bot that understands only Tinkatongue. Our findings show that LLM agents fail to establish a conversation within 100 responses, yet they adopt distinct strategies that mirror human approaches to language learning. The results suggest a new direction for evaluation benchmarks and open pathways to model designs that learn more effectively from interactive feedback.
Authors: Jeremy Bertoncini, Alberto De Marchi, Matthias Gerdts, Simon Gottschalk
Abstract: Quadratic programming is a workhorse of modern nonlinear optimization, control, and data science. Although regularized methods offer convergence guarantees under minimal assumptions on the problem data, they can exhibit the slow tail-convergence typical of first-order schemes, thus requiring many iterations to achieve high-accuracy solutions. Moreover, hyperparameter tuning significantly impacts on the solver performance but how to find an appropriate parameter configuration remains an elusive research question. To address these issues, we explore how data-driven approaches can accelerate the solution process. Aiming at high-accuracy solutions, we focus on a stabilized interior-point solver and carefully handle its two-loop flow and control parameters. We will show that reinforcement learning can make a significant contribution to facilitating the solver tuning and to speeding up the optimization process. Numerical experiments demonstrate that, after a lightweight training, the learned policy generalizes well to different problem classes with varying dimensions and to various solver configurations.
Authors: Laura Boggia, Bogdan Malaescu
Abstract: Anomaly detection in multivariate time series is crucial to ensure the quality of data coming from a physics experiment. Accurately identifying the moments when unexpected errors or defects occur is essential, yet challenging due to scarce labels, unknown anomaly types, and complex correlations across dimensions. To address the scarcity and unreliability of labelled data, we use the Lorenzetti Simulator to generate synthetic events with injected calorimeter anomalies. We then assess the sensitivity of several time series anomaly detection methods, including transformer-based and other deep learning models. The approach employed here is generic and applicable to different detector designs and defects.
Authors: Patrick Wienholt, Christiane Kuhl, Jakob Nikolas Kather, Sven Nebelung, Daniel Truhn
Abstract: Deep neural networks excel in radiological image classification but frequently suffer from poor interpretability, limiting clinical acceptance. We present MedicalPatchNet, an inherently self-explainable architecture for chest X-ray classification that transparently attributes decisions to distinct image regions. MedicalPatchNet splits images into non-overlapping patches, independently classifies each patch, and aggregates predictions, enabling intuitive visualization of each patch's diagnostic contribution without post-hoc techniques. Trained on the CheXpert dataset (223,414 images), MedicalPatchNet matches the classification performance (AUROC 0.907 vs. 0.908) of EfficientNet-B0, while substantially improving interpretability: MedicalPatchNet demonstrates substantially improved interpretability with higher pathology localization accuracy (mean hit-rate 0.485 vs. 0.376 with Grad-CAM) on the CheXlocalize dataset. By providing explicit, reliable explanations accessible even to non-AI experts, MedicalPatchNet mitigates risks associated with shortcut learning, thus improving clinical trust. Our model is publicly available with reproducible training and inference scripts and contributes to safer, explainable AI-assisted diagnostics across medical imaging domains. We make the code publicly available: https://github.com/TruhnLab/MedicalPatchNet
Authors: Zichun Hao, Raghav Kansal, Abhijith Gandrakota, Chang Sun, Ngadiuba Jennifer, Javier Duarte, Maria Spiropulu
Abstract: A common challenge with supervised machine learning (ML) in high energy physics (HEP) is the reliance on simulations for labeled data, which can often mismodel the underlying collision or detector response. To help mitigate this problem of domain shift, we propose RINO (Renormalization Group Invariance with No Labels), a self-supervised learning approach that can instead pretrain models directly on collision data, learning embeddings invariant to renormalization group flow scales. In this work, we pretrain a transformer-based model on jets originating from quantum chromodynamic (QCD) interactions from the JetClass dataset, emulating real QCD-dominated experimental data, and then finetune on the JetNet dataset -- emulating simulations -- for the task of identifying jets originating from top quark decays. RINO demonstrates improved generalization from the JetNet training data to JetClass data compared to supervised training on JetNet from scratch, demonstrating the potential for RINO pretraining on real collision data followed by fine-tuning on small, high-quality MC datasets, to improve the robustness of ML models in HEP.
Authors: Anjiang Wei, Tianran Sun, Yogesh Seenichamy, Hang Song, Anne Ouyang, Azalia Mirhoseini, Ke Wang, Alex Aiken
Abstract: GPU kernel optimization has long been a central challenge at the intersection of high-performance computing and machine learning. Efficient kernels are crucial for accelerating large language model (LLM) training and serving, yet attaining high performance typically requires extensive manual tuning. Compiler-based systems reduce some of this burden, but still demand substantial manual design and engineering effort. Recently, researchers have explored using LLMs for GPU kernel generation, though prior work has largely focused on translating high-level PyTorch modules into CUDA code. In this work, we introduce Astra, the first LLM-based multi-agent system for GPU kernel optimization. Unlike previous approaches, Astra starts from existing CUDA implementations extracted from SGLang, a widely deployed framework for serving LLMs, rather than treating PyTorch modules as the specification. Within Astra, specialized LLM agents collaborate through iterative code generation, testing, profiling, and planning to produce kernels that are both correct and high-performance. On kernels from SGLang, Astra achieves an average speedup of 1.32x using zero-shot prompting with OpenAI o4-mini. A detailed case study further demonstrates that LLMs can autonomously apply loop transformations, optimize memory access patterns, exploit CUDA intrinsics, and leverage fast math operations to yield substantial performance gains. Our work highlights multi-agent LLM systems as a promising new paradigm for GPU kernel optimization.
Authors: Gokul Karthik Kumar, Rishabh Saraf, Ludovick Lepauloux, Abdul Muneer, Billel Mokeddem, Hakim Hacid
Abstract: Large language models (LLMs) have transformed NLP, yet their integration with audio remains underexplored -- despite audio's centrality to human communication. We introduce Falcon3-Audio, a family of Audio-Language Models (ALMs) built on instruction-tuned LLMs and Whisper encoders. Using a remarkably small amount of public audio data -- less than 30K hours (5K unique) -- Falcon3-Audio-7B matches the best reported performance among open-weight models on the MMAU benchmark, with a score of 64.14, matching R1-AQA, while distinguishing itself through superior data and parameter efficiency, single-stage training, and transparency. Notably, our smallest 1B model remains competitive with larger open models ranging from 2B to 13B parameters. Through extensive ablations, we find that common complexities -- such as curriculum learning, multiple audio encoders, and intricate cross-attention connectors -- are not required for strong performance, even compared to models trained on over 500K hours of data.
Authors: Anna Van Elst, Igor Colin, Stephan Cl\'emen\c{c}on
Abstract: As decentralized AI and edge intelligence become increasingly prevalent, ensuring robustness and trustworthiness in such distributed settings has become a critical issue-especially in the presence of corrupted or adversarial data. Traditional decentralized algorithms are vulnerable to data contamination as they typically rely on simple statistics (e.g., means or sum), motivating the need for more robust statistics. In line with recent work on decentralized estimation of trimmed means and ranks, we develop gossip algorithms for computing a broad class of rank-based statistics, including L-statistics and rank statistics-both known for their robustness to outliers. We apply our method to perform robust distributed two-sample hypothesis testing, introducing the first gossip algorithm for Wilcoxon rank-sum tests. We provide rigorous convergence guarantees, including the first convergence rate bound for asynchronous gossip-based rank estimation. We empirically validate our theoretical results through experiments on diverse network topologies.
Authors: Patricia Hu, Silvan David Peter, Jan Schl\"uter, Gerhard Widmer
Abstract: Advances in neural network design and the availability of large-scale labeled datasets have driven major improvements in piano transcription. Existing approaches target either offline applications, with no restrictions on computational demands, or online transcription, with delays of 128-320 ms. However, most real-time musical applications require latencies below 30 ms. In this work, we investigate whether and how the current state-of-the-art online transcription model can be adapted for real-time piano transcription. Specifically, we eliminate all non-causal processing, and reduce computational load through shared computations across core model components and variations in model size. Additionally, we explore different pre- and postprocessing strategies, and related label encoding schemes, and discuss their suitability for real-time transcription. Evaluating the adaptions on the MAESTRO dataset, we find a drop in transcription accuracy due to strictly causal processing as well as a tradeoff between the preprocessing latency and prediction accuracy. We release our system as a baseline to support researchers in designing models towards minimum latency real-time transcription.
Authors: Paolo Combes, Stefan Weinzierl, Klaus Obermayer
Abstract: Deep learning appears as an appealing solution for Automatic Synthesizer Programming (ASP), which aims to assist musicians and sound designers in programming sound synthesizers. However, integrating software synthesizers into training pipelines is challenging due to their potential non-differentiability. This work tackles this challenge by introducing a method to approximate arbitrary synthesizers. Specifically, we train a neural network to map synthesizer presets onto an audio embedding space derived from a pretrained model. This facilitates the definition of a neural proxy that produces compact yet effective representations, thereby enabling the integration of audio embedding loss into neural-based ASP systems for black-box synthesizers. We evaluate the representations derived by various pretrained audio models in the context of neural-based nASP and assess the effectiveness of several neural network architectures, including feedforward, recurrent, and transformer-based models, in defining neural proxies. We evaluate the proposed method using both synthetic and hand-crafted presets from three popular software synthesizers and assess its performance in a synthesizer sound matching downstream task. While the benefits of the learned representation are nuanced by resource requirements, encouraging results were obtained for all synthesizers, paving the way for future research into the application of synthesizer proxies for neural-based ASP systems.
Authors: Himanshu Singh, A. V. Subramanyam, Shivank Rajput, Mohan Kankanhalli
Abstract: Deep neural networks have exhibited impressive performance in image classification tasks but remain vulnerable to adversarial examples. Standard adversarial training enhances robustness but typically fails to explicitly address inter-class feature overlap, a significant contributor to adversarial susceptibility. In this work, we introduce a novel adversarial training framework that actively mitigates inter-class proximity by projecting out inter-class dependencies from adversarial and clean samples in the feature space. Specifically, our approach first identifies the nearest inter-class neighbors for each adversarial sample and subsequently removes projections onto these neighbors to enforce stronger feature separability. Theoretically, we demonstrate that our proposed logits correction reduces the Lipschitz constant of neural networks, thereby lowering the Rademacher complexity, which directly contributes to improved generalization and robustness. Extensive experiments across standard benchmarks including CIFAR-10, CIFAR-100, and SVHN show that our method demonstrates strong performance that is competitive with leading adversarial training techniques, highlighting significant achievements in both robust and clean accuracy. Our findings reveal the importance of addressing inter-class feature proximity explicitly to bolster adversarial robustness in DNNs.
Authors: Sachit Menon, Ahmet Iscen, Arsha Nagrani, Tobias Weyand, Carl Vondrick, Cordelia Schmid
Abstract: Video understanding has seen significant progress in recent years, with models' performance on perception from short clips continuing to rise. Yet, multiple recent benchmarks, such as LVBench, Neptune, and ActivityNet-RTL, show performance wanes for tasks requiring complex reasoning on videos as queries grow more complex and videos grow longer. In this work, we ask: can existing perception capabilities be leveraged to successfully perform more complex video reasoning? In particular, we develop a large language model agent given access to video modules as subagents or tools. Rather than following a fixed procedure to solve queries as in previous work such as Visual Programming, ViperGPT, and MoReVQA, the agent uses the results of each call to a module to determine subsequent steps. Inspired by work in the textual reasoning domain, we introduce a critic to distinguish between instances of successful and unsuccessful sequences from the agent. We show that the combination of our agent and critic achieve strong performance on the previously-mentioned datasets.
Authors: Sergio Chibbaro, Cyril Furtlehner, Th\'eo Marchetta, Andrei-Tiberiu Pantea, Davide Rossetti
Abstract: Causal relationships play a fundamental role in understanding the world around us. The ability to identify and understand cause-effect relationships is critical to making informed decisions, predicting outcomes, and developing effective strategies. However, deciphering causal relationships from observational data is a difficult task, as correlations alone may not provide definitive evidence of causality. In recent years, the field of machine learning (ML) has emerged as a powerful tool, offering new opportunities for uncovering hidden causal mechanisms and better understanding complex systems. In this work, we address the issue of detecting the intrinsic causal links of a large class of complex systems in the framework of the response theory in physics. We develop some theoretical ideas put forward by [1], and technically we use state-of-the-art ML techniques to build up models from data. We consider both linear stochastic and non-linear systems. Finally, we compute the asymptotic efficiency of the linear response based causal predictor in a case of large scale Markov process network of linear interactions.
Authors: Bo Yu, Zhixiu Hua, Bo Zhao
Abstract: Background: Parkinson's disease remains a major neurodegenerative disorder with high misdiagnosis rates, primarily due to reliance on clinical rating scales. Recent studies have demonstrated a strong association between gut microbiota and Parkinson's disease, suggesting that microbial composition may serve as a promising biomarker. Although deep learning models based ongut microbiota show potential for early prediction, most approaches rely on single classifiers and often overlook inter-strain correlations or temporal dynamics. Therefore, there is an urgent need for more robust feature extraction methods tailored to microbiome data. Methods: We proposed BDPM (A Machine Learning-Based Feature Extractor for Parkinson's Disease Classification via Gut Microbiota Analysis). First, we collected gut microbiota profiles from 39 Parkinson's patients and their healthy spouses to identify differentially abundant taxa. Second, we developed an innovative feature selection framework named RFRE (Random Forest combined with Recursive Feature Elimination), integrating ecological knowledge to enhance biological interpretability. Finally, we designed a hybrid classification model to capture temporal and spatial patterns in microbiome data.
Authors: Friedrich Wolf-Monheim
Abstract: Next to decision tree and k-nearest neighbours algorithms deep convolutional neural networks (CNNs) are widely used to classify audio data in many domains like music, speech or environmental sounds. To train a specific CNN various spectral and rhythm features like mel-scaled spectrograms, mel-frequency cepstral coefficients (MFCC), cyclic tempograms, short-time Fourier transform (STFT) chromagrams, constant-Q transform (CQT) chromagrams and chroma energy normalized statistics (CENS) chromagrams can be used as digital image input data for the neural network. The performance of these spectral and rhythm features for audio category level as well as audio class level classification is investigated in detail with a deep CNN and the ESC-50 dataset with 2,000 labeled environmental audio recordings using an end-to-end deep learning pipeline. The evaluated metrics accuracy, precision, recall and F1 score for multiclass classification clearly show that the mel-scaled spectrograms and the mel-frequency cepstral coefficients (MFCC) perform significantly better then the other spectral and rhythm features investigated in this research for audio classification tasks using deep CNNs.
Authors: Shivam Sharma, Supreeth Mysore Venkatesh, Pushkin Kachroo
Abstract: Clustering financial assets based on return correlations is a fundamental task in portfolio optimization and statistical arbitrage. However, classical clustering methods often fall short when dealing with signed correlation structures, typically requiring lossy transformations and heuristic assumptions such as a fixed number of clusters. In this work, we apply the Graph-based Coalition Structure Generation algorithm (GCS-Q) to directly cluster signed, weighted graphs without relying on such transformations. GCS-Q formulates each partitioning step as a QUBO problem, enabling it to leverage quantum annealing for efficient exploration of exponentially large solution spaces. We validate our approach on both synthetic and real-world financial data, benchmarking against state-of-the-art classical algorithms such as SPONGE and k-Medoids. Our experiments demonstrate that GCS-Q consistently achieves higher clustering quality, as measured by Adjusted Rand Index and structural balance penalties, while dynamically determining the number of clusters. These results highlight the practical utility of near-term quantum computing for graph-based unsupervised learning in financial applications.
Authors: Sebastian Macaluso, Giovanni Geraci, El\'ias F. Combarro, Sergi Abadal, Ioannis Arapakis, Sofia Vallecorsa, Eduard Alarc\'on
Abstract: The complexity of large-scale 6G-and-beyond networks demands innovative approaches for multi-objective optimization over vast search spaces, a task often intractable. Quantum computing (QC) emerges as a promising technology for efficient large-scale optimization. We present our vision of leveraging QC to tackle key classes of problems in future mobile networks. By analyzing and identifying common features, particularly their graph-centric representation, we propose a unified strategy involving QC algorithms. Specifically, we outline a methodology for optimization using quantum annealing as well as quantum reinforcement learning. Additionally, we discuss the main challenges that QC algorithms and hardware must overcome to effectively optimize future networks.
Authors: Emre Sahinoglu, Shahin Shahrampour
Abstract: We study decentralized online Riemannian optimization over manifolds with possibly positive curvature, going beyond the Hadamard manifold setting. Decentralized optimization techniques rely on a consensus step that is well understood in Euclidean spaces because of their linearity. However, in positively curved Riemannian spaces, a main technical challenge is that geodesic distances may not induce a globally convex structure. In this work, we first analyze a curvature-aware Riemannian consensus step that enables a linear convergence beyond Hadamard manifolds. Building on this step, we establish a $O(\sqrt{T})$ regret bound for the decentralized online Riemannian gradient descent algorithm. Then, we investigate the two-point bandit feedback setup, where we employ computationally efficient gradient estimators using smoothing techniques, and we demonstrate the same $O(\sqrt{T})$ regret bound through the subconvexity analysis of smoothed objectives.
Authors: Christopher Brady (North Carolina State University), Xu Wu (North Carolina State University)
Abstract: The Organization for Economic Cooperation and Development (OECD) Working Party on Nuclear Criticality Safety (WPNCS) proposed a benchmark exercise to assess the performance of current nuclear data adjustment techniques applied to nonlinear applications and experiments with low correlation to applications. This work introduces Bayesian Inverse Uncertainty Quantification (IUQ) as a method for nuclear data adjustments in this benchmark, and compares IUQ to the more traditional methods of Generalized Linear Least Squares (GLLS) and Monte Carlo Bayes (MOCABA). Posterior predictions from IUQ showed agreement with GLLS and MOCABA for linear applications. When comparing GLLS, MOCABA, and IUQ posterior predictions to computed model responses using adjusted parameters, we observe that GLLS predictions fail to replicate computed response distributions for nonlinear applications, while MOCABA shows near agreement, and IUQ uses computed model responses directly. We also discuss observations on why experiments with low correlation to applications can be informative to nuclear data adjustments and identify some properties useful in selecting experiments for inclusion in nuclear data adjustment. Performance in this benchmark indicates potential for Bayesian IUQ in nuclear data adjustments.
Authors: Mihai Nadas, Laura Diosan, Andreea Tomescu, Andrei Piscoran
Abstract: Literary translation has recently gained attention as a distinct and complex task in machine translation research. However, the translation by small open models remains an open problem. We contribute to this ongoing research by introducing TINYFABULIST TRANSLATION FRAMEWORK (TF2), a unified framework for dataset creation, fine tuning, and evaluation in English-Romanian literary translations, centred on the creation and open release of both a compact, fine tuned language model (TF2-12B) and large scale synthetic parallel datasets (DS-TF2-EN-RO-3M and DS-TF2-EN-RO-15K). Building on DS-TF1-EN-3M (TF1), the largest collection of synthetic English fables to date, we address the need for rich, high quality literary datasets in low resource languages such as Romanian. Our pipeline first generates 15k high quality Romanian references from the TF1 pool using a high performing LLM. We then apply a two stage fine tuning process to a 12B parameter open weight model: (i) instruction tuning to capture genre specific narrative style, and (ii) adapter compression for efficient deployment. Evaluation combines corpus level BLEU and a five dimension LLM based rubric (accuracy, fluency, coherence, style, cultural adaptation) to provide a nuanced assessment of translation quality. Results show that our fine tuned model achieves fluency and adequacy competitive with top performing large proprietary models, while being open, accessible, and significantly more cost effective. Alongside the fine tuned model and both datasets, we publicly release all scripts and evaluation prompts. TF2 thus provides an end-to-end, reproducible pipeline for research on cost efficient translation, cross lingual narrative generation, and the broad adoption of open models for culturally significant literary content in low resource settings.
Authors: Tuo Wang, Adithya Kulkarni, Tyler Cody, Peter A. Beling, Yujun Yan, Dawei Zhou
Abstract: Uncertainty estimation is essential for enhancing the reliability of Large Language Models (LLMs), particularly in high-stakes applications. Existing methods often overlook semantic dependencies, relying on token-level probability measures that fail to capture structural relationships within the generated text. We propose GENUINE: Graph ENhanced mUlti-level uncertaINty Estimation for Large Language Models, a structure-aware framework that leverages dependency parse trees and hierarchical graph pooling to refine uncertainty quantification. By incorporating supervised learning, GENUINE effectively models semantic and structural relationships, improving confidence assessments. Extensive experiments across NLP tasks show that GENUINE achieves up to 29% higher AUROC than semantic entropy-based approaches and reduces calibration errors by over 15%, demonstrating the effectiveness of graph-based uncertainty modeling. The code is available at https://github.com/ODYSSEYWT/GUQ.
Authors: Mahmudul Islam Masum, Miad Islam, Arif I. Sarwat
Abstract: As local AI grows in popularity, there is a critical gap between the benchmark performance of object detectors and their practical viability on consumer-grade hardware. While models like YOLOv10s promise real-time speeds, these metrics are typically achieved on high-power, desktop-class GPUs. This paper reveals that on resource-constrained systems, such as laptops with RTX 4060 GPUs, performance is not compute-bound but is instead dominated by system-level bottlenecks, as illustrated by a simple bottleneck test. To overcome this hardware-level constraint, we introduce a Two-Pass Adaptive Inference algorithm, a model-independent approach that requires no architectural changes. This study mainly focuses on adaptive inference strategies and undertakes a comparative analysis of architectural early-exit and resolution-adaptive routing, highlighting their respective trade-offs within a unified evaluation framework. The system uses a fast, low-resolution pass and only escalates to a high-resolution model pass when detection confidence is low. On a 5000-image COCO dataset, our method achieves a 1.85x speedup over a PyTorch Early-Exit baseline, with a modest mAP loss of 5.51%. This work provides a practical and reproducible blueprint for deploying high-performance, real-time AI on consumer-grade devices by shifting the focus from pure model optimization to hardware-aware inference strategies that maximize throughput.
Authors: Rohan Garg (Dianxia), Yongjin Xiao (Dianxia), Jason (Dianxia), Yang, Mandar Rahurkar
Abstract: We present a budget pacing feature called Smart Fast Finish (SFF). SFF builds upon the industry standard Fast Finish (FF) feature in budget pacing systems that depletes remaining advertising budget as quickly as possible towards the end of some fixed time period. SFF dynamically updates system parameters such as start time and throttle rate depending on historical ad-campaign data. SFF is currently in use at DoorDash, one of the largest delivery platforms in the US, and is part of its budget pacing system. We show via online budget-split experimentation data and offline simulations that SFF is a robust solution for overdelivery mitigation when pacing budget.
Authors: Katsuaki Nakano, Reza Feyyazi, Shanchieh Jay Yang, Michael Zuzak
Abstract: Recent advances in Large Language Models (LLMs) have driven interest in automating cybersecurity penetration testing workflows, offering the promise of faster and more consistent vulnerability assessment for enterprise systems. Existing LLM agents for penetration testing primarily rely on self-guided reasoning, which can produce inaccurate or hallucinated procedural steps. As a result, the LLM agent may undertake unproductive actions, such as exploiting unused software libraries or generating cyclical responses that repeat prior tactics. In this work, we propose a guided reasoning pipeline for penetration testing LLM agents that incorporates a deterministic task tree built from the MITRE ATT&CK Matrix, a proven penetration testing kll chain, to constrain the LLM's reaoning process to explicitly defined tactics, techniques, and procedures. This anchors reasoning in proven penetration testing methodologies and filters out ineffective actions by guiding the agent towards more productive attack procedures. To evaluate our approach, we built an automated penetration testing LLM agent using three LLMs (Llama-3-8B, Gemini-1.5, and GPT-4) and applied it to navigate 10 HackTheBox cybersecurity exercises with 103 discrete subtasks representing real-world cyberattack scenarios. Our proposed reasoning pipeline guided the LLM agent through 71.8\%, 72.8\%, and 78.6\% of subtasks using Llama-3-8B, Gemini-1.5, and GPT-4, respectively. Comparatively, the state-of-the-art LLM penetration testing tool using self-guided reasoning completed only 13.5\%, 16.5\%, and 75.7\% of subtasks and required 86.2\%, 118.7\%, and 205.9\% more model queries. This suggests that incorporating a deterministic task tree into LLM reasoning pipelines can enhance the accuracy and efficiency of automated cybersecurity assessments
Authors: Zheyuan Hu, Robyn Wu, Naveen Enock, Jasmine Li, Riya Kadakia, Zackory Erickson, Aviral Kumar
Abstract: Modern paradigms for robot imitation train expressive policy architectures on large amounts of human demonstration data. Yet performance on contact-rich, deformable-object, and long-horizon tasks plateau far below perfect execution, even with thousands of expert demonstrations. This is due to the inefficiency of existing ``expert'' data collection procedures based on human teleoperation. To address this issue, we introduce RaC, a new phase of training on human-in-the-loop rollouts after imitation learning pre-training. In RaC, we fine-tune a robotic policy on human intervention trajectories that illustrate recovery and correction behaviors. Specifically, during a policy rollout, human operators intervene when failure appears imminent, first rewinding the robot back to a familiar, in-distribution state and then providing a corrective segment that completes the current sub-task. Training on this data composition expands the robotic skill repertoire to include retry and adaptation behaviors, which we show are crucial for boosting both efficiency and robustness on long-horizon tasks. Across three real-world bimanual control tasks: shirt hanging, airtight container lid sealing, takeout box packing, and a simulated assembly task, RaC outperforms the prior state-of-the-art using 10$\times$ less data collection time and samples. We also show that RaC enables test-time scaling: the performance of the trained RaC policy scales linearly in the number of recovery maneuvers it exhibits. Videos of the learned policy are available at https://rac-scaling-robot.github.io/.
Authors: Chiwoo Park, Robert Waelder, Bonggwon Kang, Benji Maruyama, Soondo Hong, Robert Gramacy
Abstract: Active learning of Gaussian process (GP) surrogates has been useful for optimizing experimental designs for physical/computer simulation experiments, and for steering data acquisition schemes in machine learning. In this paper, we develop a method for active learning of piecewise, Jump GP surrogates. Jump GPs are continuous within, but discontinuous across, regions of a design space, as required for applications spanning autonomous materials design, configuration of smart factory systems, and many others. Although our active learning heuristics are appropriated from strategies originally designed for ordinary GPs, we demonstrate that additionally accounting for model bias, as opposed to the usual model uncertainty, is essential in the Jump GP context. Toward that end, we develop an estimator for bias and variance of Jump GP models. Illustrations, and evidence of the advantage of our proposed methods, are provided on a suite of synthetic benchmarks, and real-simulation experiments of varying complexity.
Authors: Yanxin Yang, Ming Hu, Xiaofei Xie, Yue Cao, Pengyu Zhang, Yihao Huang, Mingsong Chen
Abstract: As a distributed machine learning paradigm, Federated Learning (FL) enables large-scale clients to collaboratively train a model without sharing their raw data. However, due to the lack of data auditing for untrusted clients, FL is vulnerable to poisoning attacks, especially backdoor attacks. By using poisoned data for local training or directly changing the model parameters, attackers can easily inject backdoors into the model, which can trigger the model to make misclassification of targeted patterns in images. To address these issues, we propose a novel data-free trigger-generation-based defense approach based on the two characteristics of backdoor attacks: i) triggers are learned faster than normal knowledge, and ii) trigger patterns have a greater effect on image classification than normal class patterns. Our approach generates the images with newly learned knowledge by identifying the differences between the old and new global models, and filters trigger images by evaluating the effect of these generated images. By using these trigger images, our approach eliminates poisoned models to ensure the updated global model is benign. Comprehensive experiments demonstrate that our approach can defend against almost all the existing types of backdoor attacks and outperform all the seven state-of-the-art defense methods with both IID and non-IID scenarios. Especially, our approach can successfully defend against the backdoor attack even when 80\% of the clients are malicious.
Authors: Peng Zhao, Yan-Feng Xie, Lijun Zhang, Zhi-Hua Zhou
Abstract: Non-stationary online learning has drawn much attention in recent years. In particular, dynamic regret and adaptive regret are proposed as two principled performance measures for online convex optimization in non-stationary environments. To optimize them, a two-layer online ensemble is usually deployed due to the inherent uncertainty of non-stationarity, in which multiple base-learners are maintained and a meta-algorithm is employed to track the best one on the fly. However, the two-layer structure raises concerns about computational complexity -- such methods typically maintain $O(\log T)$ base-learners simultaneously for a $T$-round online game and thus perform multiple projections onto the feasible domain per round, which becomes the computational bottleneck when the domain is complicated. In this paper, we present efficient methods for optimizing dynamic regret and adaptive regret that reduce the number of projections per round from $O(\log T)$ to $1$. The proposed algorithms require only one gradient query and one function evaluation at each round. Our technique hinges on the reduction mechanism developed in parameter-free online learning and requires non-trivial modifications for non-stationary online methods. Furthermore, we study an even stronger measure, namely "interval dynamic regret", and reduce the number of projections per round from $O(\log^2 T)$ to $1$ for minimizing it. Our reduction demonstrates broad generality and applies to two important applications: online stochastic control and online principal component analysis, resulting in methods that are both efficient and optimal. Finally, empirical studies verify our theoretical findings.
Authors: Pratiksha Thaker, Amrith Setlur, Zhiwei Steven Wu, Virginia Smith
Abstract: Public pretraining is a promising approach to improve differentially private model training. However, recent work has noted that many positive research results studying this paradigm only consider in-distribution tasks, and may not apply to settings where there is distribution shift between the pretraining and finetuning data -- a scenario that is likely when finetuning private tasks due to the sensitive nature of the data. In this work, we show empirically across three tasks that even in settings with large distribution shift, where both zero-shot performance from public data and training from scratch with private data give unusably weak results, public features can in fact improve private training accuracy by up to 67\% over private training from scratch. We provide a theoretical explanation for this phenomenon, showing that if the public and private data share a low-dimensional representation, public representations can improve the sample complexity of private training even if it is impossible to learn the private task from the public data alone. Altogether, our results provide evidence that public data can indeed make private training practical in realistic settings of extreme distribution shift.
Authors: Xintong Li, Junda Wu, Tong Yu, Yu Wang, Xiang Chen, Jiuxiang Gu, Lina Yao, Julian McAuley, Jingbo Shang
Abstract: Instruction tuning in multimodal large language models (MLLMs) generally involves cooperative learning between a backbone LLM and a feature encoder of non-text input modalities. The major challenge is how to efficiently find the synergy between the two modules so that LLMs can adapt their reasoning abilities to downstream tasks while feature encoders can adjust to provide more task-specific information about its modality. In this paper, we analyze the MLLM instruction tuning from both theoretical and empirical perspectives, where we find the unbalanced learning between the feature encoder and the LLM can cause problems of oscillation and biased learning that lead to sub-optimal convergence. Inspired by our findings, we propose a Multimodal Balance Coefficient that enables quantitative measurement of the balance of learning. Based on this, we further design a dynamic learning scheduler that better coordinates the learning between the LLM and feature encoder, alleviating the problems of oscillation and biased learning. In addition, we introduce an auxiliary regularization on the gradient to promote updating with larger step sizes, which potentially allows for a more accurate estimation of the proposed MultiModal Balance Coefficient and further improves the training sufficiency. Our proposed approach is agnostic to the architecture of LLM and feature encoder, so it can be generically integrated with various MLLMs. We conduct experiments on multiple downstream tasks with various MLLMs, demonstrating that the proposed method is more effective than the baselines in MLLM instruction tuning.
Authors: Manav Vora, Jonas Liang, Melkior Ornik
Abstract: Monotonic Partially Observable Markov Decision Processes (POMDPs), where the system state progressively decreases until a restorative action is performed, can be used to model sequential repair problems effectively. This paper considers the problem of solving budget-constrained multi-component monotonic POMDPs, where a finite budget limits the maximal number of restorative actions. For a large number of components, solving such a POMDP using current methods is computationally intractable due to the exponential growth in the state space with an increasing number of components. To address this challenge, we propose a two-step approach. Since the individual components of a budget-constrained multi-component monotonic POMDP are only connected via the shared budget, we first approximate the optimal budget allocation among these components using an approximation of each component POMDP's optimal value function which is obtained through a random forest model. Subsequently, we introduce an oracle-guided meta-trained Proximal Policy Optimization (PPO) algorithm to solve each of the independent budget-constrained single-component monotonic POMDPs. The oracle policy is obtained by performing value iteration on the corresponding monotonic Markov Decision Process (MDP). This two-step method provides scalability in solving truly massive multi-component monotonic POMDPs. To demonstrate the efficacy of our approach, we consider a real-world maintenance scenario that involves inspection and repair of an administrative building by a team of agents within a maintenance budget. Finally, we perform a computational complexity analysis for a varying number of components to show the scalability of the proposed approach.
Authors: Peter St. John, Dejun Lin, Polina Binder, Malcolm Greaves, Vega Shah, John St. John, Adrian Lange, Patrick Hsu, Rajesh Illango, Arvind Ramanathan, Anima Anandkumar, David H Brookes, Akosua Busia, Abhishaike Mahajan, Stephen Malina, Neha Prasad, Sam Sinai, Lindsay Edwards, Thomas Gaudelet, Cristian Regep, Martin Steinegger, Burkhard Rost, Alexander Brace, Kyle Hippe, Luca Naef, Keisuke Kamata, George Armstrong, Kevin Boyd, Zhonglin Cao, Han-Yi Chou, Simon Chu, Allan dos Santos Costa, Sajad Darabi, Eric Dawson, Kieran Didi, Cong Fu, Mario Geiger, Michelle Gill, Darren J Hsu, Gagan Kaushik, Maria Korshunova, Steven Kothen-Hill, Youhan Lee, Meng Liu, Micha Livne, Zachary McClure, Jonathan Mitchell, Alireza Moradzadeh, Ohad Mosafi, Youssef Nashed, Saee Paliwal, Yuxing Peng, Sara Rabhi, Farhad Ramezanghorbani, Danny Reidenbach, Camir Ricketts, Brian C Roland, Kushal Shah, Tyler Shimko, Hassan Sirelkhatim, Savitha Srinivasan, Abraham C Stern, Dorota Toczydlowska, Srimukh Prasad Veccham, Niccol\`o Alberto Elia Venanzi, Anton Vorontsov, Jared Wilber, Isabel Wilkinson, Wei Jing Wong, Eva Xue, Cory Ye, Xin Yu, Yang Zhang, Guoqing Zhou, Becca Zandstein, Alejandro Chacon, Prashant Sohani, Maximilian Stadler, Christian Hundt, Feiwen Zhu, Christian Dallago, Bruno Trentini, Emine Kucukbenli, Saee Paliwal, Timur Rvachov, Eddie Calleja, Johnny Israeli, Harry Clifford, Risto Haukioja, Nicholas Haemel, Kyle Tretina, Neha Tadimeti, Anthony B Costa
Abstract: Artificial Intelligence models encoding biology and chemistry are opening new routes to high-throughput and high-quality in-silico drug development. However, their training increasingly relies on computational scale, with recent protein language models (pLM) training on hundreds of graphical processing units (GPUs). We introduce the BioNeMo Framework to facilitate the training of computational biology and chemistry AI models across hundreds of GPUs. Its modular design allows the integration of individual components, such as data loaders, into existing workflows and is open to community contributions. We detail technical features of the BioNeMo Framework through use cases such as pLM pre-training and fine-tuning. On 256 NVIDIA A100s, BioNeMo Framework trains a three billion parameter BERT-based pLM on over one trillion tokens in 4.2 days. The BioNeMo Framework is open-source and free for everyone to use.
Authors: Ozgu Goksu, Nicolas Pugeault
Abstract: Federated Learning offers a solution for decentralised model training, addressing the difficulties associated with distributed data and privacy in machine learning. However, the fact of data heterogeneity in federated learning frequently hinders the global model's generalisation, leading to low performance and adaptability to unseen data. This problem is particularly critical for specialised applications such as medical imaging, where both the data and the number of clients are limited. In this paper, we empirically demonstrate that inconsistencies in client-side training distributions substantially degrade the performance of federated learning models across multiple benchmark datasets. We propose a novel FL framework using a combination of pruning and regularisation of clients' training to improve the sparsity, redundancy, and robustness of neural connections, and thereby the resilience to model aggregation. To address a relatively unexplored dimension of data heterogeneity, we further introduce a novel benchmark dataset, CelebA-Gender, specifically designed to control for within-class distributional shifts across clients based on attribute variations, thereby complementing the predominant focus on inter-class imbalance in prior federated learning research. Comprehensive experiments on many datasets like CIFAR-10, MNIST, and the newly introduced CelebA-Gender dataset demonstrate that our method consistently outperforms standard FL baselines, yielding more robust and generalizable models in heterogeneous settings.
Authors: Tianren Zhang, Guanyu Chen, Feng Chen
Abstract: Humans develop world models that capture the underlying generation process of data. Whether neural networks can learn similar world models remains an open problem. In this work, we present the first theoretical results for this problem, showing that in a multi-task setting, models with a low-degree bias provably recover latent data-generating variables under mild assumptions--even if proxy tasks involve complex, non-linear functions of the latents. However, such recovery is sensitive to model architecture. Our analysis leverages Boolean models of task solutions via the Fourier-Walsh transform and introduces new techniques for analyzing invertible Boolean transforms, which may be of independent interest. We illustrate the algorithmic implications of our results and connect them to related research areas, including self-supervised learning, out-of-distribution generalization, and the linear representation hypothesis in large language models.
Authors: Micha Livne
Abstract: Learning representations that generalize well to unknown downstream tasks is a central challenge in representation learning. Existing approaches such as contrastive learning, self-supervised masking, and denoising auto-encoders address this challenge with varying trade-offs. In this paper, we introduce the {contrastive Mutual Information Machine} (cMIM), a probabilistic framework that augments the Mutual Information Machine (MIM) with a novel contrastive objective. While MIM maximizes mutual information between inputs and latent variables and encourages clustering of latent codes, its representations underperform on discriminative tasks compared to state-of-the-art alternatives. cMIM addresses this limitation by enforcing global discriminative structure while retaining MIM's generative strengths. We present two main contributions: (1) we propose cMIM, a contrastive extension of MIM that eliminates the need for positive data augmentation and is robust to batch size, unlike InfoNCE-based methods; (2) we introduce {informative embeddings}, a general technique for extracting enriched representations from encoder--decoder models that substantially improve discriminative performance without additional training, and which apply broadly beyond MIM. Empirical results demonstrate that cMIM consistently outperforms MIM and InfoNCE in classification and regression tasks, while preserving comparable reconstruction quality. These findings suggest that cMIM provides a unified framework for learning representations that are simultaneously effective for discriminative and generative applications.
Authors: Kunyang Sun, Dorian Bagni, Joseph M. Cavanagh, Yingze Wang, Jacob M. Sawyer, Bo Zhou, Andrew Gritsevskiy, Oufan Zhang, Teresa Head-Gordon
Abstract: Generative machine learning models for exploring chemical space have shown immense promise, but many molecules they generate are too difficult to synthesize, making them impractical for further investigation or development. In this work, we present a novel approach by fine-tuning Meta's Llama3 Large Language Models (LLMs) to create SynLlama, which generates full synthetic pathways made of commonly accessible building blocks and robust organic reaction templates. SynLlama explores a large synthesizable space using significantly less data, and offers strong performance in both forward and bottom-up synthesis planning compared to other state-of-the-art methods. We find that SynLlama, even without training on external building blocks, can effectively generalize to unseen yet purchasable building blocks, meaning that its reconstruction capabilities extend to a broader synthesizable chemical space than the training data. We also demonstrate the use of SynLlama in a pharmaceutical context for synthesis planning of analog molecules and hit expansion leads for proposed inhibitors of target proteins, offering medicinal chemists a valuable tool for discovery.
Authors: Guoyou Sun, Panagiotis Karras, Qi Zhang
Abstract: Semantic communication has emerged as a promising paradigm to tackle the challenges of massive growing data traffic and sustainable data communication. It shifts the focus from data fidelity to goal-oriented or task-oriented semantic transmission. While deep learning-based methods are commonly used for semantic encoding and decoding, they struggle with the sequential nature of time series data and high computation cost, particularly in resource-constrained IoT environments. Data compression plays a crucial role in reducing transmission and storage costs, yet traditional data compression methods fall short of the demands of goal-oriented communication systems. In this paper, we propose a novel method for direct analytics on time series data compressed by the SHRINK compression algorithm. Through experimentation using outlier detection as a case study, we show that our method outperforms baselines running on uncompressed data in multiple cases, with merely 1% difference in the worst case. Additionally, it achieves four times lower runtime on average and accesses approximately 10% of the data volume, which enables edge analytics with limited storage and computation power. These results demonstrate that our approach offers reliable, high-speed outlier detection analytics for diverse IoT applications while extracting semantics from time-series data, achieving high compression, and reducing data transmission.
Authors: Junxiong Wang, Wen-Ding Li, Daniele Paliotta, Daniel Ritter, Alexander M. Rush, Tri Dao
Abstract: Effective reasoning is crucial to solving complex mathematical problems. Recent large language models (LLMs) have boosted performance by scaling test-time computation through long chain-of-thought reasoning. However, transformer-based models are inherently limited in extending context length due to their quadratic computational complexity and linear memory requirements. In this paper, we introduce a novel hybrid linear RNN reasoning model, M1, built on the Mamba architecture, which allows memory-efficient inference. Our approach leverages a distillation process from existing reasoning models and is further enhanced through RL training. Experimental results on the AIME and MATH benchmarks show that M1 not only outperforms previous linear RNN models but also matches the performance of state-of-the-art Deepseek R1 distilled reasoning models at a similar scale. We also compare our generation speed with a highly performant general purpose inference engine, vLLM, and observe more than a 3x speedup compared to a same size transformer. With throughput speedup, we are able to achieve higher accuracy compared to DeepSeek R1 distilled transformer reasoning models under a fixed generation time budget using self-consistency voting. Overall, we introduce a hybrid Mamba reasoning model and provide a more effective approach to scaling test-time generation using self-consistency or long chain of thought reasoning.
Authors: Michael Banf, Johannes Kuhn
Abstract: Large Language Models (LLMs) have shown remarkable capabilities across various domains, yet they struggle with knowledge-intensive tasks in areas that demand factual accuracy, e.g. industrial automation and healthcare. Key limitations include their tendency to hallucinate, lack of source traceability (provenance), and challenges in timely knowledge updates. Combining language models with knowledge graphs (GraphRAG) offers promising avenues for overcoming these deficits. However, a major challenge lies in creating such a knowledge graph in the first place. Here, we propose a novel approach that combines LLMs with a tripartite knowledge graph representation, which is constructed by connecting complex, domain-specific objects via a curated ontology of corresponding, domain-specific concepts to relevant sections within chunks of text through a concept-anchored pre-analysis of source documents starting from an initial lexical graph. Subsequently, we formulate LLM prompt creation as an unsupervised node classification problem allowing for the optimization of information density, coverage, and arrangement of LLM prompts at significantly reduced lengths. An initial experimental evaluation of our approach on a healthcare use case, involving multi-faceted analyses of patient anamneses given a set of medical concepts as well as a series of clinical guideline literature, indicates its potential to optimize information density, coverage, and arrangement of LLM prompts while significantly reducing their lengths, which, in turn, may lead to reduced costs as well as more consistent and reliable LLM outputs.
Authors: Guangzhi Sun, Potsawee Manakul, Xiao Zhan, Mark Gales
Abstract: Unlearning has emerged as a critical capability for large language models (LLMs) to support data privacy, regulatory compliance, and ethical AI deployment. Recent techniques often rely on obfuscation by injecting incorrect or irrelevant information to suppress knowledge. Such methods effectively constitute knowledge addition rather than true removal, often leaving models vulnerable to probing. In this paper, we formally distinguish unlearning from obfuscation and introduce a probing-based evaluation framework to assess whether existing approaches genuinely remove targeted information. Moreover, we propose DF-MCQ, a novel unlearning method that flattens the model predictive distribution over automatically generated multiple-choice questions using KL-divergence, effectively removing knowledge about target individuals and triggering appropriate refusal behaviour. Experimental results demonstrate that DF-MCQ achieves unlearning with over 90% refusal rate and a random choice-level uncertainty that is much higher than obfuscation on probing questions.
Authors: Assaf Ben-Kish, Itamar Zimerman, M. Jehanzeb Mirza, Lior Wolf, James Glass, Leonid Karlinsky, Raja Giryes
Abstract: A recent trend in LLMs is developing recurrent sub-quadratic models that improve long-context processing efficiency. We investigate leading large long-context models, focusing on how their fixed-size recurrent memory affects their performance. Our experiments reveal that, even when these models are trained for extended contexts, their use of long contexts remains underutilized. Specifically, we demonstrate that a chunk-based inference procedure, which identifies and processes only the most relevant portion of the input can mitigate recurrent memory failures and be effective for many long-context tasks: On LongBench, our method improves the overall performance of Falcon3-Mamba-Inst-7B by 14%, Falcon-Mamba-Inst-7B by 28%, RecurrentGemma-IT-9B by 50%, and RWKV6-Finch-7B by 51%. Surprisingly, this simple approach also leads to state-of-the-art results in the challenging LongBench v2 benchmark, showing competitive performance with equivalent size Transformers. Furthermore, our findings raise questions about whether recurrent models genuinely exploit long-range dependencies, as our single-chunk strategy delivers stronger performance - even in tasks that presumably require cross-context relations.
Authors: Austin H. Cheng, Chong Sun, Al\'an Aspuru-Guzik
Abstract: Generative models of 3D molecular structure play a rapidly growing role in the design and simulation of molecules. Diffusion models currently dominate the space of 3D molecule generation, while autoregressive models have trailed behind. In this work, we present Quetzal, a simple but scalable autoregressive model that builds molecules atom-by-atom in 3D. Treating each molecule as an ordered sequence of atoms, Quetzal combines a causal transformer that predicts the next atom's discrete type with a smaller Diffusion MLP that models the continuous next-position distribution. Compared to existing autoregressive baselines, Quetzal achieves substantial improvements in generation quality and is competitive with the performance of state-of-the-art diffusion models. In addition, by reducing the number of expensive forward passes through a dense transformer, Quetzal enables significantly faster generation speed, as well as exact divergence-based likelihood computation. Finally, without any architectural changes, Quetzal natively handles variable-size tasks like hydrogen decoration and scaffold completion. We hope that our work motivates a perspective on scalability and generality for generative modelling of 3D molecules.
Authors: Ziyan Luo, Tianwei Ni, Pierre-Luc Bacon, Doina Precup, Xujie Si
Abstract: A key approach to state abstraction is approximating behavioral metrics (notably, bisimulation metrics) in the observation space and embedding these learned distances in the representation space. While promising for robustness to task-irrelevant noise, as shown in prior work, accurately estimating these metrics remains challenging, requiring various design choices that create gaps between theory and practice. Prior evaluations focus mainly on final returns, leaving the quality of learned metrics and the source of performance gains unclear. To systematically assess how metric learning works in deep reinforcement learning (RL), we evaluate five recent approaches, unified conceptually as isometric embeddings with varying design choices. We benchmark them with baselines across 20 state-based and 14 pixel-based tasks, spanning 370 task configurations with diverse noise settings. Beyond final returns, we introduce the evaluation of a denoising factor to quantify the encoder's ability to filter distractions. To further isolate the effect of metric learning, we propose and evaluate an isolated metric estimation setting, in which the encoder is influenced solely by the metric loss. Finally, we release an open-source, modular codebase to improve reproducibility and support future research on metric learning in deep RL.
Authors: Xing Lei, Zifeng Zhuang, Shentao Yang, Sheng Xu, Yunhao Luo, Fei Shen, Wenyan Yang, Xuetao Zhang, Donglin Wang
Abstract: Recently, supervised learning (SL) methodology has emerged as an effective approach for offline reinforcement learning (RL) due to their simplicity, stability, and efficiency. However, recent studies show that SL methods lack the trajectory stitching capability, typically associated with temporal difference (TD)-based approaches. A question naturally surfaces: \textit{How can we endow SL methods with stitching capability and close its performance gap with TD learning?} To answer this question, we introduce $Q$-conditioned maximization supervised learning for offline goal-conditioned RL, which enhances SL with the stitching capability through $Q$-conditioned policy and $Q$-conditioned maximization. Concretely, we propose \textbf{G}oal-\textbf{C}onditioned \textbf{\textit{Rein}}forced \textbf{S}upervised \textbf{L}earning (\textbf{GC\textit{Rein}SL}), which consists of (1) estimating the $Q$-function by Normalizing Flows from the offline dataset and (2) finding the maximum $Q$-value within the data support by integrating $Q$-function maximization with Expectile Regression. In inference time, our policy chooses optimal actions based on such a maximum $Q$-value. Experimental results from stitching evaluations on offline RL datasets demonstrate that our method outperforms prior SL approaches with stitching capabilities and goal data augmentation techniques.
Authors: Max Gupta
Abstract: Rapidly learning abstract concepts from limited examples is a hallmark of human intelligence. This work investigates whether gradient-based meta-learning can equip neural networks with inductive biases for efficient few-shot acquisition of discrete concepts. I compare meta-learning methods against a supervised learning baseline on Boolean concepts (logical statements) generated by a probabilistic context-free grammar (PCFG). By systematically varying concept dimensionality (number of features) and recursive compositionality (depth of grammar recursion), I delineate between complexity regimes in which meta-learning robustly improves few-shot concept learning and regimes in which it does not. Meta-learners are much better able to handle compositional complexity than featural complexity. I highlight some reasons for this with a representational analysis of the weights of meta-learners and a loss landscape analysis demonstrating how featural complexity increases the roughness of loss trajectories, allowing curvature-aware optimization to be more effective than first-order methods. I find improvements in out-of-distribution generalization on complex concepts by increasing the number of adaptation steps in meta-SGD, where adaptation acts as a way of encouraging exploration of rougher loss basins. Overall, this work highlights the intricacies of learning compositional versus featural complexity in high dimensional concept spaces and provides a road to understanding the role of 2nd order methods and extended gradient adaptation in few-shot concept learning.
Authors: Xuanhao Mu, G\"okhan Demirel, Yuzhe Zhang, Jianlei Liu, Thorsten Schlachter, Veit Hagenmeyer
Abstract: To bridge the temporal granularity gap in energy network design and operation based on Energy System Models, resampling of time series is required. While conventional upsampling methods are computationally efficient, they often result in significant information loss or increased noise. Advanced models such as time series generation models, Super-Resolution models and imputation models show potential, but also face fundamental challenges. The goal of time series generative models is to learn the distribution of the original data to generate high-resolution series with similar statistical characteristics. This is not entirely consistent with the definition of upsampling. Time series Super-Resolution models or imputation models can degrade the accuracy of upsampling because the input low-resolution time series are sparse and may have insufficient context. Moreover, such models usually rely on supervised learning paradigms. This presents a fundamental application paradox: their training requires the high-resolution time series that is intrinsically absent in upsampling application scenarios. To address the mentioned upsampling issue, this paper introduces a new method utilizing Generative Adversarial Transformers (GATs), which can be trained without access to any ground-truth high-resolution data. Compared with conventional interpolation methods, the introduced method can reduce the root mean square error (RMSE) of upsampling tasks by 9%, and the accuracy of a model predictive control (MPC) application scenario is improved by 13%.
Authors: Xiaodong Cui, A F M Saif, Brian Kingsbury, Tianyi Chen
Abstract: Self-supervised pre-training using unlabeled data is widely used in automatic speech recognition. In this paper, we propose a new self-supervised pre-training approach to dealing with heterogeneous data. Instead of mixing all the data and minimizing the averaged global loss in the conventional way, we impose additional local constraints to ensure that the model optimizes each source of heterogeneous data to its local optimum after $K$-step gradient descent initialized from the model. We formulate this as a bilevel optimization problem, and use the first-order approximation method to solve the problem. We discuss its connection to model-agnostic meta learning. Experiments are carried out on self-supervised pre-training using multi-domain and multilingual datasets, demonstrating that the proposed approach can significantly improve the adaptivity of the self-supervised pre-trained model for the downstream supervised fine-tuning tasks.
Authors: Deddy Jobson, Muktti Shukla, Phuong Dinh, Julio Christian Young, Nick Pitton, Nina Chen, Ryan Ginstrom
Abstract: E-commerce marketplaces make use of a number of marketing channels like emails, push notifications, etc. to reach their users and stimulate purchases. Personalized emails especially are a popular touch point for marketers to inform users of latest items in stock, especially for those who stopped visiting the marketplace. Such emails contain personalized recommendations tailored to each user's interests, enticing users to buy relevant items. A common limitation of these emails is that the primary entry point, the title of the email, tends to follow fixed templates, failing to inspire enough interest in the contents. In this work, we explore the potential of large language models (LLMs) for generating thematic titles that reflect the personalized content of the emails. We perform offline simulations and conduct online experiments on the order of millions of users, finding our techniques useful in improving the engagement between customers and our emails. We highlight key findings and learnings as we productionize the safe and automated generation of email titles for millions of users.
Authors: Pranoy Panda, Raghav Magazine, Chaitanya Devaguptapu, Sho Takemori, Vishal Sharma
Abstract: Large Language Models (LLMs) have revolutionized natural language processing, but their varying capabilities and costs pose challenges in practical applications. LLM routing addresses this by dynamically selecting the most suitable LLM for each query/task. Previous approaches treat this as a supervised learning problem, assuming complete knowledge of optimal query-LLM pairings. However, real-world scenarios lack such comprehensive mappings and face evolving user queries. We thus propose to study LLM routing as a contextual bandit problem, enabling adaptive decision-making using bandit feedback without requiring exhaustive inference across all LLMs for all queries (in contrast to supervised routing). To address this problem, we develop a shared embedding space for queries and LLMs, where query and LLM embeddings are aligned to reflect their affinity. This space is initially learned from offline human preference data and refined through online bandit feedback. We instantiate this idea through Preference-prior Informed Linucb fOr adaptive rouTing (PILOT), a novel extension of LinUCB. To handle diverse user budgets for model routing, we introduce an online cost policy modeled as a multi-choice knapsack problem, ensuring resource-efficient routing.
Authors: Hongyu Wang, Jingquan Wang, Dan Negrut
Abstract: Data-driven modeling of constrained multibody systems faces two persistent challenges: high computational cost and limited long-term prediction accuracy. To address these issues, we introduce the Flow-Matching Neural Ordinary Differential Equation (FNODE), a framework that learns acceleration vector fields directly from trajectory data. By reformulating the training objective to supervise accelerations rather than integrated states, FNODE eliminates the need for backpropagation through an ODE solver, which represents a bottleneck in traditional Neural ODEs. Acceleration targets are computed efficiently using numerical differentiation techniques, including a hybrid Fast Fourier Transform (FFT) and Finite Difference (FD) scheme. We evaluate FNODE on a diverse set of benchmarks, including the single and triple mass-spring-damper systems, double pendulum, slider-crank, and cart-pole. Across all cases, FNODE consistently outperforms existing approaches such as Multi-Body Dynamic Neural ODE (MBD-NODE), Long Short-Term Memory (LSTM) networks, and Fully Connected Neural Networks (FCNN), demonstrating good accuracy, generalization, and computational efficiency.
Authors: Xiao Xue, Marco F. P. ten Eikelder, Tianyue Yang, Yiqing Li, Kan He, Shuo Wang, Peter V. Coveney
Abstract: Phase separation in binary mixtures, governed by the Cahn-Hilliard equation, plays a central role in interfacial dynamics across materials science and soft matter. While numerical solvers are accurate, they are often computationally expensive and lack flexibility across varying initial conditions and geometries. Neural operators provide a data-driven alternative by learning solution operators between function spaces, but current architectures often fail to capture multiscale behavior and neglect underlying physical symmetries. Here we show that an equivariant U-shaped neural operator (E-UNO) can learn the evolution of the phase-field variable from short histories of past dynamics, achieving accurate predictions across space and time. The model combines global spectral convolution with a multi-resolution U-shaped architecture and regulates translation equivariance to align with the underlying physics. E-UNO outperforms standard Fourier neural operator and U-shaped neural operator baselines, particularly on fine-scale and high-frequency structures. By encoding symmetry and scale hierarchy, the model generalizes better, requires less training data, and yields physically consistent dynamics. This establishes E-UNO as an efficient surrogate for complex phase-field systems.
Authors: Ibne Farabi Shihab, Sanjeda Akter, Anuj Sharma
Abstract: Sparse-reward reinforcement learning (RL) remains fundamentally hard: without structure, any agent needs $\Omega(|\mathcal{S}||\mathcal{A}|/p)$ samples to recover rewards. We introduce Policy-Aware Matrix Completion (PAMC) as a first concrete step toward a structural reward learning framework. Our key idea is to exploit approximate low-rank + sparse structure in the reward matrix, under policy-biased (MNAR) sampling. We prove recovery guarantees with inverse-propensity weighting, and establish a visitation-weighted error-to-regret bound linking completion error to control performance. Importantly, when assumptions weaken, PAMC degrades gracefully: confidence intervals widen and the algorithm abstains, ensuring safe fallback to exploration. Empirically, PAMC improves sample efficiency across Atari-26 (10M steps), DM Control, MetaWorld MT50, D4RL offline RL, and preference-based RL benchmarks, outperforming DrQ-v2, DreamerV3, Agent57, T-REX/D-REX, and PrefPPO under compute-normalized comparisons. Our results highlight PAMC as a practical and principled tool when structural rewards exist, and as a concrete first instantiation of a broader structural reward learning perspective.
Authors: Minqi Jiang, Andrei Lupu, Yoram Bachrach
Abstract: Progress in many task domains emerges from repeated revisions to previous solution attempts. Training agents that can reliably self-improve over such sequences at inference-time is a natural target for reinforcement learning (RL), yet the naive approach assumes a fixed maximum iteration depth, which can be both costly and arbitrary. We present Exploratory Iteration (ExIt), a family of autocurriculum RL methods that directly exploits the recurrent structure of self-improvement tasks to train LLMs to perform multi-step self-improvement at inference-time while only training on the most informative single-step iterations. ExIt grows a task space by selectively sampling the most informative intermediate, partial histories encountered during an episode for continued iteration, treating these starting points as new self-iteration task instances to train a self-improvement policy. ExIt can further pair with explicit exploration mechanisms to sustain greater task diversity. Across several domains, encompassing competition math, multi-turn tool-use, and machine learning engineering, we demonstrate that ExIt strategies, starting from either a single or many task instances, can produce policies exhibiting strong inference-time self-improvement on held-out task instances, and the ability to iterate towards higher performance over a step budget extending beyond the average iteration depth encountered during training.
Authors: Jie Fu, Hong Yuan, Zhili Chen, Wendy Hui Wang
Abstract: Graph Neural Networks (GNNs) have emerged as powerful models for learning from graph-structured data. However, their widespread adoption has raised serious privacy concerns. While prior research has primarily focused on edge-level privacy, a critical yet underexplored threat lies in topology privacy - the confidentiality of the graph's overall structure. In this work, we present a comprehensive study on topology privacy risks in GNNs, revealing their vulnerability to graph-level inference attacks. To this end, we propose a suite of Topology Inference Attacks (TIAs) that can reconstruct the structure of a target training graph using only black-box access to a GNN model. Our findings show that GNNs are highly susceptible to these attacks, and that existing edge-level differential privacy mechanisms are insufficient as they either fail to mitigate the risk or severely compromise model accuracy. To address this challenge, we introduce Private Graph Reconstruction (PGR), a novel defense framework designed to protect topology privacy while maintaining model accuracy. PGR is formulated as a bi-level optimization problem, where a synthetic training graph is iteratively generated using meta-gradients, and the GNN model is concurrently updated based on the evolving graph. Extensive experiments demonstrate that PGR significantly reduces topology leakage with minimal impact on model accuracy. Our code is available at https://github.com/JeffffffFu/PGR.
Authors: Christo Mathew, Wentian Wang, Jacob Feldman, Lazaros K. Gallos, Paul B. Kantor, Vladimir Menkov, Hao Wang
Abstract: We investigate reinforcement learning in the Game Of Hidden Rules (GOHR) environment, a complex puzzle in which an agent must infer and execute hidden rules to clear a 6$\times$6 board by placing game pieces into buckets. We explore two state representation strategies, namely Feature-Centric (FC) and Object-Centric (OC), and employ a Transformer-based Advantage Actor-Critic (A2C) algorithm for training. The agent has access only to partial observations and must simultaneously infer the governing rule and learn the optimal policy through experience. We evaluate our models across multiple rule-based and trial-list-based experimental setups, analyzing transfer effects and the impact of representation on learning efficiency.
Authors: Hongzong Li, Jiahao Ma, Zhanpeng Shi, Rui Xiao, Fanming Jin, Ye-Fan Hu, Jian-Dong Huang
Abstract: Antibody binding site prediction plays a pivotal role in computational immunology and therapeutic antibody design. Existing sequence or structure methods rely on single-view features and fail to identify antibody-specific binding sites on the antigens. In this paper, we propose \textbf{CAME-AB}, a novel Cross-modality Attention framework with a Mixture-of-Experts (MoE) backbone for robust antibody binding site prediction. CAME-AB integrates five biologically grounded modalities, including raw amino acid encodings, BLOSUM substitution profiles, pretrained language model embeddings, structure-aware features, and GCN-refined biochemical graphs, into a unified multimodal representation. To enhance adaptive cross-modal reasoning, we propose an \emph{adaptive modality fusion} module that learns to dynamically weight each modality based on its global relevance and input-specific contribution. A Transformer encoder combined with an MoE module further promotes feature specialization and capacity expansion. We additionally incorporate a supervised contrastive learning objective to explicitly shape the latent space geometry, encouraging intra-class compactness and inter-class separability. To improve optimization stability and generalization, we apply stochastic weight averaging during training. Extensive experiments on benchmark antibody-antigen datasets demonstrate that CAME-AB consistently outperforms strong baselines on multiple metrics, including Precision, Recall, F1-score, AUC-ROC, and MCC. Ablation studies further validate the effectiveness of each architectural component and the benefit of multimodal feature integration. The model implementation details and the codes are available on https://anonymous.4open.science/r/CAME-AB-C525
Authors: Victor Toscano-Duran, Rocio Gonzalez-Diaz, Miguel A. Guti\'errez-Naranjo
Abstract: While it is well-established that artificial neural networks are universal approximators for continuous functions on compact domains, many modern approaches rely on deep or overparameterized architectures that incur high computational costs. In this paper, a new type of small shallow neural network, called the Barycentric Neural Network (BNN), is proposed, which leverages a fixed set of base points and their barycentric coordinates to define both its structure and its parameters. We demonstrate that our BNN enables the exact representation of continuous piecewise linear functions (CPLFs), ensuring strict continuity across segments. Since any continuous function over a compact domain can be approximated arbitrarily well by CPLFs, the BNN naturally emerges as a flexible and interpretable tool for function approximation. Beyond the use of this representation, the main contribution of the paper is the introduction of a new variant of persistent entropy, a topological feature that is stable and scale invariant, called the length-weighted persistent entropy (LWPE), which is weighted by the lifetime of topological features. Our framework, which combines the BNN with a loss function based on our LWPE, aims to provide flexible and geometrically interpretable approximations of nonlinear continuous functions in resource-constrained settings, such as those with limited base points for BNN design and few training epochs. Instead of optimizing internal weights, our approach directly optimizes the base points that define the BNN. Experimental results show that our approach achieves superior and faster approximation performance compared to classical loss functions such as MSE, RMSE, MAE, and log-cosh.
Authors: Haruki Yokota, Koki Yamada, Yuichi Tanaka, Antonio Ortega
Abstract: We propose a novel framework for learning time-varying graphs from spatiotemporal measurements. Given an appropriate prior on the temporal behavior of signals, our proposed method can estimate time-varying graphs from a small number of available measurements. To achieve this, we introduce two regularization terms in convex optimization problems that constrain sparseness of temporal variations of the time-varying networks. Moreover, a computationally-scalable algorithm is introduced to efficiently solve the optimization problem. The experimental results with synthetic and real datasets (point cloud and temperature data) demonstrate our proposed method outperforms the existing state-of-the-art methods.
Authors: Gustavo Assun\c{c}\~ao, Miguel Castelo-Branco, Paulo Menezes
Abstract: Background: Exploration of the physical environment is an indispensable precursor to information acquisition and knowledge consolidation for living organisms. Yet, current artificial intelligence models lack these autonomy capabilities during training, hindering their adaptability. This work proposes a learning framework for artificial agents to obtain an intrinsic exploratory drive, based on epistemic and achievement emotions triggered during data observation. Methods: This study proposes a dual-module reinforcement framework, where data analysis scores dictate pride or surprise, in accordance with psychological studies on humans. A correlation between these states and exploration is then optimized for agents to meet their learning goals. Results: Causal relationships between states and exploration are demonstrated by the majority of agents. A 15.4\% mean increase is noted for surprise, with a 2.8\% mean decrease for pride. Resulting correlations of $\rho_{surprise}=0.461$ and $\rho_{pride}=-0.237$ are obtained, mirroring previously reported human behavior. Conclusions: These findings lead to the conclusion that bio-inspiration for AI development can be of great use. This can incur benefits typically found in living beings, such as autonomy. Further, it empirically shows how AI methodologies can corroborate human behavioral findings, showcasing major interdisciplinary importance. Ramifications are discussed.
Authors: Ali Reza Ibrahimzada, Yang Chen, Ryan Rong, Reyhaneh Jabbarvand
Abstract: Bugs are essential in software engineering; many research studies in the past decades have been proposed to detect, localize, and repair bugs in software systems. Effectiveness evaluation of such techniques requires complex bugs, i.e., those that are hard to detect through testing and hard to repair through debugging. From the classic software engineering point of view, a hard-to-repair bug differs from the correct code in multiple locations, making it hard to localize and repair. Hard-to-detect bugs, on the other hand, manifest themselves under specific test inputs and reachability conditions. These two objectives, i.e., generating hard-to-detect and hard-to-repair bugs, are mostly aligned; a bug generation technique can change multiple statements to be covered only under a specific set of inputs. However, these two objectives are conflicting for learning-based techniques: A bug should have a similar code representation to the correct code in the training data to challenge a bug prediction model to distinguish them. The hard-to-repair bug definition remains the same but with a caveat: the more a bug differs from the original code, the more distant their representations are and easier to be detected. We propose BugFarm, to transform arbitrary code into multiple complex bugs. BugFarm leverages LLMs to mutate code in multiple locations (hard-to-repair). To ensure that multiple modifications do not notably change the code representation, BugFarm analyzes the attention of the underlying model and instructs LLMs to only change the least attended locations (hard-to-detect). Our comprehensive evaluation of 435k+ bugs from over 1.9M mutants generated by BUGFARM and two alternative approaches demonstrates our superiority in generating bugs that are hard to detect by learning-based bug prediction approaches and hard-to-repair by state-of-the-art learning-based program repair technique.
Authors: Xin Jin, Bohan Li, BAAO Xie, Wenyao Zhang, Jinming Liu, Ziqiang Li, Tao Yang, Wenjun Zeng
Abstract: Representation disentanglement may help AI fundamentally understand the real world and thus benefit both discrimination and generation tasks. It currently has at least three unresolved core issues: (i) heavy reliance on label annotation and synthetic data -- causing poor generalization on natural scenarios; (ii) heuristic/hand-craft disentangling constraints make it hard to adaptively achieve an optimal training trade-off; (iii) lacking reasonable evaluation metric, especially for the real label-free data. To address these challenges, we propose a \textbf{C}losed-\textbf{L}oop unsupervised representation \textbf{Dis}entanglement approach dubbed \textbf{CL-Dis}. Specifically, we use diffusion-based autoencoder (Diff-AE) as a backbone while resorting to $\beta$-VAE as a co-pilot to extract semantically disentangled representations. The strong generation ability of diffusion model and the good disentanglement ability of VAE model are complementary. To strengthen disentangling, VAE-latent distillation and diffusion-wise feedback are interconnected in a closed-loop system for a further mutual promotion. Then, a self-supervised \textbf{Navigation} strategy is introduced to identify interpretable semantic directions in the disentangled latent space. Finally, a new metric based on content tracking is designed to evaluate the disentanglement effect. Experiments demonstrate the superiority of CL-Dis on applications like real image manipulation and visual analysis.
Authors: Ziyu Zhang, Johann Laconte, Daniil Lisus, Timothy D. Barfoot
Abstract: This paper presents a novel method for assessing the resilience of the ICP algorithm via learning-based, worst-case attacks on lidar point clouds. For safety-critical applications such as autonomous navigation, ensuring the resilience of algorithms before deployments is crucial. The ICP algorithm is the standard for lidar-based localization, but its accuracy can be greatly affected by corrupted measurements from various sources, including occlusions, adverse weather, or mechanical sensor issues. Unfortunately, the complex and iterative nature of ICP makes assessing its resilience to corruption challenging. While there have been efforts to create challenging datasets and develop simulations to evaluate the resilience of ICP, our method focuses on finding the maximum possible ICP error that can arise from corrupted measurements at a location. We demonstrate that our perturbation-based adversarial attacks can be used pre-deployment to identify locations on a map where ICP is particularly vulnerable to corruptions in the measurements. With such information, autonomous robots can take safer paths when deployed, to mitigate against their measurements being corrupted. The proposed attack outperforms baselines more than 88% of the time across a wide range of scenarios.
Authors: Zhiwei Wang, Yunji Wang, Zhongwang Zhang, Zhangchen Zhou, Hui Jin, Tianyang Hu, Jiacheng Sun, Zhenguo Li, Yaoyu Zhang, Zhi-Qin John Xu
Abstract: Large language models have consistently struggled with complex reasoning tasks, such as mathematical problem-solving. Investigating the internal reasoning mechanisms of these models can help us design better model architectures and training strategies, ultimately enhancing their reasoning capability. In this study, we constructed a symbolic multi-step reasoning task to investigate the information propagation mechanisms in Transformer models when solving the task through direct answering and Chain-of-Thought (CoT) reasoning. We introduced the concept of buffer mechanism: the model stores various information in distinct buffers and selectively extracts it through the query-key matrix. We proposed a random matrix-based algorithm to enhance the model's reasoning ability. This algorithm introduces only 132 trainable parameters, yet leads to significant performance improvements on 7 multi-step reasoning datasets, including PrOntoQA, LogicAsker, and LogicInference. These findings provide new insights into understanding the large language models.
Authors: Yurui Chang, Bochuan Cao, Yujia Wang, Jinghui Chen, Lu Lin
Abstract: Large Language Models (LLMs) have demonstrated impressive performances in complex text generation tasks. However, the contribution of the input prompt to the generated content still remains obscure to humans, underscoring the necessity of understanding the causality between input and output pairs. Existing works for providing prompt-specific explanation often confine model output to be classification or next-word prediction. Few initial attempts aiming to explain the entire language generation often treat input prompt texts independently, ignoring their combinatorial effects on the follow-up generation. In this study, we introduce a counterfactual explanation framework based on Joint Prompt Attribution, JoPA, which aims to explain how a few prompt texts collaboratively influences the LLM's complete generation. Particularly, we formulate the task of prompt attribution for generation interpretation as a combinatorial optimization problem, and introduce a probabilistic algorithm to search for the casual input combination in the discrete space. We define and utilize multiple metrics to evaluate the produced explanations, demonstrating both the faithfulness and efficiency of our framework.
Authors: Shuze Chen, David Simchi-Levi, Chonghuan Wang
Abstract: Utilizing randomized experiments to evaluate the effect of short-term treatments on the short-term outcomes has been well understood and become the golden standard in industrial practice. However, as service systems become increasingly dynamical and personalized, much focus is shifting toward maximizing long-term outcomes, such as customer lifetime value, through lifetime exposure to interventions. Our goal is to assess the impact of treatment and control policies on long-term outcomes from relatively short-term observations, such as those generated by A/B testing. A key managerial observation is that many practical treatments are local, affecting only targeted states while leaving other parts of the policy unchanged. This paper rigorously investigates whether and how such locality can be exploited to improve estimation of long-term effects in Markov Decision Processes (MDPs), a fundamental model of dynamic systems. We first develop optimal inference techniques for general A/B testing in MDPs and establish corresponding efficiency bounds. We then propose methods to harness the localized structure by sharing information on the non-targeted states. Our new estimator can achieve a linear reduction with the number of test arms for a major part of the variance without sacrificing unbiasedness. It also matches a tighter variance lower bound that accounts for locality. Furthermore, we extend our framework to a broad class of differentiable estimators, which encompasses many widely used approaches in practice. We show that all such estimators can benefit from variance reduction through information sharing without increasing their bias. Together, these results provide both theoretical foundations and practical tools for conducting efficient experiments in dynamic service systems with local treatments.
Authors: Thomas Thebaud, Anna Favaro, Casey Chen, Gabrielle Chavez, Laureano Moro-Velazquez, Ankur Butala, Najim Dehak
Abstract: Motor dysfunction is a common sign of neurodegenerative diseases (NDs) such as Parkinson's disease (PD) and Alzheimer's disease (AD), but may be difficult to detect, especially in the early stages. In this work, we examine the behavior of a wide array of explainable metrics extracted from the handwriting signals of 113 subjects performing multiple tasks on a digital tablet, as part of the Neurological Signals dataset. The aim is to measure their effectiveness in characterizing NDs, including AD and PD. To this end, task-agnostic and task-specific metrics are extracted from 14 distinct tasks. Subsequently, through statistical analysis and a series of classification experiments, we investigate which metrics provide greater discriminative power between NDs and healthy controls and amongst different NDs. Preliminary results indicate that the tasks at hand can all be effectively leveraged to distinguish between the considered set of NDs, specifically by measuring the stability, the speed of writing, the time spent not writing, and the pressure variations between groups from our handcrafted explainable metrics, which shows p-values lower than 0.0001 for multiple tasks. Using various binary classification algorithms on the computed metrics, we obtain up to 87 % accuracy for the discrimination between AD and healthy controls (CTL), and up to 69 % for the discrimination between PD and CTL.
Authors: S\'egol\`ene Martin, Anne Gagneux, Paul Hagemann, Gabriele Steidl
Abstract: In this paper, we introduce Plug-and-Play (PnP) Flow Matching, an algorithm for solving imaging inverse problems. PnP methods leverage the strength of pre-trained denoisers, often deep neural networks, by integrating them in optimization schemes. While they achieve state-of-the-art performance on various inverse problems in imaging, PnP approaches face inherent limitations on more generative tasks like inpainting. On the other hand, generative models such as Flow Matching pushed the boundary in image sampling yet lack a clear method for efficient use in image restoration. We propose to combine the PnP framework with Flow Matching (FM) by defining a time-dependent denoiser using a pre-trained FM model. Our algorithm alternates between gradient descent steps on the data-fidelity term, reprojections onto the learned FM path, and denoising. Notably, our method is computationally efficient and memory-friendly, as it avoids backpropagation through ODEs and trace computations. We evaluate its performance on denoising, super-resolution, deblurring, and inpainting tasks, demonstrating superior results compared to existing PnP algorithms and Flow Matching based state-of-the-art methods.
Authors: Yanjie Ze, Zixuan Chen, Wenhao Wang, Tianyi Chen, Xialin He, Ying Yuan, Xue Bin Peng, Jiajun Wu
Abstract: Humanoid robots capable of autonomous operation in diverse environments have long been a goal for roboticists. However, autonomous manipulation by humanoid robots has largely been restricted to one specific scene, primarily due to the difficulty of acquiring generalizable skills and the expensiveness of in-the-wild humanoid robot data. In this work, we build a real-world robotic system to address this challenging problem. Our system is mainly an integration of 1) a whole-upper-body robotic teleoperation system to acquire human-like robot data, 2) a 25-DoF humanoid robot platform with a height-adjustable cart and a 3D LiDAR sensor, and 3) an improved 3D Diffusion Policy learning algorithm for humanoid robots to learn from noisy human data. We run more than 2000 episodes of policy rollouts on the real robot for rigorous policy evaluation. Empowered by this system, we show that using only data collected in one single scene and with only onboard computing, a full-sized humanoid robot can autonomously perform skills in diverse real-world scenarios. Videos are available at https://humanoid-manipulation.github.io .
Authors: Wei Wu, Zhuoshi Pan, Chao Wang, Liyi Chen, Yunchu Bai, Tianfu Wang, Kun Fu, Zheng Wang, Hui Xiong
Abstract: Rapid advances in Large Language Models (LLMs) have spurred demand for processing extended context sequences in contemporary applications. However, this progress faces two challenges: performance degradation due to sequence lengths out-of-distribution, and excessively long inference times caused by the quadratic computational complexity of attention. These issues limit LLMs in long-context scenarios. In this paper, we propose Dynamic Token-Level KV Cache Selection (TokenSelect), a training-free method for efficient and accurate long-context inference. TokenSelect builds upon the observation of non-contiguous attention sparsity, using QK dot products to measure per-head KV Cache criticality at token-level. By per-head soft voting mechanism, TokenSelect selectively involves a few critical KV cache tokens in attention calculation without sacrificing accuracy. To further accelerate TokenSelect, we design the Selection Cache based on observations of consecutive Query similarity and implemented the efficient Paged Dot Product Kernel, significantly reducing the selection overhead. A comprehensive evaluation of TokenSelect demonstrates up to $23.84\times$ speedup in attention computation and up to $2.28\times$ acceleration in end-to-end latency, while providing superior performance compared to state-of-the-art long-context inference methods.
Authors: Ahmed Luqman, Khuzemah Qazi, Murray Patterson, Malik Jahan Khan, Imdadullah Khan
Abstract: Despite the success of CNN models on a variety of Image classification and segmentation tasks, their extensive computational and storage demands pose considerable challenges for real-world deployment on resource-constrained devices. Quantization is one technique that aims to alleviate these large storage requirements and speed up the inference process by reducing the precision of model parameters to lower-bit representations. In this paper, we introduce a novel post-training quantization method for model weights. Our method finds optimal clipping thresholds and scaling factors along with mathematical guarantees that our method minimizes quantization noise. Empirical results on real-world datasets demonstrate that our quantization scheme significantly reduces model size and computational requirements while preserving model accuracy.
Authors: John M. Hanna, Hugues Talbot, Irene E. Vignon-Clementel
Abstract: In machine learning and statistical modeling, the mean square or absolute error is commonly used as an error metric, also called a "loss function." While effective in reducing the average error, this approach may fail to address localized outliers, leading to significant inaccuracies in regions with sharp gradients or discontinuities. This issue is particularly evident in physics-informed neural networks (PINNs), where such localized errors are expected and affect the overall solution. To overcome this limitation, we propose a novel loss function that combines the mean and the standard deviation of the chosen error metric. By minimizing this combined loss function, the method ensures a more uniform error distribution and reduces the impact of localized high-error regions. The proposed loss function is easy to implement and tested on problems of varying complexity: the 1D Poisson equation, the unsteady Burgers' equation, 2D linear elastic solid mechanics, and 2D steady Navier-Stokes equations. Results demonstrate improved solution quality and lower maximum error compared to the standard mean-based loss, with minimal impact on computational time.
Authors: Zeineb Haouari, Jonas Weidner, Yeray Martin-Ruisanchez, Ivan Ezhov, Aswathi Varma, Daniel Rueckert, Bjoern Menze, Benedikt Wiestler
Abstract: Glioblastoma, a highly aggressive brain tumor, poses major challenges due to its poor prognosis and high morbidity rates. Partial differential equation-based models offer promising potential to enhance therapeutic outcomes by simulating patient-specific tumor behavior for improved radiotherapy planning. However, model calibration remains a bottleneck due to the high computational demands of optimization methods like Monte Carlo sampling and evolutionary algorithms. To address this, we recently introduced an approach leveraging a neural forward solver with gradient-based optimization to significantly reduce calibration time. This approach requires a highly accurate and fully differentiable forward model. We investigate multiple architectures, including (i) an enhanced TumorSurrogate, (ii) a modified nnU-Net, and (iii) a 3D Vision Transformer (ViT). The nnU-Net achieved the best overall results, excelling in both tumor outline matching and voxel-level prediction of tumor cell concentration. It yielded the lowest MSE in tumor cell concentration compared to ground truth numerical simulation and the highest Dice score across all tumor cell concentration thresholds. Our study demonstrates significant enhancement in forward solver performance and outlines important future research directions.
Authors: Trung-Khang Tran, Thach V. Bui
Abstract: The goal of group testing is to identify a small number of defective items within a large population. In the non-adaptive setting, tests are designed in advance and represented by a measurement matrix $\mM$, where rows correspond to tests and columns to items. A test is positive if it includes at least one defective item. Traditionally, $\mM$ remains fixed during both testing and recovery. In this work, we address the case where some entries of $\mM$ are missing, yielding a missing measurement matrix $\mG$. Our aim is to reconstruct $\mM$ from $\mG$ using available samples and their outcome vectors. The above problem can be considered as a problem intersected between Boolean matrix factorization and matrix completion, called the matrix completion in group testing (MCGT) problem, as follows. Given positive integers $t,s,n$, let $\mY:=(y_{ij}) \in \{0, 1\}^{t \times s}$, $\mM:=(m_{ij}) \in \{0,1\}^{t \times n}$, $\mX:=(x_{ij}) \in \{0,1\}^{n \times s}$, and matrix $\mG \in \{0,1 \}^{t \times n}$ be a matrix generated from matrix $\mM$ by erasing some entries in $\mM$. Suppose $\mY:=\mM \odot \mX$, where an entry $y_{ij}:=\bigvee_{k=1}^n (m_{ik}\wedge x_{kj})$, and $\wedge$ and $\vee$ are AND and OR operators. Unlike the problem in group testing whose objective is to find $\mX$ when given $\mM$ and $\mY$, our objective is to recover $\mM$ given $\mY,\mX$, and $\mG$. We first prove that the MCGT problem is NP-complete. Next, we show that certain rows with missing entries aid recovery while others do not. For Bernoulli measurement matrices, we establish that larger $s$ increases the higher the probability that $\mM$ can be recovered. We then instantiate our bounds for specific decoding algorithms and validate them through simulations, demonstrating superiority over standard matrix completion and Boolean matrix factorization methods.
Authors: Pengyu Wang, Ying Fang, Xiaofei Li
Abstract: Reverberant speech, denoting the speech signal degraded by reverberation, contains crucial knowledge of both anechoic source speech and room impulse response (RIR). This work proposes a variational Bayesian inference (VBI) framework with neural speech prior (VINP) for joint speech dereverberation and blind RIR identification. In VINP, a probabilistic signal model is constructed in the time-frequency (T-F) domain based on convolution transfer function (CTF) approximation. For the first time, we propose using an arbitrary discriminative dereverberation deep neural network (DNN) to estimate the prior distribution of anechoic speech within a probabilistic model. By integrating both reverberant speech and the anechoic speech prior, VINP yields the maximum a posteriori (MAP) and maximum likelihood (ML) estimations of the anechoic speech spectrum and CTF filter, respectively. After simple transformations, the waveforms of anechoic speech and RIR are estimated. VINP is effective for automatic speech recognition (ASR) systems, which sets it apart from most deep learning (DL)-based single-channel dereverberation approaches. Experiments on single-channel speech dereverberation demonstrate that VINP attains state-of-the-art (SOTA) performance in mean opinion score (MOS) and word error rate (WER). For blind RIR identification, experiments demonstrate that VINP achieves SOTA performance in estimating reverberation time at 60 dB (RT60) and advanced performance in direct-to-reverberation ratio (DRR) estimation. Codes and audio samples are available online.
Authors: Zilu Dong, Xiangqing Shen, Rui Xia
Abstract: As large language models continue to scale up, knowledge editing techniques that modify models' internal knowledge without full retraining have gained significant attention. MEMIT, a prominent batch editing algorithm, stands out for its capability to perform mass knowledge modifications. However, we uncover that MEMIT's editing efficacy significantly deteriorates when processing batches containing multiple edits sharing the same subject. Our analysis reveals this stems from MEMIT's key value modeling framework: identical keys (derived from the shared subject) are forced to represent different values (corresponding to different knowledge), resulting in update conflicts during editing. Addressing this issue, we propose MEMIT-Merge, an enhanced approach that merges value computation processes for facts sharing the same subject, effectively resolving the performance degradation in samesubject batch editing scenarios. Experimental results demonstrate that when MEMIT's edit success rate drops to around 50% at larger batch sizes, MEMIT-Merge maintains a success rate exceeding 90%, showcasing remarkable robustness to subject entity collisions. The code is available at https://github.com/NUSTM/ MEMIT-Merge.
Authors: Jingbiao Mei, Jinghong Chen, Guangyu Yang, Weizhe Lin, Bill Byrne
Abstract: Hateful memes have become a significant concern on the Internet, necessitating robust automated detection systems. While Large Multimodal Models (LMMs) have shown promise in hateful meme detection, they face notable challenges like sub-optimal performance and limited out-of-domain generalization capabilities. Recent studies further reveal the limitations of both supervised fine-tuning (SFT) and in-context learning when applied to LMMs in this setting. To address these issues, we propose a robust adaptation framework for hateful meme detection that enhances in-domain accuracy and cross-domain generalization while preserving the general vision-language capabilities of LMMs. Analysis reveals that our approach achieves improved robustness under adversarial attacks compared to SFT models. Experiments on six meme classification datasets show that our approach achieves state-of-the-art performance, outperforming larger agentic systems. Moreover, our method generates higher-quality rationales for explaining hateful content compared to standard SFT, enhancing model interpretability. Code available at https://github.com/JingbiaoMei/RGCL
Authors: Nobin Sarwar
Abstract: Visual Question Answering requires models to generate accurate answers by integrating visual and textual understanding. However, VQA models still struggle with hallucinations, producing convincing but incorrect answers, particularly in knowledge-driven and Out-of-Distribution scenarios. We introduce FilterRAG, a retrieval-augmented framework that combines BLIP-VQA with Retrieval-Augmented Generation to ground answers in external knowledge sources like Wikipedia and DBpedia. FilterRAG achieves 36.5% accuracy on the OK-VQA dataset, demonstrating its effectiveness in reducing hallucinations and improving robustness in both in-domain and Out-of-Distribution settings. These findings highlight the potential of FilterRAG to improve Visual Question Answering systems for real-world deployment.
Authors: Tom Kempton, Stuart Burrell
Abstract: Advances in hardware and language model architecture have spurred a revolution in natural language generation. However, autoregressive models compute probability distributions over next-token choices, and sampling from these distributions, known as decoding, has received significantly less attention than other design choices. Existing decoding strategies are largely based on heuristics, resulting in methods that are difficult to apply or improve in a principled manner. We develop the theory of decoding strategies for language models by expressing popular decoding algorithms as equilibrium states in the language of ergodic theory and stating the objective functions they optimize. Using this, we analyze the effect of the local normalization step required to make probabilities sum to one in top-k, nucleus, and temperature sampling. We argue that local normalization distortion is a fundamental defect of decoding strategies and quantify the size of this distortion and its effect on mathematical proxies for the quality and diversity of generated text. This yields conclusions for the design of decoding algorithms and the detection of machine-generated text.
Authors: Gautam Kishore Shahi, Oshani Seneviratne, Marc Spaniol
Abstract: With the shift from traditional to digital media, the online landscape now hosts not only reliable news articles but also a significant amount of unreliable content. Digital media has faster reachability by significantly influencing public opinion and advancing political agendas. While newspaper readers may be familiar with their preferred outlets political leanings or credibility, determining unreliable news articles is much more challenging. The credibility of many online sources is often opaque, with AI generated content being easily disseminated at minimal cost. Unreliable news articles, particularly those that followed the Russian invasion of Ukraine in 2022, closely mimic the topics and writing styles of credible sources, making them difficult to distinguish. To address this, we introduce SemCAFE, a system designed to detect news reliability by incorporating entity relatedness into its assessment. SemCAFE employs standard Natural Language Processing techniques, such as boilerplate removal and tokenization, alongside entity level semantic analysis using the YAGO knowledge base. By creating a semantic fingerprint for each news article, SemCAFE could assess the credibility of 46,020 reliable and 3,407 unreliable articles on the 2022 Russian invasion of Ukraine. Our approach improved the macro F1 score by 12% over state of the art methods. The sample data and code are available on GitHub
Authors: Akhiad Bercovich, Itay Levy, Izik Golan, Mohammad Dabbah, Ran El-Yaniv, Omri Puny, Ido Galil, Zach Moshe, Tomer Ronen, Najeeb Nabwani, Ido Shahaf, Oren Tropp, Ehud Karpas, Ran Zilberstein, Jiaqi Zeng, Soumye Singhal, Alexander Bukharin, Yian Zhang, Tugrul Konuk, Gerald Shen, Ameya Sunil Mahabaleshwarkar, Bilal Kartal, Yoshi Suhara, Olivier Delalleau, Zijia Chen, Zhilin Wang, David Mosallanezhad, Adi Renduchintala, Haifeng Qian, Dima Rekesh, Fei Jia, Somshubra Majumdar, Vahid Noroozi, Wasi Uddin Ahmad, Sean Narenthiran, Aleksander Ficek, Mehrzad Samadi, Jocelyn Huang, Siddhartha Jain, Igor Gitman, Ivan Moshkov, Wei Du, Shubham Toshniwal, George Armstrong, Branislav Kisacanin, Matvei Novikov, Daria Gitman, Evelina Bakhturina, Prasoon Varshney, Makesh Narsimhan, Jane Polak Scowcroft, John Kamalu, Dan Su, Kezhi Kong, Markus Kliegl, Rabeeh Karimi Mahabadi, Ying Lin, Sanjeev Satheesh, Jupinder Parmar, Pritam Gundecha, Brandon Norick, Joseph Jennings, Shrimai Prabhumoye, Syeda Nahida Akter, Mostofa Patwary, Abhinav Khattar, Deepak Narayanan, Roger Waleffe, Jimmy Zhang, Bor-Yiing Su, Guyue Huang, Terry Kong, Parth Chadha, Sahil Jain, Christine Harvey, Elad Segal, Jining Huang, Sergey Kashirsky, Robert McQueen, Izzy Putterman, George Lam, Arun Venkatesan, Sherry Wu, Vinh Nguyen, Manoj Kilaru, Andrew Wang, Anna Warno, Abhilash Somasamudramath, Sandip Bhaskar, Maka Dong, Nave Assaf, Shahar Mor, Omer Ullman Argov, Scot Junkin, Oleksandr Romanenko, Pedro Larroy, Monika Katariya, Marco Rovinelli, Viji Balas, Nicholas Edelman, Anahita Bhiwandiwalla, Muthu Subramaniam, Smita Ithape, Karthik Ramamoorthy, Yuting Wu, Suguna Varshini Velury, Omri Almog, Joyjit Daw, Denys Fridman, Erick Galinkin, Michael Evans, Shaona Ghosh, Katherine Luna, Leon Derczynski, Nikki Pope, Eileen Long, Seth Schneider, Guillermo Siman, Tomasz Grzegorzek, Pablo Ribalta, Monika Katariya, Chris Alexiuk, Joey Conway, Trisha Saar, Ann Guan, Krzysztof Pawelec, Shyamala Prayaga, Oleksii Kuchaiev, Boris Ginsburg, Oluwatobi Olabiyi, Kari Briski, Jonathan Cohen, Bryan Catanzaro, Jonah Alben, Yonatan Geifman, Eric Chung
Abstract: We introduce the Llama-Nemotron series of models, an open family of heterogeneous reasoning models that deliver exceptional reasoning capabilities, inference efficiency, and an open license for enterprise use. The family comes in three sizes -- Nano (8B), Super (49B), and Ultra (253B) -- and performs competitively with state-of-the-art reasoning models such as DeepSeek-R1 while offering superior inference throughput and memory efficiency. In this report, we discuss the training procedure for these models, which entails using neural architecture search from Llama 3 models for accelerated inference, knowledge distillation, and continued pretraining, followed by a reasoning-focused post-training stage consisting of two main parts: supervised fine-tuning and large scale reinforcement learning. Llama-Nemotron models are the first open-source models to support a dynamic reasoning toggle, allowing users to switch between standard chat and reasoning modes during inference. To further support open research and facilitate model development, we provide the following resources: 1. We release the Llama-Nemotron reasoning models -- LN-Nano, LN-Super, and LN-Ultra -- under the commercially permissive NVIDIA Open Model License Agreement. 2. We release the complete post-training dataset: Llama-Nemotron-Post-Training-Dataset. 3. We also release our training codebases: NeMo, NeMo-Aligner, and Megatron-LM.
Authors: Xiaoyu Xu, Minxin Du, Qingqing Ye, Haibo Hu
Abstract: Large language models (LLMs) trained over extensive corpora risk memorizing sensitive, copyrighted, or toxic content. To address this, we propose \textbf{OBLIVIATE}, a robust unlearning framework that removes targeted data while preserving model utility. The framework follows a structured process: extracting target tokens, building retain sets, and fine-tuning with a tailored loss function comprising three components -- masking, distillation, and world fact. Using low-rank adapters (LoRA) ensures efficiency without compromising unlearning quality. We conduct experiments on multiple datasets, including Harry Potter series, WMDP, and TOFU, using a comprehensive suite of metrics: \emph{forget quality} (via a new document-level memorization score), \emph{model utility}, and \emph{fluency}. Results demonstrate its effectiveness in resisting membership inference attacks, minimizing the impact on retained data, and maintaining robustness across diverse scenarios.
Authors: Francesco Mori, Francesca Mignacco
Abstract: Dropout is a regularization technique widely used in training artificial neural networks to mitigate overfitting. It consists of dynamically deactivating subsets of the network during training to promote more robust representations. Despite its widespread adoption, dropout probabilities are often selected heuristically, and theoretical explanations of its success remain sparse. Here, we analytically study dropout in two-layer neural networks trained with online stochastic gradient descent. In the high-dimensional limit, we derive a set of ordinary differential equations that fully characterize the evolution of the network during training and capture the effects of dropout. We obtain a number of exact results describing the generalization error and the optimal dropout probability at short, intermediate, and long training times. Our analysis shows that dropout reduces detrimental correlations between hidden nodes, mitigates the impact of label noise, and that the optimal dropout probability increases with the level of noise in the data. Our results are validated by extensive numerical simulations.
Authors: Yiran Yang, Rui Chen
Abstract: In this paper, we consider a score-based Integer Programming (IP) approach for solving the Bayesian Network Structure Learning (BNSL) problem. State-of-the-art BNSL IP formulations suffer from the exponentially large number of variables and constraints. A standard approach in IP to address such challenges is to employ row and column generation techniques, which dynamically generate rows and columns, while the complex pricing problem remains a computational bottleneck for BNSL. For the general class of $\ell_0$-penalized likelihood scores, we show how the pricing problem can be reformulated as a difference of submodular optimization problem, and how the Difference of Convex Algorithm (DCA) can be applied as an inexact method to efficiently solve the pricing problems. Empirically, we show that, for continuous Gaussian data, our row and column generation approach yields solutions with higher quality than state-of-the-art score-based approaches, especially when the graph density increases, and achieves comparable performance against benchmark constraint-based and hybrid approaches, even when the graph size increases.
Authors: Mingxuan Li, Junzhe Zhang, Elias Bareinboim
Abstract: A key task in Artificial Intelligence is learning effective policies for controlling agents in unknown environments to optimize performance measures. Off-policy learning methods, like Q-learning, allow learners to make optimal decisions based on past experiences. This paper studies off-policy learning from biased data in complex and high-dimensional domains where \emph{unobserved confounding} cannot be ruled out a priori. Building on the well-celebrated Deep Q-Network (DQN), we propose a novel deep reinforcement learning algorithm robust to confounding biases in observed data. Specifically, our algorithm attempts to find a safe policy for the worst-case environment compatible with the observations. We apply our method to twelve confounded Atari games, and find that it consistently dominates the standard DQN in all games where the observed input to the behavioral and target policies mismatch and unobserved confounders exist.
Authors: Qi Feng
Abstract: Video-based spatial cognition is vital for robotics and embodied AI but challenges current Vision-Language Models (VLMs). This paper makes two key contributions. First, we introduce ViCA (Visuospatial Cognitive Assistant)-322K, a diverse dataset of 322,003 QA pairs from real-world indoor videos (ARKitScenes, ScanNet, ScanNet++), offering supervision for 3D metadata-grounded queries and video-based complex reasoning. Second, we develop ViCA-7B, fine-tuned on ViCA-322K, which achieves new state-of-the-art on all eight VSI-Bench tasks, outperforming existing models, including larger ones (e.g., +26.1 on Absolute Distance). For interpretability, we present ViCA-Thinking-2.68K, a dataset with explicit reasoning chains, and fine-tune ViCA-7B to create ViCA-7B-Thinking, a model that articulates its spatial reasoning. Our work highlights the importance of targeted data and suggests paths for improved temporal-spatial modeling. We release all resources to foster research in robust visuospatial intelligence.
Authors: Qi Feng
Abstract: While Multimodal Large Language Models (MLLMs) excel at general vision-language tasks, visuospatial cognition - reasoning about spatial layouts, relations, and dynamics - remains a significant challenge. Existing models often lack the necessary architectural components and specialized training data for fine-grained spatial understanding. We introduce ViCA2 (Visuospatial Cognitive Assistant 2), a novel MLLM designed to enhance spatial reasoning. ViCA2 features a dual vision encoder architecture integrating SigLIP for semantics and Hiera for spatial structure, coupled with a token ratio control mechanism for efficiency. We also developed ViCA-322K, a new large-scale dataset with over 322,000 spatially grounded question-answer pairs for targeted instruction tuning. On the challenging VSI-Bench benchmark, our ViCA2-7B model achieves a state-of-the-art average score of 56.8, significantly surpassing larger open-source models (e.g., LLaVA-NeXT-Video-72B, 40.9) and leading proprietary models (Gemini-1.5 Pro, 45.4). This demonstrates the effectiveness of our approach in achieving strong visuospatial intelligence with a compact model. We release ViCA2, its codebase, and the ViCA-322K dataset to facilitate further research.
Authors: Ander Artola Velasco, Stratis Tsirtsis, Nastaran Okati, Manuel Gomez-Rodriguez
Abstract: State-of-the-art large language models require specialized hardware and substantial energy to operate. As a consequence, cloud-based services that provide access to large language models have become very popular. In these services, the price users pay for an output provided by a model depends on the number of tokens the model uses to generate it -- they pay a fixed price per token. In this work, we show that this pricing mechanism creates a financial incentive for providers to strategize and misreport the (number of) tokens a model used to generate an output, and users cannot prove, or even know, whether a provider is overcharging them. However, we also show that, if an unfaithful provider is obliged to be transparent about the generative process used by the model, misreporting optimally without raising suspicion is hard. Nevertheless, as a proof-of-concept, we develop an efficient heuristic algorithm that allows providers to significantly overcharge users without raising suspicion. Crucially, we demonstrate that the cost of running the algorithm is lower than the additional revenue from overcharging users, highlighting the vulnerability of users under the current pay-per-token pricing mechanism. Further, we show that, to eliminate the financial incentive to strategize, a pricing mechanism must price tokens linearly on their character count. While this makes a provider's profit margin vary across tokens, we introduce a simple prescription under which the provider who adopts such an incentive-compatible pricing mechanism can maintain the average profit margin they had under the pay-per-token pricing mechanism. Along the way, to illustrate and complement our theoretical results, we conduct experiments with several large language models from the $\texttt{Llama}$, $\texttt{Gemma}$ and $\texttt{Ministral}$ families, and input prompts from the LMSYS Chatbot Arena platform.
Authors: Yu Zhang, Yuqi Xie, Huihan Liu, Rutav Shah, Michael Wan, Linxi Fan, Yuke Zhu
Abstract: Imitation learning advances robot capabilities by enabling the acquisition of diverse behaviors from human demonstrations. However, large-scale datasets used for policy training often introduce substantial variability in quality, which can negatively impact performance. As a result, automatically curating datasets by filtering low-quality samples to improve quality becomes essential. Existing robotic curation approaches rely on costly manual annotations and perform curation at a coarse granularity, such as the dataset or trajectory level, failing to account for the quality of individual state-action pairs. To address this, we introduce SCIZOR, a self-supervised data curation framework that filters out low-quality state-action pairs to improve the performance of imitation learning policies. SCIZOR targets two complementary sources of low-quality data: suboptimal data, which hinders learning with undesirable actions, and redundant data, which dilutes training with repetitive patterns. SCIZOR leverages a self-supervised task progress predictor for suboptimal data to remove samples lacking task progression, and a deduplication module operating on joint state-action representation for samples with redundant patterns. Empirically, we show that SCIZOR enables imitation learning policies to achieve higher performance with less data, yielding an average improvement of 15.4% across multiple benchmarks. More information is available at: https://ut-austin-rpl.github.io/SCIZOR/
Authors: Dimitrios Bralios, Paris Smaragdis, Jonah Casebeer
Abstract: Neural audio autoencoders create compact latent representations that preserve perceptually important information, serving as the foundation for both modern audio compression systems and generation approaches like next-token prediction and latent diffusion. Despite their prevalence, most audio processing operations, such as spatial and spectral up-sampling, still inefficiently operate on raw waveforms or spectral representations rather than directly on these compressed representations. We propose a framework that performs audio processing operations entirely within an autoencoder's latent space, eliminating the need to decode to raw audio formats. Our approach dramatically simplifies training by operating solely in the latent domain, with a latent L1 reconstruction term, augmented by a single latent adversarial discriminator. This contrasts sharply with raw-audio methods that typically require complex combinations of multi-scale losses and discriminators. Through experiments in bandwidth extension and mono-to-stereo up-mixing, we demonstrate computational efficiency gains of up to 100x while maintaining quality comparable to post-processing on raw audio. This work establishes a more efficient paradigm for audio processing pipelines that already incorporate autoencoders, enabling significantly faster and more resource-efficient workflows across various audio tasks.
Authors: Sahiti Yerramilli, Nilay Pande, Rynaa Grover, Jayant Sravan Tamarapalli
Abstract: This paper introduces GeoChain, a large-scale benchmark for evaluating step-by-step geographic reasoning in multimodal large language models (MLLMs). Leveraging 1.46 million Mapillary street-level images, GeoChain pairs each image with a 21-step chain-of-thought (CoT) question sequence (over 30 million Q&A pairs). These sequences guide models from coarse attributes to fine-grained localization across four reasoning categories - visual, spatial, cultural, and precise geolocation - annotated by difficulty. Images are also enriched with semantic segmentation (150 classes) and a visual locatability score. Our benchmarking of contemporary MLLMs (GPT-4.1 variants, Claude 3.7, Gemini 2.5 variants) on a diverse 2,088-image subset reveals consistent challenges: models frequently exhibit weaknesses in visual grounding, display erratic reasoning, and struggle to achieve accurate localization, especially as the reasoning complexity escalates. GeoChain offers a robust diagnostic methodology, critical for fostering significant advancements in complex geographic reasoning within MLLMs.
Authors: Rynaa Grover, Jayant Sravan Tamarapalli, Sahiti Yerramilli, Nilay Pande
Abstract: Multimodal Large Language Models (MLLMs) excel at high-level visual reasoning, but their performance on nuanced perceptual tasks remains surprisingly limited. We present HueManity, a benchmark designed to assess visual perception in MLLMs. The dataset comprises 83,850 images featuring two-character alphanumeric strings embedded in Ishihara test style dot patterns, challenging models on precise pattern recognition. Our evaluation of nine state-of-the-art MLLMs on HueManity demonstrates a significant performance deficit compared to human and traditional computer vision baselines. The best-performing MLLM achieved a 33.6% accuracy on the numeric `easy' task and a striking 3% on the alphanumeric `hard' task. In contrast, human participants achieved near-perfect scores (100% and 95.6%), and a fine-tuned ResNet50 model reached accuracies of 96.5% and 94.5%. These results highlight a critical gap in the visual capabilities of current MLLMs. Our analysis further explores potential architectural and training-paradigm factors contributing to this perceptual gap in MLLMs. We open-source HueManity dataset and code to foster further research in improving perceptual robustness of MLLMs.
Authors: Ke Wang, Yiming Qin, Nikolaos Dimitriadis, Alessandro Favero, Pascal Frossard
Abstract: Language models deployed in real-world systems often require post-hoc updates to incorporate new or corrected knowledge. However, editing such models efficiently and reliably-without retraining or forgetting previous information-remains a major challenge. Existing methods for lifelong model editing either compromise generalization, interfere with past edits, or fail to scale to long editing sequences. We propose MEMOIR, a novel scalable framework that injects knowledge through a residual memory, i.e., a dedicated parameter module, while preserving the core capabilities of the pre-trained model. By sparsifying input activations through sample-dependent masks, MEMOIR confines each edit to a distinct subset of the memory parameters, minimizing interference among edits. At inference, it identifies relevant edits by comparing the sparse activation patterns of new queries to those stored during editing. This enables generalization to rephrased queries by activating only the relevant knowledge while suppressing unnecessary memory activation for unrelated prompts. Experiments on question answering, hallucination correction, and out-of-distribution generalization benchmarks for LLaMA-3 and Mistral backbones demonstrate that MEMOIR achieves state-of-the-art performance across reliability, generalization, and locality metrics, scaling to thousands of sequential edits with minimal forgetting.
Authors: Haley Stone, Jing Du, Hao Xue, Matthew Scotch, David Heslop, Andreas Z\"ufle, Chandini Raina MacIntyre, Flora Salim
Abstract: Pathogen genome data offers valuable structure for spatial models, but its utility is limited by incomplete sequencing coverage. We propose a probabilistic framework for inferring genetic distances between unsequenced cases and known sequences within defined transmission chains, using time-aware evolutionary distance modeling. The method estimates pairwise divergence from collection dates and observed genetic distances, enabling biologically plausible imputation grounded in observed divergence patterns, without requiring sequence alignment or known transmission chains. Applied to highly pathogenic avian influenza A/H5 cases in wild birds in the United States, this approach supports scalable, uncertainty-aware augmentation of genomic datasets and enhances the integration of evolutionary information into spatiotemporal modeling workflows.
Authors: Mathukumalli Vidyasagar
Abstract: In this paper, we present a unified algorithm for stochastic optimization that makes use of a "momentum" term; in other words, the stochastic gradient depends not only on the current true gradient of the objective function, but also on the true gradient at the previous iteration. Our formulation includes the Stochastic Heavy Ball (SHB) and the Stochastic Nesterov Accelerated Gradient (SNAG) algorithms as special cases. In addition, in our formulation, the momentum term is allowed to vary as a function of time (i.e., the iteration counter). The assumptions on the stochastic gradient are the most general in the literature, in that it can be biased, and have a conditional variance that grows in an unbounded fashion as a function of time. This last feature is crucial in order to make the theory applicable to "zero-order" methods, where the gradient is estimated using just two function evaluations. We present a set of sufficient conditions for the convergence of the unified algorithm. These conditions are natural generalizations of the familiar Robbins-Monro and Kiefer-Wolfowitz-Blum conditions for standard stochastic gradient descent. We also analyze another method from the literature for the SHB algorithm with a time-varying momentum parameter, and show that it is impracticable.
Authors: Eman Alqudah, Ashfaq Khokhar
Abstract: Ensuring packet-level communication quality is vital for ultra-reliable, low-latency communications (URLLC) in large-scale industrial wireless networks. We enhance the Local Deadline Partition (LDP) algorithm by introducing a Graph Convolutional Network (GCN) integrated with a Deep Q-Network (DQN) reinforcement learning framework for improved interference coordination in multi-cell, multi-channel networks. Unlike LDP's static priorities, our approach dynamically learns link priorities based on real-time traffic demand, network topology, remaining transmission opportunities, and interference patterns. The GCN captures spatial dependencies, while the DQN enables adaptive scheduling decisions through reward-guided exploration. Simulation results show that our GCN-DQN model achieves mean SINR improvements of 179.6\%, 197.4\%, and 175.2\% over LDP across three network configurations. Additionally, the GCN-DQN model demonstrates mean SINR improvements of 31.5\%, 53.0\%, and 84.7\% over our previous CNN-based approach across the same configurations. These results underscore the effectiveness of our GCN-DQN model in addressing complex URLLC requirements with minimal overhead and superior network performance.
Authors: Dimitrios Bralios, Jonah Casebeer, Paris Smaragdis
Abstract: Neural audio codecs and autoencoders have emerged as versatile models for audio compression, transmission, feature-extraction, and latent-space generation. However, a key limitation is that most are trained to maximize reconstruction fidelity, often neglecting the specific latent structure necessary for optimal performance in diverse downstream applications. We propose a simple, post-hoc framework to address this by modifying the bottleneck of a pre-trained autoencoder. Our method introduces a "Re-Bottleneck", an inner bottleneck trained exclusively through latent space losses to instill user-defined structure. We demonstrate the framework's effectiveness in three experiments. First, we enforce an ordering on latent channels without sacrificing reconstruction quality. Second, we align latents with semantic embeddings, analyzing the impact on downstream diffusion modeling. Third, we introduce equivariance, ensuring that a filtering operation on the input waveform directly corresponds to a specific transformation in the latent space. Ultimately, our Re-Bottleneck framework offers a flexible and efficient way to tailor representations of neural audio models, enabling them to seamlessly meet the varied demands of different applications with minimal additional training.
Authors: Christopher F. Brown, Michal R. Kazmierski, Valerie J. Pasquarella, William J. Rucklidge, Masha Samsikova, Chenhui Zhang, Evan Shelhamer, Estefania Lahera, Olivia Wiles, Simon Ilyushchenko, Noel Gorelick, Lihui Lydia Zhang, Sophia Alj, Emily Schechter, Sean Askay, Oliver Guinan, Rebecca Moore, Alexis Boukouvalas, Pushmeet Kohli
Abstract: Unprecedented volumes of Earth observation data are continually collected around the world, but high-quality labels remain scarce given the effort required to make physical measurements and observations. This has led to considerable investment in bespoke modeling efforts translating sparse labels into maps. Here we introduce AlphaEarth Foundations, an embedding field model yielding a highly general, geospatial representation that assimilates spatial, temporal, and measurement contexts across multiple sources, enabling accurate and efficient production of maps and monitoring systems from local to global scales. The embeddings generated by AlphaEarth Foundations are the only to consistently outperform a suite of other well-known/widely accepted featurization approaches tested on a diverse set of mapping evaluations without re-training. We have released a dataset of global, annual, analysis-ready embedding field layers from 2017 through 2024.
Authors: Rubing Chen, Jiaxin Wu, Jian Wang, Xulu Zhang, Wenqi Fan, Chenghua Lin, Xiao-Yong Wei, Qing Li
Abstract: The increasing demand for domain-specific evaluation of large language models (LLMs) has led to the development of numerous benchmarks. These efforts often adhere to the principle of data scaling, relying on large corpora or extensive question-answer (QA) sets to ensure broad coverage. However, the impact of corpus and QA set design on the precision and recall of domain-specific LLM performance remains poorly understood. In this paper, we argue that data scaling is not always the optimal principle for domain-specific benchmark construction. Instead, we introduce Comp-Comp, an iterative benchmarking framework grounded in the principle of comprehensiveness and compactness. Comprehensiveness ensures semantic recall by covering the full breadth of the domain, while compactness improves precision by reducing redundancy and noise. To demonstrate the effectiveness of our approach, we present a case study conducted at a well-renowned university, resulting in the creation of PolyBench, a large-scale, high-quality academic benchmark. Although this study focuses on academia, the Comp-Comp framework is domain-agnostic and readily adaptable to a wide range of specialized fields. The source code and datasets can be accessed at https://github.com/Anya-RB-Chen/COMP-COMP.
Authors: Nilay Pande, Sahiti Yerramilli, Jayant Sravan Tamarapalli, Rynaa Grover
Abstract: A key frontier for Multimodal Large Language Models (MLLMs) is the ability to perform deep mathematical and spatial reasoning directly from images, moving beyond their established success in semantic description. Mathematical surface plots provide a rigorous testbed for this capability, as they isolate the task of reasoning from the semantic noise common in natural images. To measure progress on this frontier, we introduce MaRVL-QA (Mathematical Reasoning over Visual Landscapes), a new benchmark designed to quantitatively evaluate these core reasoning skills. The benchmark comprises two novel tasks: Topological Counting, identifying and enumerating features like local maxima; and Transformation Recognition, recognizing applied geometric transformations. Generated from a curated library of functions with rigorous ambiguity filtering, our evaluation on MaRVL-QA reveals that even state-of-the-art MLLMs struggle significantly, often resorting to superficial heuristics instead of robust spatial reasoning. MaRVL-QA provides a challenging new tool for the research community to measure progress, expose model limitations, and guide the development of MLLMs with more profound reasoning abilities.
Authors: Qiming Guo, Jinwen Tang, Xingran Huang
Abstract: We introduce Advertisement Embedding Attacks (AEA), a new class of LLM security threats that stealthily inject promotional or malicious content into model outputs and AI agents. AEA operate through two low-cost vectors: (1) hijacking third-party service-distribution platforms to prepend adversarial prompts, and (2) publishing back-doored open-source checkpoints fine-tuned with attacker data. Unlike conventional attacks that degrade accuracy, AEA subvert information integrity, causing models to return covert ads, propaganda, or hate speech while appearing normal. We detail the attack pipeline, map five stakeholder victim groups, and present an initial prompt-based self-inspection defense that mitigates these injections without additional model retraining. Our findings reveal an urgent, under-addressed gap in LLM security and call for coordinated detection, auditing, and policy responses from the AI-safety community.
Authors: Darius A. Faroughy, Manfred Opper, Cesar Ojeda
Abstract: Generative modeling of high-energy collisions at the Large Hadron Collider (LHC) offers a data-driven route to simulations, anomaly detection, among other applications. A central challenge lies in the hybrid nature of particle-cloud data: each particle carries continuous kinematic features and discrete quantum numbers such as charge and flavor. We introduce a transformer-based multimodal flow that extends flow-matching with a continuous-time Markov jump bridge to jointly model LHC jets with both modalities. Trained on CMS Open Data, our model can generate high fidelity jets with realistic kinematics, jet substructure and flavor composition.
Authors: Haichen Hu, David Simchi-Levi
Abstract: We study the problem of evaluating the excess risk of classical penalized empirical risk minimization (ERM) with Bregman losses. We show that by leveraging the recently proposed wild refitting procedure (Wainwright, 2025), one can efficiently upper bound the excess risk through the so-called "wild optimism," without relying on the global structure of the underlying function class. This property makes our approach inherently model-free. Unlike conventional analyses, our framework operates with just one dataset and black-box access to the training procedure. The method involves randomized vector-valued symmetrization with an appropriate scaling of the prediction residues and constructing artificially modified outcomes, upon which we retrain a second predictor for excess risk estimation. We establish high-probability performance guarantees both under the fixed design setting and the random design setting, demonstrating that wild refitting under Bregman losses, with an appropriately chosen wild noise scale, yields a valid upper bound on the excess risk. This work thus is promising for theoretically evaluating modern opaque ML and AI models such as deep neural networks and large language models, where the model class is too complex for classical learning theory and empirical process techniques to apply.
Authors: Tejaswini Medi, Hsien-Yi Wang, Arianna Rampini, Margret Keuper
Abstract: Latent generative models have shown remarkable progress in high-fidelity image synthesis, typically using a two-stage training process that involves compressing images into latent embeddings via learned tokenizers in the first stage. The quality of generation strongly depends on how expressive and well-optimized these latent embeddings are. While various methods have been proposed to learn effective latent representations, generated images often lack realism, particularly in textured regions with sharp transitions, due to loss of fine details governed by high frequencies. We conduct a detailed frequency decomposition of existing state-of-the-art (SOTA) latent tokenizers and show that conventional objectives inherently prioritize low-frequency reconstruction, often at the expense of high-frequency fidelity. Our analysis reveals these latent tokenizers exhibit a bias toward low-frequency information during optimization, leading to over-smoothed outputs and visual artifacts that diminish perceptual quality. To address this, we propose a wavelet-based, frequency-aware variational autoencoder (FA-VAE) framework that explicitly decouples the optimization of low- and high-frequency components. This decoupling enables improved reconstruction of fine textures while preserving global structure. Moreover, we integrate our frequency-preserving latent embeddings into a SOTA latent diffusion model, resulting in sharper and more realistic image generation. Our approach bridges the fidelity gap in current latent tokenizers and emphasizes the importance of frequency-aware optimization for realistic image synthesis, with broader implications for applications in content creation, neural rendering, and medical imaging.
Authors: Jian Yang, Jiahui Wu, Li Fang, Hongchao Fan, Bianying Zhang, Huijie Zhao, Guangyi Yang, Rui Xin, Xiong You
Abstract: Transforming road network data into vector representations using deep learning has proven effective for road network analysis. However, urban road networks' heterogeneous and hierarchical nature poses challenges for accurate representation learning. Graph neural networks, which aggregate features from neighboring nodes, often struggle due to their homogeneity assumption and focus on a single structural scale. To address these issues, this paper presents MSRFormer, a novel road network representation learning framework that integrates multi-scale spatial interactions by addressing their flow heterogeneity and long-distance dependencies. It uses spatial flow convolution to extract small-scale features from large trajectory datasets, and identifies scale-dependent spatial interaction regions to capture the spatial structure of road networks and flow heterogeneity. By employing a graph transformer, MSRFormer effectively captures complex spatial dependencies across multiple scales. The spatial interaction features are fused using residual connections, which are fed to a contrastive learning algorithm to derive the final road network representation. Validation on two real-world datasets demonstrates that MSRFormer outperforms baseline methods in two road network analysis tasks. The performance gains of MSRFormer suggest the traffic-related task benefits more from incorporating trajectory data, also resulting in greater improvements in complex road network structures with up to 16% improvements compared to the most competitive baseline method. This research provides a practical framework for developing task-agnostic road network representation models and highlights distinct association patterns of the interplay between scale effects and flow heterogeneity of spatial interactions.
Authors: Francesco Caravelli
Abstract: These notes offer a unified introduction to spectral methods for the study of complex systems. They are intended as an operative manual rather than a theorem-proof textbook: the emphasis is on tools, identities, and perspectives that can be readily applied across disciplines. Beginning with a compendium of matrix identities and inversion techniques, the text develops the connections between spectra, dynamics, and structure in finite-dimensional systems. Applications range from dynamical stability and random walks on networks to input-output economics, PageRank, epidemic spreading, memristive circuits, synchronization phenomena, and financial stability. Throughout, the guiding principle is that eigenvalues, eigenvectors, and resolvent operators provide a common language linking problems in physics, mathematics, computer science, and beyond. The presentation is informal, accessible to advanced undergraduates, yet broad enough to serve as a reference for researchers interested in spectral approaches to complex systems.
Authors: Yuming Li, Yikai Wang, Yuying Zhu, Zhongyu Zhao, Ming Lu, Qi She, Shanghang Zhang
Abstract: Recent advancements in aligning image and video generative models via GRPO have achieved remarkable gains in enhancing human preference alignment. However, these methods still face high computational costs from on-policy rollouts and excessive SDE sampling steps, as well as training instability due to sparse rewards. In this paper, we propose BranchGRPO, a novel method that introduces a branch sampling policy updating the SDE sampling process. By sharing computation across common prefixes and pruning low-reward paths and redundant depths, BranchGRPO substantially lowers the per-update compute cost while maintaining or improving exploration diversity. This work makes three main contributions: (1) a branch sampling scheme that reduces rollout and training cost; (2) a tree-based advantage estimator incorporating dense process-level rewards; and (3) pruning strategies exploiting path and depth redundancy to accelerate convergence and boost performance. Experiments on image and video preference alignment show that BranchGRPO improves alignment scores by 16% over strong baselines, while cutting training time by 50%.
Authors: Jack Wilkie, Hanan Hindy, Ivan Andonovic, Christos Tachtatzis, Robert Atkinson
Abstract: Malware classification is a contemporary and ongoing challenge in cyber-security: modern obfuscation techniques are able to evade traditional static analysis, while dynamic analysis is too resource intensive to be deployed at a large scale. One prominent line of research addresses these limitations by converting malware binaries into 2D images by heuristically reshaping them into a 2D grid before resizing using Lanczos resampling. These images can then be classified based on their textural information using computer vision approaches. While this approach can detect obfuscated malware more effectively than static analysis, the process of converting files into 2D images results in significant information loss due to both quantisation noise, caused by rounding to integer pixel values, and the introduction of 2D dependencies which do not exist in the original data. This loss of signal limits the classification performance of the downstream model. This work addresses these weaknesses by instead resizing the files into 1D signals which avoids the need for heuristic reshaping, and additionally these signals do not suffer from quantisation noise due to being stored in a floating-point format. It is shown that existing 2D CNN architectures can be readily adapted to classify these 1D signals for improved performance. Furthermore, a bespoke 1D convolutional neural network, based on the ResNet architecture and squeeze-and-excitation layers, was developed to classify these signals and evaluated on the MalNet dataset. It was found to achieve state-of-the-art performance on binary, type, and family level classification with F1 scores of 0.874, 0.503, and 0.507, respectively, paving the way for future models to operate on the proposed signal modality.
Authors: Xiangwei Shen, Zhimin Li, Zhantao Yang, Shiyi Zhang, Yingfang Zhang, Donghao Li, Chunyu Wang, Qinglin Lu, Yansong Tang
Abstract: Recent studies have demonstrated the effectiveness of directly aligning diffusion models with human preferences using differentiable reward. However, they exhibit two primary challenges: (1) they rely on multistep denoising with gradient computation for reward scoring, which is computationally expensive, thus restricting optimization to only a few diffusion steps; (2) they often need continuous offline adaptation of reward models in order to achieve desired aesthetic quality, such as photorealism or precise lighting effects. To address the limitation of multistep denoising, we propose Direct-Align, a method that predefines a noise prior to effectively recover original images from any time steps via interpolation, leveraging the equation that diffusion states are interpolations between noise and target images, which effectively avoids over-optimization in late timesteps. Furthermore, we introduce Semantic Relative Preference Optimization (SRPO), in which rewards are formulated as text-conditioned signals. This approach enables online adjustment of rewards in response to positive and negative prompt augmentation, thereby reducing the reliance on offline reward fine-tuning. By fine-tuning the FLUX model with optimized denoising and online reward adjustment, we improve its human-evaluated realism and aesthetic quality by over 3x.
Authors: Wenxuan Huang, Shuang Chen, Zheyong Xie, Shaosheng Cao, Shixiang Tang, Yufan Shen, Qingyu Yin, Wenbo Hu, Xiaoman Wang, Yuntian Tang, Junbo Qiao, Yue Guo, Yao Hu, Zhenfei Yin, Philip Torr, Yu Cheng, Wanli Ouyang, Shaohui Lin
Abstract: Unified multimodal understanding and generation models recently have achieve significant improvement in image generation capability, yet a large gap remains in instruction following and detail preservation compared to systems that tightly couple comprehension with generation such as GPT-4o. Motivated by recent advances in interleaving reasoning, we explore whether such reasoning can further improve Text-to-Image (T2I) generation. We introduce Interleaving Reasoning Generation (IRG), a framework that alternates between text-based thinking and image synthesis: the model first produces a text-based thinking to guide an initial image, then reflects on the result to refine fine-grained details, visual quality, and aesthetics while preserving semantics. To train IRG effectively, we propose Interleaving Reasoning Generation Learning (IRGL), which targets two sub-goals: (1) strengthening the initial think-and-generate stage to establish core content and base quality, and (2) enabling high-quality textual reflection and faithful implementation of those refinements in a subsequent image. We curate IRGL-300K, a dataset organized into six decomposed learning modes that jointly cover learning text-based thinking, and full thinking-image trajectories. Starting from a unified foundation model that natively emits interleaved text-image outputs, our two-stage training first builds robust thinking and reflection, then efficiently tunes the IRG pipeline in the full thinking-image trajectory data. Extensive experiments show SoTA performance, yielding absolute gains of 5-10 points on GenEval, WISE, TIIF, GenAI-Bench, and OneIG-EN, alongside substantial improvements in visual quality and fine-grained fidelity. The code, model weights and datasets will be released in: https://github.com/Osilly/Interleaving-Reasoning-Generation .
URLs: https://github.com/Osilly/Interleaving-Reasoning-Generation