Ax Mayur Kurup, Hyunjae Suh, Swathi Vaidyanathan, Pranesh Vyas, Srinidhi Madabhushi, Yegor Silyutin 22d ago

Deployment Risk Assessment Using Diff-Aware Features: A Case Study at Prime Video

Diff-aware ML model for predicting deployment risk at Prime Video to reduce unnecessary code freezes.

Ax Yashal Shakti Kanungo, Sumit Negi, Aruna Rajan 22d ago

Ad Headline Generation using Self-Critical Masked Language Model

Reinforcement learning application using policy gradient methods on masked language models for e-commerce ad headline generation.

Ax Aristotelis Papatheodorou, Pranav Vaidhyanathan, Natalia Ares, Ioannis Havoutis, Gerard J. Milburn 22d ago

CaLiSym: Learning Symplectic Dynamics of Real-World Systems through Structured Canonical Lifts

Framework extending symplectic learning to robotic systems with actuation and dissipation for stable dynamics prediction.

Ax Albert Zeyer, Ralf Schl\"uter, Hermann Ney 22d ago

Gradient-Based Speech-to-Text Alignment for Any ASR Model: From CTC to Speech LLMs

Gradient-based method for speech-to-text alignment compatible with CTC, transducer, and speech LLM models.

Ax Robert Gruhlke, Julius Berner, David Sommer, Lorenz Richter 22d ago

Tensor Train Diffusion: Leveraging Low-Rank Structures for High-Dimensional Score-Based Sampling

Low-rank tensor train approach for high-dimensional sampling in diffusion models using score-based methods.

Ax Michael King, Aravindh Mahendran, Matthew Koichi Grimes, Fedor Kitashov, Adham Elarabawy, Pedro Velez, Maks Ovsjanikov, Viorica P\u{a}tr\u{a}ucean 22d ago

Gen4U: Unifying Video Generation and Understanding via Diffusion

Analysis of video diffusion model representations revealing structured latent spaces across network depth and noise levels.

Ax Jie Wang 22d ago

Computing with Stochastic Oracles in AI-Augmented Computation

Theoretical framework modeling AI-augmented computation as probabilistic Turing machines interacting with stochastic oracles.

Ax Sojung An, Junha Lee, Sujeong You, Nam Ik Cho, Donghyun Kim 22d ago

LoCA: Spatially-Aware Low-Rank Convolutional Adaptation of Vision Foundation Models

Parameter-efficient fine-tuning method extending LoRA to vision foundation models with spatial awareness for downstream tasks.

Ax Denis Belomestny, Alexander Gasnikov, Egor Gladin, Alexey Naumov, Artemy Rubtsov, Yuri Sapronov, Daniil Tiapkin, Nikita Yudin 22d ago

Mathematical methods of reinforcement learning

Survey of mathematical foundations in reinforcement learning, covering MDPs, Bellman operators, and convergence guarantees for value/policy iteration algorithms.

Ax Seulbin Hwang, Kiyoung Om, Daejung Kim, Jinhan Lee 22d ago

Flow-ERD: Agent-type Aware Flow Matching with Entropy-Regularized Distillation for Diverse Traffic Simulation

Flow-ERD combines flow matching with entropy-regularized distillation for diverse and realistic multi-agent traffic simulation in autonomous driving.

Ax Ruilin Tong, Dong Gong 22d ago

MILES: Modular Instruction Memory with Learnable Selection for Self-Improving LLM Reasoning

MILES proposes modular instruction memory with learnable selection enabling LLMs to accumulate and reuse reasoning experience across sequential problems.

Ax Hao Kong, Di Liu, Shuo Huai, Xiangzhong Luo, Ravi Subramaniam, Christian Makaya, Qian Lin, Weichen Liu 22d ago

EdgeCompress: Coupling Multidimensional Model Compression and Dynamic Inference for EdgeAI

EdgeCompress framework combining dynamic image cropping and model compression techniques to reduce CNN computational overhead for resource-constrained embedded devices.

Ax Roxana Barrios, Ioannis Sgouralis 22d ago

Tensorized algorithms and scalable filtering methods for hidden Markov and factorial hidden Markov models

Tensorized algorithms and scalable filtering methods for hidden Markov models and factorial HMMs to efficiently represent multi-factor time-series systems.

Ax Jiabin Shen, Guang Chen, Chengjun Mao 22d ago

Behavior Leverage Imbalance in Multi-Teacher On-Policy Distillation

Analysis of behavior shift in multi-teacher on-policy distillation for LLM agents learning tool use, showing invisible distribution shifts in aggregate losses.

Ax Beomgu Kang, Hyunseok Seo 22d ago

Prior-aware and Context-guided Group Sampling for Active Probabilistic Subsampling

Active deep probabilistic subsampling approach with prior-aware context-guided group sampling for reducing measurement overhead and data transfer.

Ax Miguel Lopez-Duran, Elena Marrero, Julian Fierrez, Marta Robledo-Moreno, Ruben Vera-Rodriguez, Daniel DeAlcala, Aythami Morales, Ruben Tolosana, Oscar Delgado, Alvaro Ortigosa, Javier Ortega-Garcia 22d ago

Comparative Study of Domain-adapted VLMs for General Document Visual Question Answering

Comparative evaluation of 8 open-source VLMs on document visual question answering, assessing robustness and transferability across document domains.

Ax T-H. Hubert Chan, Elaine Shi, Mengshi Zhao, Mingxun Zhou 22d ago

Continual Learning With Participation Privacy: An Auditable Buffering-Aggregation Recipe

Privacy-preserving continual learning framework with auditable buffering-aggregation recipe for federated and streaming systems under adaptive interaction.

Ax Sergio Rozada, Yiming Qin, Manuel Madeira, Pascal Frossard, Alejandro Ribeiro 22d ago

DiPhon: Diffusion on Graphons for Scalable Graph Generation

DiPhon applies diffusion models to large-scale graph generation using graphon theory to study structural statistics across different node scales.

Ax Sergei Zorkaltsev, Maciej Haranczyk, Christina Schenk 22d ago

Bayesian Optimization of Genetic Algorithm Hyperparameters in a Multi-Fidelity Framework for Efficient Lattice Material Design

Multi-fidelity Bayesian optimization framework for tuning genetic algorithm hyperparameters using FFT, CNN surrogates, and Gaussian processes for lattice material design.

Ax Diab W. Abueidda, Bilal Ahmed, Panos Pantidis, Mostafa E. Mobasher 22d ago

Physics-Audited Agentic Discovery in Scientific Machine Learning

PA-SciML introduces physics-audited verification workflow for LLM agents discovering scientific surrogate models, ensuring predictions satisfy physical constraints beyond error metrics.

Ax Yutang Ma, Kecheng Huang, Xikun Jiang, Zili Shao 22d ago

TF-Engram: A Train-Free Engram with SSD-Backed Memory for Large Language Models

TF-Engram proposes SSD-backed memory architecture for LLMs to enable efficient knowledge expansion without retraining, using engram-style hidden-state injection.

Ax Tanay Sodha, Aditya Sharma, Ramya Hebbalaguppe, Vinti Agarwal, Pranav Murthy Yeluripaty 22d ago

When Prompts Ignore Structure: Graph-Based Attribute Reasoning for Calibrated VLMs

ARGTCA improves confidence calibration in vision-language models by using graph-based attribute reasoning to represent class relationships, addressing overconfidence from prompt tuning.

Ax Jieying Wang, Shuyuan Fan, Mingkai Zheng, Zhao Zhang 22d ago

GIFT: Geometry-Informed Low-precision Gradient Communication for LLM Pretraining

GIFT: geometry-informed low-precision gradient quantization for distributed LLM pretraining reducing communication bottlenecks.

Ax Yazdan Jamshidi, Alexey Shvets 22d ago

PALS: Percentile-Aware Layerwise Sparsity for LLM Pruning

PALS: layer-aware sparsity pruning for LLMs using percentile-based activation thresholds, improving perplexity over uniform pruning methods.

Ax Grzegorz Brzezinka 22d ago

Does Bielik Know What It Doesn't Know? Activation Dispersion Separates Entity Familiarity from Factual Reliability Across Model Scale

Studies activation patterns in Polish Bielik LLMs to identify entity familiarity and factual reliability signals before answer generation across model scales.

Ax Hyunjae Kim, Dain Kim, Pan Xiao, Serina S. Applebaum, Younjoon Chung, Xuguang Ai, Yu Yin, Roy Jiang, Yuexi Du, Yawen Wei, Yiming Kong, Tuo Guo, Zhiyuan Cao, Mengmeng Du, Yuelei Fu, Yan Hu, Rui Shi, Gui Yang, Kevin W. Jin, Yuntian Liu, Yuxuan Tian, Jonathan Marquez, Zhen Chen, Sheng Zhang, Hoifung Poon, Hua Xu, Jaewoo Kang, Qingyu Chen 22d ago

MedPMC: A Systematic Framework for Scaling High-Fidelity Medical Multimodal Data for Foundation Models

MedPMC framework scales high-fidelity medical multimodal data from PubMed Central for foundation model development with clinical validation.

Ax Eitan Levin, Venkat Chandrasekaran 22d ago

Any-Dimensional Learning by Sampling

Studies generalization of variable-size input models (point clouds, sequences, graphs) from small training sizes to unseen larger sizes.

Ax Yair Feldman, Linxi Zhao, Nathan Godey, Dongyoung Go, Yilun Hua, Kilian Q. Weinberger, Jennifer J. Sun, Yoav Artzi 22d ago

Co-LMLM: Continuous-Query Limited Memory Language Models

Co-LMLM externalizes factual knowledge to continuous-query knowledge bases during LLM generation, enabling knowledge control beyond conventional weight-based approaches.

Ax Jiancong Xiao, Yanbo Fan, Ruoyu Sun, Zhi-Quan Luo 22d ago

Adversarial Rademacher Complexity of Deep Neural Networks

Analyzes adversarial Rademacher complexity of DNNs to provide generalization guarantees for models robust to adversarial perturbations.

Ax Yilei Chen, Vittorio Giammarino, James Queeney, Ioannis Ch. Paschalidis 22d ago

Provably Efficient Off-Policy Adversarial Imitation Learning with Convergence Guarantees

Studies off-policy adversarial imitation learning convergence and sample complexity, showing sample reuse improves efficiency without importance sampling correction.

Ax Yuelin Xu, Xiao Zhang 22d ago

How Learning Dynamics Drive Adversarially Robust Generalization?

Analyzes robust overfitting in adversarial training using dynamical systems and PAC-Bayesian theory, providing mechanistic interpretation of generalization in robust models.

Ax Jinho Chang, Changsun Lee, Hyungjin Chung, Jong Chul Ye 22d ago

ContrastiveCFG: Guiding Diffusion Sampling by Contrasting Positive and Negative Concepts

ContrastiveCFG improves diffusion model sampling by contrasting positive and negative concepts, addressing limitations of naive negative prompting in classifier-free guidance.

Ax Yuchen Lin, Yong Zhang, Sihan Feng, Hong Zhao 22d ago

Network Dynamics-Based Framework for Understanding Deep Neural Networks

Proposes theoretical framework analyzing deep neural network learning dynamics through dynamical systems theory, introducing order-preserving transformations at neuron level.

Ax Songqiao Hu, Zeyi Liu, Lufeng Hao, Yinzhong Cheng, Xiao He 22d ago

PB-OEL: A Performance-Bounded Online Ensemble Learning Framework With Mixed Feedback for Real-Time Safety Assessment

PB-OEL framework for real-time safety assessment using online ensemble learning with mixed feedback and performance guarantees under concept drift.

Ax Zhiqiang He, Zhi Liu 22d ago

Silent Neuron Theory and Plasticity Preservation for Deep Reinforcement Learning in Adaptive Video Streaming

Studies silent neurons and plasticity in deep reinforcement learning for adaptive video streaming to improve generalization across heterogeneous network conditions.

Ax Sameera Ramasinghe, Thalaiyasingam Ajanthan, Gil Avraham, Yan Zuo, Alexander Long 22d ago

Protocol Models: Scaling Decentralized Training with Communication-Efficient Model Parallelism

Protocol Models proposes communication-efficient model parallelism for decentralized training by compressing activations and activation gradients in distributed deep learning setups.

Ax Boris van Breugel, Yelysei Bondarenko, Paul Whatmough, Markus Nagel 22d ago

FPTQuant: Function-Preserving Transforms for LLM Quantization

FPTQuant introduces function-preserving transforms to enable efficient quantization of LLMs by handling magnitude outliers, improving inference efficiency without significant performance degradation.

Ax Changcun Huang 22d ago

Understanding Two-Layer Neural Networks with Smooth Activation Functions

Theoretical analysis of two-layer neural networks with smooth activation functions using Taylor series and spline methods.

Ax Luis Roque, Vitor Cerqueira, Carlos Soares, Luis Torgo 22d ago

L-GTA: Latent Generative Modeling for Time Series Augmentation

L-GTA: VAE-based generative model for time series augmentation using Bi-LSTM and temporal self-attention.

Ax Seunghyeon Kim, Taesun Yeom, Jinho Kim, Wonpyo Park, Kyuyeun Kim, Jaeho Lee 22d ago

Activation Quantization of Vision Encoders Needs Prefixing Registers

Quantization technique for vision encoders using register prefixing to handle outliers and reduce inference cost.

Ax Zhiwei Han, Stefan Matthes, Hao Shen 22d ago

Provable Affine Identifiability of Nonlinear CCA under Latent Distributional Priors

Theoretical analysis proving affine identifiability in nonlinear CCA under latent distributional priors.

Ax Rameen Mahmood, Tousif Ahmed, Sai Teja Peddinti, Danny Yuxing Huang 22d ago

What's on My Network? Using Large Language Models to Identify Real-World IoT Devices at Scale

LLM-based semantic inference pipeline for identifying IoT devices from network traffic metadata at scale.

Ax Bowen Wang, Matteo Zecchin, Osvaldo Simeone 22d ago

Distributed Dynamic Associative Memory via Online Convex Optimization

Distributed associative memory system for multiple agents with time-varying data streams based on online optimization.

Ax Haozhe Wu 22d ago

FDRMFL: Multimodal Federated Feature Extraction Model Based on Information Maximization and Contrastive Learning

Federated feature extraction framework for multimodal regression with non-IID data using information maximization.

Ax Aryan Karmore 22d ago

ButterflyMoE: Compression-Scalable Ternary Experts via Structured Butterfly Orbits

ButterflyMoE compression method for Mixture of Experts architectures achieving memory-efficient scaling.

Ax Shovito Barua Soumma, Asiful Arefeen, Stephanie M. Carpenter, Melanie Hingle, Hassan Ghasemzadeh 22d ago

Counterfactual Modeling with Fine-Tuned LLMs for Health Intervention Design and Sensor Data Augmentation

Framework using fine-tuned LLMs to generate counterfactual explanations for health interventions and data augmentation.

Ax Jaime Vale, Vanessa Freitas Silva, Maria Eduarda Silva, Fernando Silva 22d ago

Synthetic Time Series Generation via Complex Networks

Method for generating synthetic time series using complex network representations to address privacy and data availability.

Ax Magamed Taimeskhanov, Samuel Vaiter, Damien Garreau 22d ago

Towards Understanding Steering Strength

Research on understanding how to set the magnitude of latent representation steering for controlling LLM behavior.

Ax Chongyi Zheng, Royina Karegoudra Jayanth, Benjamin Eysenbach 22d ago

Can We Really Learn One Representation to Optimize All Rewards?

Theoretical analysis of unsupervised representation learning in reinforcement learning via forward-backward methods.

Ax Haokun Liu, Gyung Hyun Je, Marco Ciccone, Zhenlin Xu, Prasanth YSS, Colin Raffel 22d ago

The Appeal and Reality of Recycling LoRAs with Adaptive Merging

Study of adaptive merging methods for combining fine-tuned LoRA modules to improve model performance across tasks.