Ax Ning Liu, Kalle Kujanp\"a\"a, Zhaoxuan Zhu, P Aditya Sreekar, Kaiwen Liu, Chuanneng Sun, Jorge Marchena Menendez, Matthew Bales, Tianyu Yang, Shahnawaz Alam, Rose Yu, Baoyuan Liu, Kristina Klinkner, Shervin Malmasi 16d ago

Eluna: An Agentic LLM System for Automating Warehouse Operations with Reasoning and Task Execution

Production-deployed agentic LLM system for warehouse operations using graph-guided multi-agent framework for SOP execution.

Ax Berkay Anahtarci 16d ago

NL-PAC: Specification Ambiguity and Certified Minimax Risk Floors in LLM-Mediated Supervision

Framework for handling specification ambiguity in LLM-based labeling with certified minimax risk bounds.

Ax Bishmoy Paul, Youngmin Yi, Hoeseok Yang 16d ago

Sensitivity-Aware Thresholding and Token Routing for Activation Sparsification in Large Language Models

Method for efficient LLM inference via MLP activation sparsification and token routing using sensitivity-aware thresholding.

Ax Ajay Narayanan Sridhar, Ronak Singh, Mehrdad Mahdavi, Vijaykrishnan Narayanan 16d ago

Correlation-Aware Contextual Bandits with Surrogate Rewards for LLM Routing

Contextual bandit algorithm with correlated arms for LLM routing using surrogate reward signals from learned models.

Ax Chaoran Cheng, Zhanghan Ni, Yanru Qu, Yuxin Chen, Ruihan Guo, Jiajun Fan, Ge Liu 16d ago

Variable-Length Generative Protein Design via Generalized Poisson Flow

Introduces Generalized Poisson Flow for variable-length protein generation without requiring pre-specified length via flow-based models.

Ax Iris Xu, Sunshine Jiang, John Marangola, Nitish Dashora, Richard Li, Thomas Liu, Zexue He, Yuheng Zhi, Alex Pentland, Pulkit Agrawal, Zhang-Wei Hong 16d ago

Learning More from Less: Reinforcement Learning from Hindsight

Improves sample efficiency in vision-language-action model RL post-training by learning from hindsight on sparse-reward manipulation tasks.

Ax Alexander Tian, Aditya Ghai, Sanjit Neelam, Zaal Vasania, Akshay Mishra 16d ago

COBS: Cumulant Order Block Sparse Attention

Analyzes block sparse attention in LLMs as KV cache optimization, studying DeepSeek's Native Sparse Attention block selection mechanism.

Ax Shuo Huai, Hao Kong, Shiqing Li, Xiangzhong Luo, Ravi Subramaniam, Christian Makaya, Qian Lin, Weichen Liu 16d ago

EvoLP: Self-Evolving Latency Predictor for Model Compression in Real-Time Edge Systems

EvoLP framework for predicting neural network inference latency on edge devices to optimize model compression for real-time deployment.

Ax Carl Hvarfner, Sam Daulton, Max Balandat, Eytan Bakshy 16d ago

Pitfalls and Remedies for Multi-Task Bayesian Optimization

Identifies pitfalls in multi-task Bayesian optimization with Gaussian processes and proposes remedies for cross-task correlation estimation.

Ax Linhui Xiao, Guiping Cao, Mingyue Guo, Xianchao Guan, Fan Yang, Ming Tao, Xin Li, Yuxin Peng, Yaowei Wang 16d ago

A Survey on the Green Development of Large Models: From Resource-Efficient Architectures to Hardware-Software Co-Design

Comprehensive survey on resource-efficient large model architectures and hardware-software co-design for sustainability and computational efficiency.

Ax Xu Liu, Kai Wan, Zihao Lu 16d ago

EXHOLD: Experience-Aware Real-Time Hold Control for Large-Scale Ride-Hailing Matching at DiDi

Industrial hold control system for ride-hailing using experience-aware optimization to reduce cancellations and improve matching.

Ax Lucas Pinto 16d ago

Present but Rescaled: Chat-to-Agent Transfer of Additive Activation Steering

Studies transfer of additive activation steering from chat to ReAct agent deployment in LLMs with behavioral measurement and representation analysis.

Ax Jiseok Chae, Donghwan Kim 16d ago

Understanding Schedule-Free Methods in Nonconvex Optimization: Rate Guarantees and Escaping Saddles

Provides convergence theory and worst-case analysis for schedule-free optimization methods in nonconvex settings without manual learning rate tuning.

Ax Julius St\"ork 16d ago

Interference and Retention in Continual Learning

Studies task interference and forgetting in continual learning through interference energy and path-averaged curvature in deep networks.

Ax Konrad \"Ozdemir, Julia Gastinger, Lukas Kirchdorfer, Heiner Stuckenschmidt 16d ago

Temporal Knowledge Graph Forecasting under Distribution Shifts: A Synthetic Evaluation

Evaluates temporal knowledge graph forecasting models under distribution shifts using synthetic data generation to test robustness.

Ax Jayadeva, Madhur Aswani 16d ago

All you need is SAMPAT

SAMPAT: three-layer interpretable neural architecture using multivariate polynomials for scientific data analysis with provable continuous function learning.

Ax Amit Peleg, Naman Deep Singh, Naama Pearl, Bibhabasu Mohapatra, Matthias Hein 16d ago

Forget Narrowly, Retain Broadly: Unlearning as an Asymmetric Generalization Problem

Machine unlearning in LLMs with improved evaluation metrics detecting under-forgetting through paraphrasing and over-forgetting through semantic probes.

Ax Kris Atallah (New York University, New York, USA) 16d ago

LionVote: Per-Layer Learning Rate Adaptation for Lion

Per-layer learning rate adaptation mechanism for Lion optimizer addressing 2.6-2.8x scaling disparity across attention, MLP, and normalization layers.

Ax Federico Ottomano, Gaopeng Ren, Yingzhen Li, Kim E. Jelfs, Alex M. Ganose 16d ago

Autoregressive latent diffusion for 3D molecule generation

Autoregressive latent diffusion approach for 3D molecule generation supporting variable-length generation and partial context conditioning.

Ax Ivan Ilin, Philip Zmushko, Peter Richt\'arik 16d ago

Super-Tuning: From Activation-Aware Pruning to Sparse Fine-Tuning

Super: sparse parameter-efficient fine-tuning method for LLMs using pruning saliency signals to select trainable parameters, reducing memory and storage.

Ax Pedro P. Santos, F\'abio Vital, Alberto Sardinha, Francisco S. Melo 16d ago

Risk-Aware General-Utility Markov Decision Processes

Risk-aware Markov Decision Processes framework enabling agents to optimize risk measures of objective value distributions while trading off expected performance.

Ax Guanquan Wang, Yoshimasa Tsuruoka 16d ago

Shortcut Trajectory Planning for Efficient Offline Reinforcement Learning

Shortcut Trajectory Planning: consistency-based offline RL planner reducing iterative denoising cost without two-stage distillation pipeline.

Ax Samra Sana, Giorgio Mantica, Saul Imbrici 16d ago

Graph Neural Networks for Scalable and Transferable Node Centrality Approximation

Graph Neural Networks for scalable approximation of betweenness and closeness centrality, studying transferability across different graphs.

Ax Foundation Model Team 16d ago

Mach-Mind-4-Flash Technical Report

Mach-Mind-4-Flash: 35B-parameter Mixture-of-Experts agentic model with 3B activated parameters achieving 100B-class performance through post-training and agentic RL.

Ax Rodolphe Barlogis, Ferhat Tamssaouet, Quentin Falcoz, St\'ephane Grieu 16d ago

Learning Physics-Informed Surrogate Model of Linear Elastic Displacement Fields from Geometry

Physics-informed DeepONet surrogate model for predicting elastic displacement fields in fractured domains without finite-element training data.

Ax Wout Mommen, Lars Keuninckx, Matthias Hartmann, Werner Van Leekwijck, Piet Wambacq 16d ago

Fully Trainable Deep Differentiable Logic Gate Networks and Lookup Table Networks

Novel training method for optimizing connections and gate types in deep differentiable logic gate networks and lookup table networks using probabilistic selection.

Ax Avik Bhatnagar, Anton Paule, Tobias Schuermann, Sebastian Reiter, Oliver Bringmann 16d ago

On-Device Adaptive Battery Power Prediction for Electric Vehicles

On-device adaptive learning approach for battery power prediction in electric vehicles using continuous fine-tuning under distribution shift.

Ax Ofir Kruzel, Itzik Klien 16d ago

Data-Efficient Deep Learning: Empirical Guidelines for Training Set Size Estimation in Inertial Sensor Classification

Empirical guidelines for determining minimum training set size for deep learning models in inertial sensor classification tasks like activity recognition.

Ax Nicolai Dinh Khang Truong, Richard R\"ottger 16d ago

SYNRARE: Synthetic Rare Disease EHR Generation for ML Benchmarking

Synthetic data generation method for creating electronic health records of rare diseases to address privacy concerns in ML-based diagnosis research.

Ax Nick Whiteley 16d ago

Similarity search generalisation in contrastive learning with InfoNCE loss

Theoretical analysis of contrastive learning with InfoNCE loss, proving population risk converges as O(1/k) with k negative samples for similarity search.

Ax Edwin De Nicolo, Rahul Marchand, Cornelius Carlsson, Pranav Vaidhyanathan, Natalia Ares 16d ago

Action-Factored Multi-Agent Reinforcement Learning for Scalable Quantum Device Tuning

Multi-agent reinforcement learning approach using factored action representation for tuning quantum-dot arrays with exploitable local structure.

Ax Bruno Abrahao 16d ago

Neural Collapse Is Forbidden: Information Floors in Language Models

Research on within-class variance in language model representations, arguing it represents allocated information storage rather than incomplete neural collapse across 14 models.

Ax Pan Li 16d ago

All Explanations are Wrong, But Many Are Useful: Exploring the Rashomon Explanation Set with Large Language Models

Explores Rashomon explanation sets using LLMs to generate multiple valid model explanations, decoupling explanation from prediction as complementary objectives.

Ax Refat Ishrak Hemel, Ehsan Hallaji, Roozbeh Razavi-Far 16d ago

TSAI-MetaFraud: A Benchmark Dataset for Financial Fraud Transaction and Behavioral Risk Detection in Metaverse Ecosystems

TSAI-MetaFraud benchmark dataset for multimodal fraud and behavioral risk detection in metaverse virtual economies.

Ax Qitai Tan, Ruiwen Gu, Yilin Su, Mo Li, Xu Lin, Xiao-Ping Zhang 16d ago

GatedLinear: Adaptive Routing of Complementary Linear Bases for Time Series Forecasting

GatedLinear implements adaptive routing across complementary linear basis functions for time series forecasting, addressing diverse temporal dynamics.

Ax Gabriel Mahuas, Victoria Shevchenko, Ugo Tanielian, Yassir Bendou, Richard Gao 16d ago

CoCoT-EEG: Contrastive-Pretrained Multiscale Convolutional Transformer for EEG Decoding

CoCoT-EEG uses contrastive pretraining with multiscale convolutional transformers for improved EEG signal decoding, replacing masked reconstruction approaches.

Ax Cl\'audio L\'ucio do Val Lopes, Lucca Machado da Silva 16d ago

Semantic Pareto-DQN: A Multi-Objective Reinforcement Learning Framework for Financial Anomaly Detection

Semantic Pareto-DQN applies multi-objective reinforcement learning to financial fraud detection, addressing class imbalance without data resampling.

Ax Shilin Ou, Yifan Xu, Luyao Zhang 16d ago

SolarChain-Eval: A Physics-Constrained Benchmark for Trustworthy Economic Agents in Decentralized Energy Markets

SolarChain-Eval is a physics-constrained benchmark for evaluating trustworthiness and performance of autonomous agents in decentralized energy markets.

Ax Merkouris Papamichail, Konstantinos Varsos, Giorgos Flouris, Jo\~ao Marques-Silva 16d ago

Interval Certifications for Multilayered Perceptrons via Lattice Traversal

Theoretical framework reducing adversarial robustness verification to lattice traversal problems for multilayer perceptrons, advancing AI safety certification methods.

Ax Micah Zhang 16d ago

HALO: Hybrid Adaptive Latent Reasoning for Language Models

HALO adds adaptive latent refinement layers to frozen pretrained language models, enabling selective extra computation for improved transfer learning efficiency.

Ax Joshua Pickard, Wei Qi, Na Li, Ann Woolley, Lisa Cosimi, Roy Kishony, Deborah Hung 16d ago

EHR-MPC: Inference-Time Control for Sepsis Treatment with Generative Patient Digital Twins

EHR-MPC framework uses generative patient digital twins for inference-time adaptive sepsis treatment optimization, decoupling dynamics learning from policy optimization.

Ax Hyunjin Seo, Hyeon Hwang, Gyubok Lee, Jay Shin, Jimin Park, Taesoo Kim, Sanghoon Lee, Hongjoon Ahn, Sungjun Han, Sangwon Jung 16d ago

TheBioCollection: Unified Pre-Training Scale LLM Corpus for Biology

TheBioCollection unifies scattered biological databases and resources into a cohesive pretraining corpus for biology-focused large language models.

Ax Dominick Reilly, Qiyu Wu, Hiromi Wakaki, Srijan Das, Yuki Mistufuji 16d ago

Mixture of Probes: Learning from Privileged Modalities in Multimodal LLMs Through Probing

Mixture of Probes technique allows multimodal LLMs to learn from auxiliary modalities available only during training, improving inference-time performance with limited modalities.

Ax Michael Murray, Daphne Chen, Simran Bagaria, Dean Fortier, Tess Hellebrekers, Galen Mullins, Harshavardhan Gajarla, Oier Mees, Maya Cakmak, Andrey Kolobov 16d ago

FlowDAgger: Human-in-the-Loop Adaptation of Generative Robot Policies in Latent Space

FlowDAgger enables human-in-the-loop adaptation of pretrained generative robot policies using diffusion models, allowing rapid real-world deployment without large-scale retraining.

Ax Maureese Williams, Dymitr Nowicki 16d ago

GATS: Graph-Augmented Tree Search with Layered World Models for Efficient Agent Planning

GATS: Planning framework for LLM agents combining tree search with layered world models to reduce inference costs during planning.

Ax Alex Beaudin, Hanna Krasowski, Eric Palanques-Tost, Calin Belta, Murat Arack 16d ago

Learning-enabled Parameter Synthesis for Nonlinear Systems from Signal Temporal Logic

Parameter synthesis for nonlinear systems satisfying Signal Temporal Logic specifications using gradient-based optimization.

Ax Xin Che, Lingyang Chu, Qiqi Zhang, Xinyu Ma, Xuan Luo, Jian Pei 16d ago

RaMark: Radioactive Watermarking for Generated Tabular Data

Watermarking method for generated tabular data that resists retraining attacks for ownership verification.

Ax Shikhar Bharadwaj, Kwanghee Choi, Stephen McIntosh, Chin-Jou Li, Eunjung Yeo, Daisuke Saito, Nobuaki Minematsu, Shinji Watanabe, Jian Zhu, David Harwath, David R. Mortensen 16d ago

Phone Segmentation and Recognition through Phonological Activation Mapping

Phone segmentation and recognition using phonological activation mapping from self-supervised speech models.

Ax Pulkit Madan, Sanjay Haresh, Reza Ebrahimi, Sunny Panchal, Apratim Bhattacharyya, Roland Memisevic 16d ago

On Locality and Length Generalization in Visual Reasoning

Analysis of local sequential vision models versus global models, investigating computational benefits for visual reasoning tasks.

Ax Tanapol Nuatho, Narisorn Sangnakara, Prapong Prechaprapranwong, Rajchawit Sarochawikasit 16d ago

Quantum-Enhanced Synthetic Data Generation Using Quantum Circuit Born Machines for Imbalanced Tabular Learning

Quantum-classical hybrid framework using quantum circuit born machines for synthetic data generation in imbalanced learning.