Isolater - Feed

Ax Giang Son Nguyen, Zi Pong Lim, Sarthak Ketanbhai Modi, Yon Shin Teo, Wenya Wang 19d ago

An Online Reference-Free Evaluation Framework for Flowchart Image-to-Code Generation

Reference-free evaluation framework for flowchart image-to-code generation using vision-language models without ground truth.

Ax Bowen Yu, Maolin Wang, Sheng Zhang, Binhao Wang, Yi Wen, Jingtong Gao, Bowen Liu, Zimo Zhao, Wanyu Wang, Xiangyu Zhao 19d ago

Curriculum Learning for Efficient Chain-of-Thought Distillation via Structure-Aware Masking and GRPO

Curriculum learning framework for distilling chain-of-thought reasoning from large to compact student models via structure-aware masking.

Ax Theophilus Amaefuna, Hitesh Vaidya, Anshuman Chhabra, Ankur Mali 19d ago

Curvature-Weighted Capacity Allocation: A Minimum Description Length Framework for Layer-Adaptive Large Language Model Optimization

MDL-based framework for layer-wise capacity allocation and pruning in large language models using curvature-aware scoring.

Ax Mumuksh Tayal, Manan Tayal, Ravi Prakash 19d ago

Safe Flow Q-Learning: Offline Safe Reinforcement Learning with Reachability-Based Flow Policies

Safe Flow Q-Learning extends offline RL with Hamilton-Jacobi reachability for safety-critical control under constraints.

Ax Dibakar Sigdel 19d ago

The Phasor Transformer: Resolving Attention Bottlenecks on the Unit Circle

Phasor Transformer using phase-native unit circle representation to reduce quadratic attention bottleneck in long-context sequence learning.

Ax Zhiyuan Chen, Yuxuan Zhong, Fan Wang, Bo Yu, Pengtao Shao, Shaoshan Liu, Ning Ding 19d ago

StateLinFormer: Stateful Training Enhancing Long-term Memory in Navigation

StateLinFormer: linear-attention transformer with stateful training for long-term memory in navigation tasks.

Ax Yongchao Wu, Aron Henriksson 19d ago

Persona Matters: Effects of Activation Steering on Short Answer Generation and Scoring

Study of activation-based persona steering on LLM short-answer generation and automated scoring across architectures.

Ax Yvon Apedo, Martyna Poreba, Michal Szczepanski, Samia Bouchafa 19d ago

Beyond Attention Scores: SVD-Based Vision Token Pruning for Efficient Vision-Language Models

SVD-based vision token pruning method for efficient Vision-Language Models addressing computational bottlenecks in long sequences.

Ax Shi Feng, Hanlin Zhang, Fan Nie, Sham Kakade, Yiling Chen 19d ago

Peer-Predictive Self-Training for Language Model Reasoning

Peer-Predictive Self-Training (PST) framework enables collaborative self-improvement of language models without external supervision using cross-model aggregation.

Ax Peter Walther, Hongrui Sheng, Xinxin Liu, Bin Feng, Reid Coyle, Xinhua Yan, Kyle Smith, Harrison Kayal, Shyam Chand Pal, Zhiling Zheng 19d ago

Predicting Scale-Up of Metal-Organic Framework Syntheses with Large Language Models

Fine-tuned LLM with positive-unlabeled learning for predicting metal-organic framework synthesis scale-up from literature, achieving 93.5% accuracy.

Ax Bingxi Zhao, Jiahao Zhang, Xubin Ren, Zirui Guo, Tianzhe Chu, Yi Ma, Chao Huang 19d ago

DeepTutor: Towards Agentic Personalized Tutoring

Open-source agentic tutoring framework combining LLMs with personalized feedback, difficulty calibration, and citation-grounded problem solving.

Ax Liang Luo, Yinbin Ma, Quanyu Zhu, Vasiliy Kuznetsov, Yuxin Chen, Neng Shi, Jian Jiao, Jiecao Yu, Buyun Zhang, Tongyi Tang, Xiaohan Wei, Yanli Zhao, Zeliang Chen, Yuchen Hao, Venkatesh Ranganathan, Sandeep Parab, Yantao Yao, Maxim Naumov, Chunzhi Yang, Shen Li, Ellie Wen, Wenlin Chen, Santanu Kolay, Chunqiang Tang 19d ago

LoKA: Low-precision Kernel Applications for Recommendation Models At Scale

Low-precision FP8 arithmetic optimization techniques for large recommendation models, addressing numerical sensitivity and training efficiency.

Ax Xuefei Wang, Jialu Wang, Fengbo Zhang, Yihan Hu, Di Zhang, Yutong Ye, Yikun Ban, Jun Han, Ruijie Wang 19d ago

MasFACT: Continual Multi-Agent Topology Learning via Geometry-Aware Posterior Transfer

Continual learning method for multi-agent LLM systems to maintain and adapt inter-agent communication topologies across evolving task streams.

Ax Zhenyu Li, Aleksandar Cvejic, Zehui Chen, Peter Wonka 19d ago

CriterAlign: Criterion-Centric Rationale Alignment for Code Preference Judging

Criterion-centric approach for pairwise preference prediction in code generation using LLM judges with explicit rubric alignment.

Ax Han Zhang, Wanting Jiang, Tomasz Kornuta, Tian Zheng, Vidya Murali 19d ago

MAVEN: A Multi-stage Agentic Annotation Pipeline for Video Reasoning Tasks

Multi-stage agentic pipeline for generating structured video annotations and Chain-of-Thought reasoning traces for training vision-language models.

Ax Quanjiang Li, Zhiming Liu, Wei Luo, Tingjin Luo, Chenping Hou 19d ago

Correcting Visual Blur Induced by Attention Distraction to Reduce Hallucinations: Algorithm and Theory

Analysis of hallucinations in multimodal LLMs as attention distraction phenomenon, with proposed correction mechanism.

Ax Zelalem Abahana, David Evans, Satish Mahadevan Srinivasan, Matjaz Gams 19d ago

When RLHF Fails: A Mechanistic Taxonomy of Reward Hacking, Collapse, and Evaluator Gaming

Mechanistic study of RLHF failures including reward hacking and policy collapse, with diagnostic tools and multiple evaluation methods.

Ax Zhilong Song, Zongmin Zhang, Lixue Cheng 19d ago

Autonomous heterogeneous catalyst discovery with a self-evolving multi-agent digital twin

Multi-agent digital twin system for autonomous heterogeneous catalyst discovery using machine learning and gas-solid/liquid-solid modeling.

Ax Ian Rios-Sialer, Shantanu Darveshi, Shuai Jiang, Avigya Paudel, Anastasiia Pronina, Ipshita Bandyopadhyay, Justin Shenk 19d ago

Temporal Preference Concepts and their Functions in a Large Language Model

arXiv paper identifying and analyzing temporal preference subgraphs in LLMs through causal localization, showing how models represent temporal tradeoffs internally.

Ax Qiwei Zeng, Hao Wang, Jinghao Lin, Shuchang Ye, Yuezhe Yang, Yige Peng, Haoyuan Che, Jinman Kim, Lei Bi 19d ago

EasyLens: A Training-Free Plug-and-Play Subtle-Lesion Representation Amplifier for Medical Vision-Language Models

arXiv paper on EasyLens, a training-free method to amplify subtle-lesion representations in medical vision-language models for improved clinical detection.

Ax Dohwan Kim, Jung-Woo Choi 19d ago

MeCo: One-Step MeanFlow-based Corrector for Multi-Channel Speech Separation

arXiv paper proposing MeCo, a one-step generative corrector using MeanFlow for multi-channel speech separation with improved perceptual quality.

Ax Tingyu Li, Le Zhou, Siyuan Li, Yujun Wu, Xinglong Xu, Jingxuan Wei, Conghui He, Cheng Tan 19d ago

Bridging Modal Isolation in Interleaved Thinking: Supervising Modality Transitions via Stepwise Reinforcement

Addresses modal isolation in multimodal interleaved thinking models where text and vision diverge, proposing stepwise reinforcement for modality supervision.

Ax Abhinav Agarwal, Adam Wei, Taylan Kargin, Michael Zeng, Cole Becker, Arif Kerem Dayi, Pablo Parrilo, Asuman Ozdaglar, Russ Tedrake 19d ago

Training and Evaluating Diffusion Policies with Long Context Lengths

Benchmark of diffusion policies for robotic manipulation with incrementally increasing context lengths to enable memory and long-horizon task performance.

Ax Jannik H\"osch, Alessandro Sestini, Florian Fuchs, Amir Baghi, Joakim Bergdahl, Iolanda Leite, Konrad Tollmar, Jean-Philippe Barrette-LaPierre, Linus Gissl\'en 19d ago

Hierarchical Control in Multi-Agent Games: LLM-based Planning and RL Execution

Hierarchical multi-agent architecture combining LLM-based strategic planning with specialized RL skill policies for complex coordinated decision-making.

Ax Alfarizy Alfarizy, Hung Truong Thanh Nguyen, Ren\'e Richard, Roozbeh Razavi-Far, Hung Cao 19d ago

Does Mixture-of-Experts Actually Help Inference on Consumer and Edge Hardware? An Empirical Study

Empirical study of Mixture-of-Experts language models on consumer/edge hardware, questioning whether FLOP advantages translate to actual speedups and cost reductions.

Ax Yixuan Li, Guangzhi Sun, Yudong Yang, Chao Zhang 19d ago

video-SALMONN-R$^3$: Learning to ReWatch, ReAsk, and ReAnswer for Efficient Video Understanding

Video-SALMONN-R³ implements two-stage video understanding with LLMs: coarse localization then high-fidelity re-watching for efficient question answering.

Ax Tyler Ga Wei Lum, Kushal Kedia, C. Karen Liu, Jeannette Bohg 19d ago

Play2Perfect: What Matters in Dexterous Play Pretraining for Precise Assembly?

Play2Perfect investigates pretraining strategies for dexterous multi-finger robotic assembly tasks combining imitation learning and reinforcement learning.

Ax Yanis Labrak, Dairazalia Sanchez-Cortes, Sergio Burdisso, S\'everin Baroudi, Shashi Kumar, Esa\'u Villatoro-Tello, Srikanth Madikeri, Manjunath K E, Old\v{r}ich Plchot, Kadri Hacio\u{g}lu, Petr Motlicek, Andreas Stolcke 19d ago

How to Leverage Synthetic Speech for LLM-Based ASR Systems?

Study on leveraging synthetic TTS speech for training ASR systems in privacy-constrained domains like banking and healthcare, addressing synthetic-real data gaps.

Ax R\'ois\'in Luo, Christian Gagn\'e, Jonas Ngnaw\'e, Ihsan Ullah, Karyn Morrissey 19d ago

A Stochastic--Geometric Theory of Scaling Laws in Grokking

Theoretical characterization of grokking phenomenon using stochastic-geometric analysis of solution space topology and delayed generalization in neural networks.

Ax Zhuoxuan Zhang (Yang), Kangqi Ni (Yang), Yuhang Chen (Yang), Mingfu Liang (Yang), Xiaohan Wei (Yang), Yunchen Pu (Yang), Fei Tian (Yang), Chonglin Sun (Yang), Frank Shyu (Yang), Adam (Yang), Song, Sandeep Pandey, Luke Simon, Tianlong Chen, Xi Liu 19d ago

Diffusion-GR2: Diffusion Generative Reasoning Re-ranker

Diffusion-GR2 uses block-diffusion language models for faster generative reasoning re-ranking with parallel decoding instead of sequential autoregressive inference.

Ax Yaniv Melamed, Yoni Zukerman, Michal Shechter, Miri Weissler, Ashwin Patil, Hani Neuvirth-Telem 19d ago

TAG: A Lightweight Framework for Test-Driven Agentic Artifact Generation

TAG framework for reliable artifact generation with LLMs using test-driven validation, shifting focus from generation quality to validation rigor.

Ax Vaishnavi Sinha, Pooja Guttal, Pranay Deep Reddy Katike, Vishal Sinha, Gerald Ndawula, Lira Yoon, Andrea Kleinsmith, Manas Gaur 19d ago

Where do LLMs Fall Short in CBT-Guided Affective Reasoning?

Study analyzing LLM failures in applying Cognitive Behavioral Therapy frameworks despite high theoretical knowledge, highlighting reasoning gaps in practical application.

Ax Alexis Kafantaris 19d ago

LLM for the development of FCM

Approach using local LLMs to extract quantitative data from text for developing fuzzy cognitive maps.

Ax Haozhe Wang, Weijia Feng, Jinpeng Yu, Che Liu, Ping Nie, Fangzhen Lin, Jiaming Liu, Ruihua Huang, Jimmy Lin, Wenhu Chen, Cong Wei 19d ago

Search Beyond What Can Be Taught: Evolving the Knowledge Boundary in Agentic Visual Generation

Framework for agentic visual generation that extends knowledge boundaries by searching for out-of-distribution content beyond training data.

Ax Archie Chaudhury 19d ago

ResonatorLM: Causal Resonant Field Mixing for Efficient Long-Context Language Modeling

Alternative to transformer architecture using causal resonant field mixing for efficient long-context language modeling.

Ax Wentao Lu 19d ago

ReCoLoRA: Spectrum-Aware Recursive Consolidation for Continual LLM Fine-Tuning

Parameter-efficient continual fine-tuning framework for LLMs using spectrum-aware recursive consolidation of LoRA adapters.

Ax Haozhan Tang, Zerui Wang, Yuxian Gu, Song Han, Han Cai 19d ago

Jet-Long: Efficient Long-Context Extension with Dynamic Bifocal RoPE

Method for extending LLM context length beyond pretraining windows using dynamic bifocal RoPE for long-context applications and agentic workflows.

Ax Didula Samaraweera, Anjana Supun, Srinath Perera 19d ago

Selective Left-Shift: Turning Test-Time Compute and Difficulty-based Curation into Training Data for Low-Resource Code Generation

Method for improving code generation in low-resource programming languages using test-time compute and difficulty-based data curation for fine-tuning.

Ax Syed Usama Imtiaz, Mitra Nasr Azadani, Nasrin Alamdari 19d ago