Isolater - Feed

Ax Gregory N. Frank 9d ago

How Alignment Routes: Localizing, Scaling, and Controlling Policy Circuits in Language Models

Analysis of policy routing circuits in alignment-trained LLMs, localizing attention gates and amplifier heads controlling refusal behavior.

Ax Zheng Li, Jerry Cheng, Huanying Helen Gu 9d ago

StableTTA: Training-Free Test-Time Adaptation that Improves Model Accuracy on ImageNet1K to 96%

StableTTA: training-free test-time adaptation improving image classification accuracy via novel ensemble strategies.

Ax Shuzhen Bi, Mingzi Zhang, Zhuoxuan Li, Xiaolong Wang, Keqian Li, Aimin Zhou 9d ago

EduIllustrate: Towards Scalable Automated Generation Of Multimodal Educational Content

EduIllustrate: benchmark evaluating LLMs on automated generation of diagram-rich educational content combining visuals and reasoning.

Ax Honghao Fu, Miao Xu, Yiwei Wang, Dailing Zhang, Liu Jun, Yujun Cai 9d ago

VideoStir: Understanding Long Videos via Spatio-Temporally Structured and Intent-Aware RAG

VideoStir: retrieval-augmented generation system for understanding long videos with multimodal LLMs using spatio-temporal structure.

Ax Manish Bhatt, Sarthak Munshi, Vineeth Sai Narajala, Idan Habler, Ammar Al-Kahfah, Ken Huang, Joel Webb, Blake Gatto, Md Tamjidul Hoque 9d ago

The Defense Trilemma: Why Prompt Injection Defense Wrappers Fail?

Theoretical analysis proving limitations of continuous wrapper defenses against prompt injection attacks in LLMs.

Ax Zhixiong Zhao, Zukang Xu, Zhixuan Chen, Dawei Yang 9d ago

MoBiE: Efficient Inference of Mixture of Binary Experts under Post-Training Quantization

MoBiE: binarization framework for efficient inference in Mixture-of-Experts LLMs via post-training quantization.

Ax Benjamin J. Choi, Melanie Weber 9d ago

Latent Structure of Affective Representations in Large Language Models

Analysis of emotional representation geometry in LLM latent spaces for transparency and safety.

Ax Xiangru Jian, Hao Xu, Wei Pang, Xinjian Zhao, Chengyu Tao, Qixin Zhang, Xikun Zhang, Chao Zhang, Guanzhi Deng, Alex Xue, Juan Du, Tianshu Yu, Garth Tarr, Linqi Song, Qiuzhuang Sun, Dacheng Tao 9d ago

FORGE: Fine-grained Multimodal Evaluation for Manufacturing Scenarios

Fine-grained benchmark evaluating multimodal LLMs on manufacturing scenarios.

Ax Qian Ma, Sarah Rajtmajer 9d ago

Private Seeds, Public LLMs: Realistic and Privacy-Preserving Synthetic Data Generation

Privacy-preserving synthetic data generation using LLMs with differential privacy mechanisms.

Ax Ziyi Chen, Yasir Khan, Mengyuan Zhang, Cheng Peng, Mengxian Lyu, Yiyang Liu, Krishna Vaddiparti, Robert L Cook, Mattia Prosperi, Yonghui Wu 9d ago

Detecting HIV-Related Stigma in Clinical Narratives Using Large Language Models

LLM-based tool for identifying HIV-related stigma in clinical narratives.

Ax Bo Li, Shikun Zhang, Wei Ye 9d ago

Data Selection for Multi-turn Dialogue Instruction Tuning

Data selection framework for multi-turn dialogue instruction tuning addressing dataset noise and inconsistency.

Ax Benjamin L\'eger, Kazem Meidani, Christian Gagn\'e 9d ago

Multi-Modal Learning meets Genetic Programming: Analyzing Alignment in Latent Space Optimization

Genetic programming combined with latent space optimization for symbolic regression via neural encoders.

Ax Arth Singh 9d ago

Re-Mask and Redirect: Exploiting Denoising Irreversibility in Diffusion Language Models

Attack exploiting denoising irreversibility in diffusion language models to bypass safety alignment.

Ax Xinmiao Xiong, Bangya Liu, Hao Wang, Dayou Li, Nuo Chen, Andrew Feng, Mingyu Ding, Suman Banerjee, Yang Zhou, Zhiwen Fan 9d ago

Accelerating Transformer-Based Monocular SLAM via Geometric Utility Scoring

Optimization for geometric foundation models in monocular SLAM via efficient keyframe selection.

Ax Keyu Li, Jin Gao, Dequan Wang 9d ago

Aligned Agents, Biased Swarm: Measuring Bias Amplification in Multi-Agent Systems

Empirical study measuring bias amplification in multi-agent systems topologies and feedback loops.

Ax Changi Hong, Yoonah Song, Hwayoung Park, Chaewoon Bang, Dayeon Gu, Do Hyun Lee, Hong Kook Kim 9d ago

PS-TTS: Phonetic Synchronization in Text-to-Speech for Achieving Natural Automated Dubbing

Text-to-speech synchronization method for automated dubbing with phonetic alignment.

Ax Peng Wang (The Chinese University of Hong Kong, Shenzhen), Yanqiao Zhu (X-LANCE Lab, Shanghai Jiao Tong University), Zixuan Jiang (Xi'an Jiaotong University), Qinyuan Chen (Fudan University), Xingjian Zhao (Fudan University), Xipeng Qiu (Fudan University), Wupeng Wang (Tongyi Fun Team, Alibaba Group), Zhifu Gao (Tongyi Fun Team, Alibaba Group), Xiangang Li (Tongyi Fun Team, Alibaba Group), Kai Yu (X-LANCE Lab, Shanghai Jiao Tong University), Xie Chen (X-LANCE Lab, Shanghai Jiao Tong University) 9d ago

Interactive ASR: Towards Human-Like Interaction and Semantic Coherence Evaluation for Agentic Speech Recognition

Interactive ASR system with semantic coherence evaluation and human-like correction mechanisms.

Ax Yuning Wang, Pol Suarez, Mathis Bode, Ricardo Vinuesa 9d ago

Physics-guided surrogate learning enables zero-shot control of turbulent wings

Physics-guided surrogate learning for turbulent flow control without heavy RL computational costs.

Ax Jingyu Zhang, Tianjian Li, William Jurayj, Hongyuan Zhan, Benjamin Van Durme, Daniel Khashabi 9d ago

Many-Tier Instruction Hierarchy in LLM Agents

Framework for managing hierarchical instruction conflicts in multi-source LLM agent environments.

Ax Julio Candanedo 9d ago

The Diffusion-Attention Connection

Theoretical connection between Transformers, diffusion maps, and magnetic Laplacians through Markov geometry.

Ax James K. Ruffle, Samia Mohinta, Chris Foulon, Mohamad Zeina, Zicheng Wang, Sebastian Brandner, Harpreet Hyare, Parashkev Nachev 9d ago

Fairboard: a quantitative framework for equity assessment of healthcare models

Framework for evaluating fairness and equity across patient subgroups in brain tumor segmentation models.

Ax Pankayaraj Pathmanathan, Furong Huang 9d ago

Deliberative Alignment is Deep, but Uncertainty Remains: Inference time safety improvement in reasoning via attribution of unsafe behavior to base model

Study of deliberative alignment for deeper safety in reasoning LLMs via attribution analysis.

Ax Hua-Dong Xiong (School of Psychological and Brain Sciences, Georgia Tech), Li Ji-An (Department of Psychology, New York University), Jiaqi Huang (Department of Cognitive Science, Indiana University Bloomington, Honda Research Institute), Robert C. Wilson (School of Psychological and Brain Sciences, Georgia Tech, Center of Excellence for Computational Cognition, Georgia Tech), Kwonjoon Lee (Honda Research Institute), Xue-Xin Wei (Departments of Neuroscience and Psychology, The University of Texas at Austin) 9d ago

Human-like Working Memory Interference in Large Language Models

Analysis of working memory limitations in LLMs and comparison with biological systems.

Ax Liu Xiao 9d ago

Belief-State RWKV for Reinforcement Learning under Partial Observability

RWKV-based RL approach with explicit belief state representation for partial observability problems.

Ax Dongmin Kim, Hoshinori Kanazawa, Yasuo Kuniyoshi 9d ago

Active Inference with a Self-Prior in the Mirror-Mark Task

Computational model using Transformer self-prior to simulate mirror self-recognition behavior without external rewards.

Ax Ming Lei, Christophe Baehr 9d ago

A Comparative Theoretical Analysis of Entropy Control Methods in Reinforcement Learning

Theoretical analysis comparing entropy regularization and covariance-based mechanisms for controlling policy collapse in RL-enhanced LLMs.

Ax Samah Fodeh, Ganesh Puthiaraju, Elyas Irankhah, Linhai Ma, Srivani Talakokkul, Afshan Khan, Sreeraj Ramachandran, Jordan Alpert, Sarah Schellhorn 9d ago

STaR-DRO: Stateful Tsallis Reweighting for Group-Robust Structured Prediction

Framework for robust structured prediction using Tsallis reweighting and task-agnostic prompting with XML structure for group-robust fine-tuning.

Ax Vijay Lingam, Aditya Golatkar, Anwesan Pal, Ben Vo, Narayanan Sadagopan, Alessandro Achille, Jun Huan, Anoop Deoras, Stefano Soatto 9d ago

ExecTune: Effective Steering of Black-Box LLMs with Guide Models

Guide-Core Policies framework for black-box LLM agents where guide models generate structured strategies executed by core models reducing inference costs.

Ax Mainak Kundu, Catherine Chen, Rifatul Islam, Ismail Uysal, Ria Kanjilal 9d ago

Explainable Human Activity Recognition: A Unified Review of Concepts and Mechanisms

Review of explainable AI mechanisms for human activity recognition in healthcare, assistive living, and smart environments applications.

Ax Weijian Mai, Mu Nan, Yu Zhu, Jiahang Cao, Rui Zhang, Yuqin Dai, Chunfeng Song, Andrew F. Luo, Jiamin Wu 9d ago

NeuroFlow: Toward Unified Visual Encoding and Decoding from Neural Activity

Unified visual encoding and decoding framework from neural activity modeling consistency between brain stimulus prediction and reconstruction.

Ax Robin Young, Michael E. Van Nuland, E. Toby Kiers, Tom\'a\v{s} V\v{e}trovsk\'y, Petr Kohout, Petr Baldrian, Srinivasan Keshav 9d ago

Below-ground Fungal Biodiversity Can be Monitored Using Self-Supervised Learning Satellite Features

Self-supervised learning on satellite imagery predicting mycorrhizal fungal biodiversity at landscape scales for ecosystem monitoring.

Ax Jan Kirin 9d ago

Relational Preference Encoding in Looped Transformer Internal States

Analysis of preference encoding in looped transformer internal states using lightweight evaluator heads on RLHF dataset.

Ax Yi-Hao Peng, Samarth Das, Jeffrey P. Bigham, Jason Wu 9d ago

Efficient Personalization of Generative User Interfaces

Methods for personalizing generative user interfaces addressing subjective preferences through preference divergence analysis and sparse feedback.

Ax Halil Ibrahim Gulluk, Olivier Gevaert 9d ago

SemEnrich: Self-Supervised Semantic Enrichment of Radiology Reports for Vision-Language Learning

Self-supervised semantic enrichment method for medical vision-language datasets addressing reporting bias in radiology reports.

Ax Tyler Yang, Romal Mitr 9d ago

Improving Pediatric Emergency Department Triage with Modality Dropout in Late Fusion Multimodal EHR Models

Multimodal EHR model for pediatric emergency triage using modality dropout to improve generalizability across demographics.

Ax Micha{\l} Derezi\'nski, Xiaoyu Dong 9d ago

Last-Iterate Convergence of Randomized Kaczmarz and SGD with Greedy Step Size

Convergence analysis of randomized Kaczmarz and SGD with greedy step size proving O(1/t^3/4) last-iterate convergence rate.

Ax Joseph Liu, Nameer Hirschkind, Xiao Yu, Mahesh Kumar Nandwana 9d ago

Regularized Entropy Information Adaptation with Temporal-Awareness Networks for Simultaneous Speech Translation

Temporal-aware network improving simultaneous speech translation policy to balance quality and latency with temporal context awareness.

Ax Theo X. Olausson, Metod Jazbec, Xi Wang, Armando Solar-Lezama, Christian A. Naesseth, Stephan Mandt, Eric Nalisnick 9d ago

A Tale of Two Temperatures: Simple, Efficient, and Diverse Sampling from Diffusion Language Models

Sampling methods for diffusion language models balancing speed, quality, and diversity through tempered confidence-based remasking.

Ax Maryam Ahang, Todd Charter, Masoud Jalayer, Homayoun Najjaran 9d ago

A Hybrid Intelligent Framework for Uncertainty-Aware Condition Monitoring of Industrial Systems

Hybrid condition monitoring framework combining data-driven learning with physics-based insights for industrial systems reliability.

Ax Smita Deb, Shirin Panahi, Mulugeta Haile, Ying-Cheng Lai 9d ago

Vestibular reservoir computing

Physical reservoir computing inspired by biological vestibular system addressing hardware complexity with designed uncoupled topology.

Ax Renjini R. Nair (Microsoft), Damian K. Kowalczyk (Microsoft), Marco Gaudesi (Microsoft), Chhaya Methani (Microsoft) 9d ago

SLM Finetuning for Natural Language to Domain Specific Code Generation in Production

Fine-tuning small language models for domain-specific code generation in production environments with strict latency requirements.

Ax James Nguyen 9d ago

From Recency Bias to Stable Convergence Block Kaczmarz Methods for Online Preference Learning in Matchmaking Applications

Kaczmarz-based preference learning algorithms for real-time matchmaking with stable convergence replacing recency-biased normalization.

Ax Ziyue Liu, Ruijie Zhang, Zhengyang Wang, Yequan Zhao, Yupeng Su, Zi Yang, Zheng Zhang 9d ago