Isolater - Feed

Ax Yurui Zheng, Ying Jin 29d ago

Prediction Sets for Counterfactual Decisions: Coverage, Optimality, and Conformal Prediction

Studies uncertainty quantification via conformal prediction for counterfactual decision-making in high-stakes applications.

Ax Zhanming Shen, Jintao Tong, Shaotian Yan, Chen Shen, Hao Chen, Wentao Ye, Xiaomeng Hu, Rui Miao, Haobo Wang, Junbo Zhao, Gang Chen, Jieping Ye 29d ago

Purified OPSD: On-Policy Self-Distillation Without Losing How to Think

Addresses failure of on-policy self-distillation on long chain-of-thought reasoning, proposing method to maintain model thinking capability.

Ax Mikael M{\o}ller H{\o}gsgaard, Patrick Rebeschini, Tobias Wegel 29d ago

Aggregation with Exponential Weights is Optimal in Expectation

Proves aggregation with exponential weights is minimax-rate optimal in expectation for model selection, settling open problem from 2013.

Ax Juwei Shen, Yujie Wu, Changwen Chen 29d ago

Dendritic In-Context Learning in a Single-Layer Spiking Neural Network

Enables in-context learning in spiking neural networks via dendritic computation, making biologically plausible SNNs pass Garg-2022 ICL benchmark.

Ax Zichao Wei 29d ago

On the Role of Directionality in Structural Generalization

Redesigns symbolic parser backend using CCG directed types for improved structural generalization on SLOG benchmark with 30K parameters.

Ax Minghao Li, Raghav Mittal, Sanjivni Rana, Suraj Shetiya, Gautam Das, Nick Koudas 29d ago

HNSW with Accuracy Guarantees Using Graph Spanners -- A Technical Report

HNSW search framework adding theoretical correctness guarantees to hierarchical navigable small world graphs via graph spanner verification.

Ax Kuo-Chung Peng, Jiun-Cheng Jiang, Chun-Hua Lin, Yifeng Peng, Junghoon Justin Park, Huan-Hsin Tseng, Hsin-Yi Lin, Kuan-Cheng Chen, Chen-Yu Liu, Shinjae Yoo, Samuel Yen-Chi Chen 29d ago

Stable Self-Modulating Quantum Fast-Weight Programmers with Bounded Memory Gates

Proposes quantum sequence modeling using variational circuits with self-modulating gates and bounded memory for stable long-sequence processing.

Ax Yuan Yuan 29d ago

The Dual Nature of LLM Persona: Aggregated Tendencies and Frame-Dependent Geometry

Studies whether LLM personas from psychometric questionnaires are intrinsic or frame-dependent using geometric analysis on manifolds.

Ax Minghan Yu, Youran Sun, Chugang Yi, Yixin Wen, Haizhao Yang 29d ago

Bringing Agentic Search to Earth Observation Data Discovery

NASA deploys agentic search system using LLMs to help geoscience researchers discover relevant datasets and tools from thousands of available resources.

Ax Mauricio Fadel Argerich, Jonathan F\"urst, Marta Pati\~no-Mart\'inez 29d ago

WattGPU: Predicting Inference Power and Latency on Unseen GPUs and LLMs

WattGPU predicts power consumption and latency for LLM inference across unseen GPUs without exhaustive profiling, addressing data center energy optimization.

Ax Thomas Winninger 29d ago

Fast Multi-dimensional Refusal Subspaces via RFM-AGOP

arXiv paper on fast multi-dimensional refusal subspace extraction in LLMs for safety and interpretability.

Ax Jakob Geusen, Ender Konukoglu 29d ago

Object-centric LeJEPA

arXiv paper on object-centric LeJEPA for more data-efficient self-supervised image representation learning.

Ax M. Doris, S. Guo, S. M. Koh, L. Ritter, A. R. Fritsch, S. Mukherjee, I. B. Spielman, J. P. Zwolak 29d ago

Q-GAIN: A Python Package for Machine Learning and Physically Informed Analysis Applications

Q-GAIN Python package for machine learning and physics-informed analysis of cold-atom experiment images with classification and detection.

Ax Donghyun Lee, Jitesh Chavan, Duy Nguyen, Sam Huang, Liming Jiang, Priyadarshini Panda, Timo Mertens, Saurabh Shukla 29d ago

OrbitQuant: Data-Agnostic Quantization for Image and Video Diffusion Transformers

OrbitQuant data-agnostic quantization method for diffusion transformers handling activation shifts across timesteps without recalibration.

Ax Juanwu Lu, Junyu Zhu, Ziran Wang 29d ago

Controllable Sim Agents with Behavior Latents

CNeVA framework for controllable simulated traffic agents with interpretable behavior latents enabling edge case testing and variable isolation.

Ax Arman Ghaffarizadeh, Danyal Mohaddes, Aliakbar Izadkhah, Shahriar Noroozizadeh 29d ago

What LLM Agents Say When No One Is Watching: Social Structure and Latent Objective Emergence in Multi-Agent Debates

Study of how social structure and audience context affect what LLM agents express in multi-agent debate settings using dual-channel framework.

Ax Mona Schirmer, Metod Jazbec, Alexander Timans, Christian Naesseth, Maja Waldron, Eric Nalisnick 29d ago

Online Safety Monitoring for LLMs

Real-time safety monitoring framework for LLMs using external verifiers with risk-calibrated thresholds to detect unsafe outputs at deployment.

Ax Matteo Boglioni, Thibault Rousset, Siva Reddy, Marius Mosbach, Verna Dankers 29d ago

LACUNA: A Testbed for Evaluating Localization Precision for LLM Unlearning

LACUNA testbed evaluates parameter-level localization precision for LLM unlearning, addressing memorized sensitive training data removal.

Ax Javier Lopez-Piqueres, Pranav Deshpande, Archan Ray, Mattia J. Villani, Marco Pistoia, Niraj Kumar 29d ago

MetaTT: A Global Tensor-Train Adapter for Parameter-Efficient Fine-Tuning

MetaTT tensor-train adapter for parameter-efficient fine-tuning of transformers with flexible factorization across layers and task dimensions.

Ax David Gonz\'alez-Mart\'inez 29d ago

BALF: Budgeted Activation-Aware Low-Rank Factorization for Fine-Tuning-Free Model Compression

BALF framework for parameter-efficient model compression using activation-aware low-rank factorization beyond linear layers.

Ax Shivam Singhal, Priyadarsi Mishra, Eran Malach, Tomer Galanti 29d ago

LLM Priors for ERM over Programs

Method using LLM priors to enable efficient program learning through empirical risk minimization with fewer samples and less computation.

Ax Ronald Katende 29d ago

Geometry as a Missing Axis of Representation Quality: The Variational Geometric Information Bottleneck under Data Scarcity

Framework incorporating latent geometry as explicit representation quality component under data scarcity through variational information bottleneck.

Ax Yulong Lu, Tong Mao, Jinchao Xu, Yahong Yang 29d ago

On the Dimension-Free Approximation of Deep Neural Networks for Symmetric Korobov Functions

Theoretical analysis of deep neural network approximation rates for symmetric Korobov functions with polynomial dimension dependence.

Ax Naveen George, Naoki Murata, Yuhta Takida, Konda Reddy Mopuri, Yuki Mitsufuji 29d ago

Locality-Aware Continual Unlearning for Diffusion Models

Method for continual unlearning in diffusion models to progressively remove concepts while maintaining generation quality across multiple removal steps.

Ax Long Lian, Sida Wang, Felix Juefei-Xu, Tsu-Jui Fu, Xiuyu Li, Adam Yala, Trevor Darrell, Alane Suhr, Yuandong Tian, Xi Victoria Lin 29d ago

ThreadWeaver: Adaptive Threading for Efficient Parallel Reasoning in Language Models

ThreadWeaver enables parallel reasoning in LLMs through adaptive threading to reduce inference latency while maintaining output quality.

Ax Dhrubo Saha 29d ago

ZENITH: Automated Gradient Norm Informed Stochastic Optimization

ZENITH optimizer for automatic learning rate scheduling in deep vision models with lower computational overhead than existing adaptive optimizers.

Ax Lukas Sch\"afer, Pallavi Choudhury, Abdelhak Lemkhenter, Chris Lovett, Somjit Nath, Luis Fran\c{c}a, Matheus Ribeiro Furtado de Mendon\c{c}a, Alex Lamb, Riashat Islam, Siddhartha Sen, John Langford, Katja Hofmann, Sergio Valcarcel Macua 29d ago

When Does Predictive Inverse Dynamics Outperform Behavior Cloning?

Theoretical analysis comparing predictive inverse dynamics models to behavior cloning for offline imitation learning with limited demonstrations.

Ax Hao Gu, Mao-Lin Luo, Zi-Hao Zhou, Han-Chen Zhang, Min-Ling Zhang, Tong Wei 29d ago

Spectral Imbalance Causes Forgetting in Low-Rank Continual Adaptation

Research on spectral imbalance in low-rank continual learning for parameter-efficient model adaptation without catastrophic forgetting.

Ax Kanghyun Noh, Jinheon Choi, Yulhwa Kim 29d ago

QTALE: Quantization-Robust Token-Adaptive Layer Execution for LLMs

Efficient LLM deployment technique combining token-adaptive layer execution with quantization for reduced computation and memory.

Ax Yuxin Ma, Nan Chen, Mateo D\'iaz, Soufiane Hayou, Dmitriy Kunisky, Soledad Villar 29d ago

$\mu$pscaling small models: Principled warm starts and hyperparameter transfer

Principled approach for upscaling smaller trained models to larger ones with hyperparameter transfer and warm starts.

Ax Chenxiao Yang, Nathan Srebro, Zhiyuan Li 29d ago

Recursive Models for Long-Horizon Reasoning

Framework enabling language models to overcome context limitations by recursively invoking themselves to solve long-horizon reasoning problems.

Ax Nils Gr\"unefeld, Jes Frellsen, Christian Hardmeier 29d ago

An Isotropic Approach to Efficient Uncertainty Quantification with Gradient Norms

Lightweight uncertainty quantification method for neural networks using gradient norms and isotropy assumptions.

Ax Abbas Zeitoun, Lucas Torroba-Hennigen, Yoon Kim 29d ago

Hyperloop Transformers

Parameter-efficient LLM architecture using looped transformers to improve memory efficiency for edge and on-device deployment.

Ax Changyu Li, Shuanghong Huang, Jiashen Liu, Ming Lei, Jidu Xing, Kaishun Wu, Lu Wang, Fei Luo 29d ago

FED-FSTQ: Fisher-Guided Token Quantization for Communication-Efficient Federated Fine-Tuning of LLMs on Edge Devices

Federated fine-tuning framework using Fisher-guided token quantization to reduce communication for LLM adaptation on edge devices.

Ax Peter Racioppo 29d ago

The Transformer as a Polar State Estimator

Geometric interpretation of transformer components showing attention and normalization emerge from polar state estimation.

Ax Younghun Go, Jaehoon Han, Changyong Shin, Chuck Yoo, Gyeongsik Yang 29d ago

Enabling KV Caching of Shared Prefix for Diffusion Language Models

Technique for KV caching shared prefixes in diffusion language models with bidirectional attention mechanisms.

Ax Wanghan Xu, Shuo Li, Tianlin Ye, Qinglong Cao, Yixin Chen, Hengjian Gao, Yiheng Wang, Qi Li, Kun Li, Sheng Xu, Shengdu Chai, Fangchen Yu, Xiangyu Zhao, Zhangrui Zhao, Weijie Ma, Zijie Guo, Koutian Wu, Haoyu Zhou, Haoxiang Yin, Lixue Cheng, Chaofan Hu, Haoxuan Li, Lu Mi, Xuxuan Xie, Yifan Zhou, Ruizhe Chen, Zhiwang Zhou, Xingjian Guo, Yuhao Zhou, Xuming He, Shengyuan Xu, Xinyu Gu, Jiamin Wu, Mianxin Liu, Chunfeng Song, Fenghua Ling, Dongzhan Zhou, Shixiang Tang, Yuqiang Li, Mao Su, Peng Ye, Siqi Sun, Bin Wang, Xue Yang, Zhenfei Yin, Tianfan Fu, Guangtao Zhai, Wanli Ouyang, Bo Zhang, Lei Bai, Wenlong Zhang 29d ago

ResearchClawBench: A Benchmark for End-to-End Autonomous Scientific Research

Benchmark with 40 tasks across 10 scientific domains for evaluating end-to-end autonomous research capabilities of AI coding agents.

Ax Hongbo Wang 29d ago

When Do Conservation Laws Survive Learned Representations? Certified Horizons for Latent World Models

Framework for certifying when conservation laws remain valid in learned latent representations of physical systems.

Ax Jingwei Song, Haofeng Xu, Jie Xiao, Chengke Bao, Jingwei Shi, Pengbin Feng, Weixun Wang, Yuhang Han, Chuan Wu, Linfeng Zhang, Bill Shi 29d ago

Staleness-Learning Rate Scaling Laws for Asynchronous RLHF

Analysis of learning rate scaling laws for asynchronous RLHF with stale rollouts in high-throughput LLM training.

Ax Zijian Zhang, Rizhen Hu, Athanasios Glentis, Dawei Li, Chung-Yiu Yau, Hongzhou Lin, Mingyi Hong 29d ago

Is One Layer Enough? Training A Single Transformer Layer Can Match Full-Parameter RL Training

Study showing single transformer layer RL training matches full-parameter fine-tuning for LLM post-training with GRPO.

Ax Tong Xiao, Jingbo Zhu 29d ago

Introduction to Transformers: an NLP Perspective

Introduction to Transformer architecture, key refinements, and applications in natural language processing.

Ax Alexander Ororbia, Karl Friston, Rajesh P. N. Rao 29d ago

Meta-Representational Predictive Coding: Neuroscience-Informed Self-Supervised Learning

Self-supervised learning approach inspired by neuroscience using predictive coding with biologically plausible credit assignment.

Ax Wenji Fang, Jing Wang, Yao Lu, Shang Liu, Yuchao Wu, Yuzhe Ma, Zhiyao Xie 29d ago

A Survey of Circuit Foundation Model: Foundation AI Models for VLSI Circuit Design and EDA

Survey of foundation models for VLSI circuit design and EDA using self-supervised pre-training on circuit data.

Ax Abdullah Burkan Bereketoglu 29d ago

Composite Reward Design in PPO-Driven Adaptive Filtering

PPO-driven adaptive filtering with composite reward design for denoising in dynamic, non-stationary environments like wireless signals and biomedical monitoring.

Ax Salahuddin Salahuddin, Ahmed Hussain, Jussi L\"opp\"onen, Toni Jutila 29d ago

Less Data, More Security: Advancing Cybersecurity LLMs Specialization via Resource-Efficient Domain-Adaptive Continuous Pre-training with Minimal Tokens

Domain-adaptive continuous pre-training specializes LLMs for cybersecurity analysis with minimal tokens and HPC efficiency for reduced computational requirements.

Ax Milan Marocchi, Matthew Fynn, Kayapanda Mandana, Yue Rong 29d ago