Isolater - Feed

Ax Ahmed Heakl, Gustavo Bertolo Stahl, Sarim Hashmi, Seung Hun Eddie Han, Mukul Ranjan, Arina Kharlamova, Salman Khan, Abdulrahman Mahmoud 4/22/2026

CASS: Nvidia to AMD Transpilation with Data, Models, and Benchmark

Dataset and ML models for transpiling GPU code between CUDA and HIP architectures, with 60k verified code pairs.

Ax Shashank Sharma, Janina Hoffmann, Vinay Namboodiri 4/22/2026

MRS: Multi-Resolution Skills for HRL Agents

Hierarchical RL approach using multi-resolution skills to improve manager subgoal selection and performance on agile tasks.

Ax Ba-Hien Tran, Van Minh Nguyen 4/22/2026

Highly Efficient and Effective LLMs with Multi-Boolean Architectures

Novel multi-Boolean architecture framework for binarizing LLM weights during training for efficiency without full-precision latents.

Ax Minh-Anh Nguyen, Bao Nguyen, Ha Lan N. T., Tuan Anh Hoang, Duc-Trong Le, Dung D. Le 4/22/2026

VoteGCL: Enhancing Graph-based Recommendations with Majority-Voting LLM-Rerank Augmentation

LLM-based data augmentation framework for recommendation systems using majority-voting reranking to address data sparsity.

Ax Jingfeng Pan, Jiahao Chen 4/22/2026

QTMRL: An Agent for Quantitative Trading Decision-Making Based on Multi-Indicator Guided Reinforcement Learning

Reinforcement learning agent for quantitative trading combining multi-indicator technical analysis with RL for adaptive decision-making.

Ax Pin-Yen Chiu, I-Sheng Fang, Jun-Cheng Chen 4/22/2026

Text Slider: Efficient and Plug-and-Play Continuous Concept Control for Image/Video Synthesis via LoRA Adapters

LoRA-based method for efficient continuous concept control in diffusion models for image/video synthesis with plug-and-play design.

Ax Zhen Zhu, Yiming Gong, Yao Xiao, Yaoyao Liu, Derek Hoiem 4/22/2026

How to Teach Large Multimodal Models New Skills

Research on sequential fine-tuning of large multimodal models showing skill recovery across different tasks and model families.

Ax Conor Hassan, Nasrulloh Loka, Cen-You Li, Daolang Huang, Paul E. Chang, Yang Yang, Francesco Silvestrin, Samuel Kaski, Luigi Acerbi 4/22/2026

Efficient Autoregressive Inference for Transformer Probabilistic Models

Paper on efficient autoregressive inference for transformer probabilistic models balancing set-conditioning with joint distributions.

Ax Yang Yang, Severi Rissanen, Paul E. Chang, Nasrulloh Loka, Daolang Huang, Arno Solin, Markus Heinonen, Luigi Acerbi 4/22/2026

PriorGuide: Test-Time Prior Adaptation for Simulation-Based Inference

Paper on test-time prior adaptation for simulation-based inference using diffusion models for Bayesian inference.

Ax Xiqiao Xiong, Ouxiang Li, Zhuo Liu, Moxin Li, Wentao Shi, Fengbin Zhu, Qifan Wang, Fuli Feng 4/22/2026

TROJail: Trajectory-Level Optimization for Multi-Turn Large Language Model Jailbreaks with Process Rewards

TROJail uses trajectory-level optimization with process rewards to learn multi-turn jailbreak strategies against LLMs.

Ax Simon Welker, Bunlong Lay, Maris Hillemann, Tal Peer, Timo Gerkmann 4/22/2026

Real-Time Streamable Generative Speech Restoration with Flow Matching

Stream.FM applies flow matching for real-time streamable speech restoration with 32ms algorithmic latency.

Ax Qiushi Han, David Simchi-Levi, Renfei Tan, Zishuo Zhao 4/22/2026

Multi-agent Adaptive Mechanism Design

DRAM framework combines mechanism design and online learning for sequential multi-agent truthful reporting.

Ax Lars van der Laan, Nathan Kallus 4/22/2026

Fitted Q Evaluation Without Bellman Completeness via Stationary Weighting

Fitted Q-evaluation theory for off-policy reinforcement learning without requiring Bellman completeness using stationary weighting.

Ax Rachmad Vidya Wicaksana Putra, Pasindu Wickramasinghe, Muhammad Shafique 4/22/2026

QSLM: A Performance- and Memory-aware Quantization Framework with Tiered Search Strategy for Spike-driven Language Models

QSLM quantization framework with tiered search optimizes spike-driven language models for embedded deployment.

Ax Janvijay Singh, Dilek Hakkani-T\"ur 4/22/2026

Do LLMs Encode Functional Importance of Reasoning Tokens?

Study investigates whether LLMs encode functional importance of individual reasoning tokens for reasoning chain compression.

Ax Zuang Wang, Yongqiang Wang 4/22/2026

Local Updates in Distributed Optimization: Provable Acceleration and Topology Effects

Theoretical analysis of local updates in distributed optimization showing acceleration benefits and topology effects in federated settings.

Ax Basab Jha, Firoj Paudel, Ujjwal Puri, Ethan Henkel, Zhang Yuting, Mateusz Kowalczyk, Mei Huang, Choi Donghyuk, Wang Junhao 4/22/2026

SAGE-32B: Agentic Reasoning via Iterative Distillation

SAGE-32B is a 32B parameter model fine-tuned via iterative distillation for agentic reasoning, task decomposition, and tool usage.

Ax Meiru Zhang, Zaiqiao Meng, Nigel Collier 4/22/2026

Failure Modes in Multi-Hop QA: The Weakest Link Effect and the Recognition Bottleneck

Multi-Focus Attention Instruction probe disentangles recognition vs synthesis failures in multi-hop LLM reasoning.

Ax Hyunjong Ok, Jaeho Lee 4/22/2026

Lost in the Prompt Order: Revealing the Limitations of Causal Attention in Language Models

Study reveals LLMs are highly sensitive to prompt order in multiple-choice QA due to causal attention limitations.

Ax Zhaoyan Gong, Zhiqiang Liu, Songze Li, Xiaoke Guo, Yuanxiang Liu, Xinle Deng, Zhizhen Liu, Lei Liang, Huajun Chen, Wen Zhang 4/22/2026

Temp-R1: A Unified Autonomous Agent for Complex Temporal KGQA via Reverse Curriculum Reinforcement Learning

Temp-R1 is an autonomous agent for temporal knowledge graph question answering trained via reverse curriculum reinforcement learning.

Ax Jiale Zhao, Ke Fang, Lu Cheng 4/22/2026

When and What to Ask: AskBench and Rubric-Guided RLVR for LLM Clarification

AskBench evaluates and improves LLM ability to request clarification on ambiguous prompts using reinforcement learning with rubric guidance.

Ax Mainak Singha, Sarthak Mehrotra, Paolo Casari, Subhasis Chaudhuri, Elisa Ricci, Biplab Banerjee 4/22/2026

CLIPoint3D: Language-Grounded Few-Shot Unsupervised 3D Point Cloud Domain Adaptation

CLIPoint3D adapts vision-language models like CLIP for 3D point cloud domain adaptation with few-shot unsupervised learning.

Ax Zongyue Qin, Raghavv Goel, Mukul Gagrani, Risheek Garrepalli, Mingu Lee, Yizhou Sun 4/22/2026

ConFu: Contemplate the Future for Better Speculative Sampling

ConFu improves speculative decoding for LLM inference acceleration by enhancing draft model quality to propose better candidate tokens for verification.

Ax Jeonghye Kim, Xufang Luo, Minbeom Kim, Sangmook Lee, Dohyung Kim, Jiwon Jeon, Dongsheng Li, Yuqing Yang 4/22/2026

Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs?

Self-distillation degrades LLM reasoning by suppressing epistemic verbalization of uncertainty; controlled experiments isolate degradation mechanisms.

Ax Caio Vicentino 4/22/2026

PolarQuant: Optimal Gaussian Weight Quantization via Hadamard Rotation for LLM Compression

PolarQuant post-training quantization for LLMs uses Hadamard rotation and Gaussian weight distribution for near-lossless compression.

Ax Aastha A K Verma, Anwoy Chatterjee, Mehak Gupta, Tanmoy Chakraborty 4/22/2026

Multilingual Language Models Encode Script Over Linguistic Structure

Analysis reveals multilingual language models organize internal representations by orthographic script rather than linguistic structure across language families.

Ax Tanzim Ahad, Ismail Hossain, Md Jahangir Alam, Sai Puppala, Yoonpyo Lee, Syed Bahauddin Alam, Sajedul Talukder 4/22/2026

Semantic Intent Fragmentation: A Single-Shot Compositional Attack on Multi-Agent AI Pipelines

Semantic Intent Fragmentation attack exploits LLM orchestration systems where composed subtasks violate security policy despite individual benignness.

Ax Olga Chetverina 4/22/2026

A Triadic Suffix Tokenization Scheme for Numerical Reasoning

Triadic Suffix Tokenization improves LLM numerical reasoning by partitioning digits into three-digit triads with explicit magnitude markers.

Ax Ryan Faulkner, Anushka Deshpande, David Guzman Piedrahita, Joel Z. Leibo, Zhijing Jin 4/22/2026

Evaluating Cooperation in LLM Social Groups through Elected Leadership

Multi-agent study evaluating whether LLMs can cooperate on resource governance through elected leadership and self-governance mechanisms.

Ax Hongyi Jin, Bohan Hou, Guanjie Wang, Ruihang Lai, Jinqi Chen, Zihao Ye, Yaxing Cai, Yixin Dong, Xinhao Cheng, Zhihao Zhang, Yilong Zhao, Yingyi Huang, Lijie Yang, Jinchen Jiang, Gabriele Oliaro, Jianan Ji, Xupeng Miao, Vinod Grover, Todd C. Mowry, Zhihao Jia, Tianqi Chen 4/22/2026

Event Tensor: A Unified Abstraction for Compiling Dynamic Megakernel

Event Tensor abstraction eliminates kernel launch overheads in LLM inference by fusing operators into persistent kernels handling dynamic shapes.

Ax Jon-Paul Cacioli 4/22/2026

The Metacognitive Monitoring Battery: A Cross-Domain Benchmark for LLM Self-Monitoring

Cross-domain metacognitive benchmark for LLMs with 524 items across six cognitive domains using human psychometric methodology.

Ax Baoyou Chen, Hanchen Xia, Peng Tu, Haojun Shi, Shan Mu, Weihao Yuan, Siyu Zhu 4/22/2026

BARD: Bridging AutoRegressive and Diffusion Vision-Language Models Via Highly Efficient Progressive Block Merging and Stage-Wise Distillation

BARD framework bridges autoregressive and diffusion vision-language models via progressive block merging and stage-wise distillation for efficient inference.

Ax Samrendra Roy, Syed Bahauddin Alam 4/22/2026