Isolater - Feed

Ax Afrozah Nadeem, Mark Dras, Usman Naseem 4/24/2026

Fairness Evaluation and Inference Level Mitigation in LLMs

Inference-level mitigation techniques for fairness issues in LLMs without expensive retraining.

Ax Ruiyi Yang, Hao Xue, Imran Razzak, Hakim Hacid, Flora D. Salim 4/24/2026

RELOOP: Recursive Retrieval with Multi-Hop Reasoner and Planners for Heterogeneous QA

RAG framework using hierarchical sequences for multi-hop reasoning over heterogeneous sources (documents, tables, KGs).

Ax Luca Carlini, Dennis Pierantozzi, Mauro Orazio Drago, Chiara Lena, Cesare Hassan, Elena De Momi, Danail Stoyanov, Sophia Bano, Mobarak I. Hoque 4/24/2026

When to Trust the Answer: Question-Aligned Semantic Nearest Neighbor Entropy for Safer Surgical VQA

Uncertainty estimation method for surgical VQA systems accounting for question context in safety-critical applications.

Ax Yiling He, Junchi Lei, Hongyu She, Shuo Shao, Xinran Zheng, Yiping Liu, Zhan Qin, Lorenzo Cavallaro 4/24/2026

Retrofit: Continual Learning with Controlled Forgetting for Binary Security Detection and Analysis

Continual learning framework for binary malware detection with controlled forgetting to adapt as threats evolve.

Ax Haotian Dong, Ye Li, Rongwei Lu, Chen Tang, Shu-Tao Xia, Zhi Wang 4/24/2026

VVS: Accelerating Speculative Decoding for Visual Autoregressive Generation via Partial Verification Skipping

Accelerates visual autoregressive generation via partial verification skipping in speculative decoding.

Ax Kushal Agrawal, Frank Xiao, Guido Bergman, Asa Cooper Stickland 4/24/2026

Why Do Language Model Agents Whistleblow?

Studies why LLM agents disclose information to external parties contrary to user instructions, examining alignment behavior.

Ax Congren Dai, Yue Yang, Krinos Li, Huichi Zhou, Shijie Liang, Bo Zhang, Enyang Liu, Ge Jin, Hongran An, Haosen Zhang, Peiyuan Jing, Kinhei Lee, Z henxuan Zhang, Xiaobing Li, Maosong Sun 4/24/2026

Musical Score Understanding Benchmark: Evaluating Large Language Models' Comprehension of Complete Musical Scores

Benchmark for evaluating LLM and vision-language model comprehension of musical scores in ABC and PDF formats.

Ax Kevin Stowe, Svetlana Afanaseva, Rodolfo Raimundo, Yitao Sun, Kailash Patil 4/24/2026

Identifying Bias in Machine-generated Text Detection

Study identifying and characterizing biases in machine-generated text detection systems across different text types and domains.

Ax Ming Li, Chenrui Fan, Yize Cheng, Soheil Feizi, Tianyi Zhou 4/24/2026

Schoenfeld's Anatomy of Mathematical Reasoning by Language Models

Framework abstracting LLM reasoning traces into functional steps using Schoenfeld's theory to analyze mathematical reasoning structure.

Ax Yun Bian, Yi Chen, HaiQuan Wang, ShiHao Li, Zhe Cui 4/24/2026

Focus on What Matters: Fisher-Guided Adaptive Multimodal Fusion for Vulnerability Detection

Multimodal fusion approach using Fisher information for vulnerability detection in code combining sequences and property graphs.

Ax Yueqing Hu, Xinyang Peng, Shuting Peng, Hanqi Wang, Tianhong Wang 4/24/2026

H\'an D\=an Xu\'e B\`u (Mimicry) or Q\=ing Ch\=u Y\'u L\'an (Mastery)? A Cognitive Perspective on Reasoning Distillation in Large Language Models

Analysis showing reasoning distillation via supervised fine-tuning fails to transmit cognitive structure of reinforcement-trained reasoning models.

Ax Zehua Liu, Shuqi Liu, Tao Zhong, Mingxuan Yuan 4/24/2026

RIFT: Repurposing Negative Samples via Reward-Informed Fine-Tuning

Framework reweighting negative samples during LLM fine-tuning to improve data efficiency beyond supervised fine-tuning and rejection sampling.

Ax Jiaying Zhang, Lei Shi, Jiguo Li, Jun Xu, Jiuchong Gao, Jinghua Hao, Renqing He 4/24/2026

GeoRA: Geometry-Aware Low-Rank Adaptation for RLVR

Parameter-efficient adaptation method preserving geometric structure of pre-trained models during reinforcement learning with verifiable rewards.

Ax Binglei Lou, Haoran Wu, Kevin Lau, Gregor MacDonald, Jiayi Nie, Yao Lai, Can Xiao, Xuan Guo, Jianyi Cheng, Rika Antonova, Robert Mullins, Aaron Zhao 4/24/2026

NPU Design for Diffusion Language Model Inference

NPU architecture design optimized for diffusion-based LLM inference with bidirectional attention and block-wise KV cache patterns.

Ax Abhinav Gupta, Toben H. Mintz, Jesse Thomason 4/24/2026

Words that make SENSE: Sensorimotor Norms in Learned Lexical Token Representations

Projection model predicting sensorimotor norms from word embeddings to ground language understanding in embodied experience.

Ax Muhammad Ahmed Mohsin, Muhammad Umer, Ahsan Bilal, Zihao He, Muhammad Usman Rafique, Asad Aali, Muhammad Ali Jamshed, John M. Cioffi, Emily Fox 4/24/2026

Continuous-Utility Direct Preference Optimization

Framework replacing binary preference labels with continuous utility scores for fine-grained alignment of LLM reasoning capabilities.

Ax Zhuoyuan Wang, Hanjiang Hu, Xiyu Deng, Saviz Mowlavi, Yorie Nakahira 4/24/2026

OpInf-LLM: Parametric PDE Solving with LLMs via Operator Inference

Framework leveraging LLMs for solving parametric PDEs through code generation and operator inference techniques.

Ax Miao Ye, Yanye Chen, Yong Wang, Cheng Zhu, Qiuxiang Jiang, Gai Huang, Feng Ding 4/24/2026

An Overlay Multicast Routing Method Based on Network Situational Awareness and Hierarchical Multi-Agent Reinforcement Learning

Multi-agent reinforcement learning approach for overlay multicast routing with network situational awareness.

Ax Shahriar Golchin, Marc Wetter 4/24/2026

Intent Laundering: AI Safety Datasets Are Not What They Seem

Evaluation of adversarial safety datasets revealing they rely on triggering cues rather than reflecting real-world attacks.

Ax Yangyi Fang, Jiaye Lin, Xiaoliang Fu, Cong Qin, Haolin Shi, Chaowen Hu, Lu Pan, Ke Zeng, Xunliang Cai 4/24/2026

How to Allocate, How to Learn? Dynamic Rollout Allocation and Advantage Modulation for Policy Optimization

Methods for dynamic rollout allocation and advantage modulation in reinforcement learning with verifiable rewards for LLM reasoning.

Ax Ishraq Tashdid, Kimia Tasnia, Alexander Garcia, Jonathan Valamehr, Sazadur Rahman 4/24/2026

ATLAS: AI-Assisted Threat-to-Assertion Learning for System-on-Chip Security Verification

LLM-driven framework combining threat modeling and formal verification for SoC security property generation and assertion checking.

Ax Eduardo Di Santi 4/24/2026

Cognitive Amplification vs Cognitive Delegation in Human-AI Systems: A Metric Framework

Framework distinguishing cognitive amplification from cognitive delegation in human-AI systems with metrics for hybrid performance.

Ax Dimitris Mitropoulos, Nikolaos Alexopoulos, Georgios Alexopoulos, Diomidis Spinellis 4/24/2026

Measuring and Exploiting Contextual Bias in LLM-Assisted Security Code Review

Study of contextual bias and framing effects in LLM-based code review systems for vulnerability detection in CI/CD pipelines.

Ax Rui Yang Tan, Yujia Hu, Roy Ka-Wei Lee 4/24/2026

Structured Visual Narratives Undermine Safety Alignment in Multimodal Large Language Models

Benchmark studying comic-template jailbreaks that exploit visual narratives to compromise multimodal LLM safety alignment.

Ax Hannah Collison, Benjamin Van Durme, Daniel Khashabi 4/24/2026

Crystal: Characterizing Relative Impact of Scholarly Publications

System using LLMs to jointly rank cited papers within a citing paper to measure relative impact with positional bias mitigation.

Ax Ravi Ranjan, Agoritsa Polyzou 4/24/2026

VLA-Forget: Vision-Language-Action Unlearning for Embodied Foundation Models

Unlearning framework for vision-language-action embodied foundation models to remove unsafe or privacy-sensitive behaviors from robotic policies.

Ax Wuyang Luan, Junhui Li, Weiguang Zhao, Wenjian Zhang, Tieru Wu, Rui Ma 4/24/2026

Rectified Schr\"odinger Bridge Matching for Few-Step Visual Navigation

Fast visual navigation approach using rectified Schrödinger bridge matching for embodied AI agents with few integration steps.

Ax Junxiao Yang, Haoran Liu, Jinzhe Tu, Jiale Cheng, Zhexin Zhang, Shiyao Cui, Jiaqi Weng, Jialing Tao, Hui Xue, Hongning Wang, Han Qiu, Minlie Huang 4/24/2026

LASA: Language-Agnostic Semantic Alignment at the Semantic Bottleneck for LLM Safety

Method for improving LLM safety alignment across low-resource languages by addressing semantic bottleneck and language-agnostic understanding.

Ax Syed Md Mukit Rashid, Abdullah Al Ishtiaq, Kai Tu, Yilu Dong, Tianwei Wu, Ali Ranjbar, Tianchang Yang, Najrin Sultana, Shagufta Mehnaz, Syed Rafiul Hussain 4/24/2026

LogicEval: A Systematic Framework for Evaluating Automated Repair Techniques for Logical Vulnerabilities in Real-World Software

Framework for evaluating automated repair of logical vulnerabilities in software using recent LLM techniques for semantic understanding.

Ax Yuseon Choi, Jingu Lee, Jungjun Oh, Sunjoo Whang, Byeongcheol Kim, Minsung Kim, Hoi-Jun Yoo, Sangjin Kim 4/24/2026

ELMoE-3D: Leveraging Intrinsic Elasticity of MoE for Hybrid-Bonding-Enabled Self-Speculative Decoding in On-Premises Serving

Optimization technique leveraging mixture-of-experts elasticity for self-speculative decoding in LLM serving to reduce memory bottlenecks.

Ax Seyedreza Mohseni, Sarvesh Baskar, Edward Raff, Manas Gaur 4/24/2026

Analyzing Chain of Thought (CoT) Approaches in Control Flow Code Deobfuscation Tasks

Study of chain-of-thought prompting with LLMs for code deobfuscation, using step-by-step reasoning for control flow analysis.

Ax Robert Feldt, Per Lenberg, Julian Frattini, Dhasarathy Parthasarathy 4/24/2026

The Semi-Executable Stack: Agentic Software Engineering and the Expanding Scope of SE

Analysis of agentic AI systems and LLMs in software engineering, discussing AI threats to SE tasks like testing, bug fixing, and integration work.

Ax Rajveer Bachkaniwala, Chengqi Luo, Richard So, Divya Mahajan, Kexin Rong 4/24/2026

Stream2LLM: Overlap Context Streaming and Prefill for Reduced Time-to-First-Token (TTFT)

Technique for reducing time-to-first-token in LLM inference by overlapping context streaming and prefill under concurrent requests.

Ax Jiaoyang Ruan, Xin Gao, Yinda Chen, Hengyu Zeng, Liang Du, Guanghao Li, Jie Fu, Jian Pu 4/24/2026

Reasoning on the Manifold: Bidirectional Consistency for Self-Verification in Diffusion Language Models

Method for verifying reasoning traces in diffusion language models using geometric perspective and bidirectional consistency checking.

Ax Sua Lee, Sanghee Park, Jinbae Im 4/24/2026

MM-JudgeBias: A Benchmark for Evaluating Compositional Biases in MLLM-as-a-Judge

Benchmark for evaluating biases in multimodal LLMs used as automatic judges, identifying failures in integrating visual and textual evidence.

Ax Alankrit Chona, Igor Kozlov, Ambuj Kumar 4/24/2026

Cyber Defense Benchmark: Agentic Threat Hunting Evaluation for LLMs in SecOps

Benchmark for evaluating LLM agents on threat hunting in security operations, using 106 real attack procedures from Windows event logs across MITRE ATT&CK techniques.

Ax Kihyuk Lee 4/24/2026

Cross-Model Consistency of AI-Generated Exercise Prescriptions: A Repeated Generation Study Across Three Large Language Models

Comparative analysis of output consistency across GPT-4, Claude, and Gemini for repeated exercise prescription generation in clinical scenarios.

Ax Michael Richter 4/24/2026

Model Capability Assessment and Safeguards for Biological Weaponization

Benchmarking study measuring biological misuse capability across advanced LLMs using STEM prompts to assess AI safety safeguards.

Ax Jeonghyeon Kim, Byeongjun Joung, Junwon Lee, Joohyung Lee, Taehoon Min, Sunjae Lee 4/24/2026

AgentLens: Adaptive Visual Modalities for Human-Agent Interaction in Mobile GUI Agents

Studies visual communication modalities for mobile GUI agents, balancing user transparency and multitasking through adaptive display approaches.

Ax Changho Han, Songsoo Kim, Dong Won Kim, Leo Anthony Celi, Jaewoong Kim, SungA Bae, Dukyong Yoon 4/24/2026

Surrogate modeling for interpreting black-box LLMs in medical predictions

Surrogate modeling framework using simplified models to interpret and explain black-box LLM behavior in medical prediction tasks.

Ax Markus Knauer, Edoardo Fiorini, Maximilian M\"uhlbauer, Stefan Schneyer, Promwat Angsuratanawech, Florian Samuel Lay, Timo Bachmann, Samuel Bustamante, Korbinian Nottensteiner, Freek Stulp, Alin Albu-Sch\"affer, Jo\~ao Silv\'erio, Thomas Eiband 4/24/2026