Isolater - Feed

Ax Chengshuai Shi, Wenzhe Li, Xinran Liang, Yizhou Lu, Wenjia Yang, Ruirong Feng, Seth Karten, Ziran Yang, Zihan Ding, Gabriel Sarch, Danqi Chen, Karthik Narasimhan, Chi Jin 5/5/2026

Odysseus: Scaling VLMs to 100+ Turn Decision-Making in Games via Reinforcement Learning

Odysseus extends vision-language models to 100+ turn decision-making in games using reinforcement learning, improving long-horizon performance.

Ax Tianyu Hu, Weikai Lin, Weizhi Zhang, Jing Ma, Song Wang 5/5/2026

MemRouter: Memory-as-Embedding Routing for Long-Term Conversational Agents

MemRouter decouples memory management from LLM generation in conversational agents using embedding-based routing for long-term memory decisions.

Ax Juvy C. Grume, John Paul P. Miranda, Aileen P. De Leon, Jordan L. Salenga, Hilene E. Hernandez, Mark Anthony A. Castro, Vernon Grace M. Maniago, Joel D. Canlas, Joel B. Quiambao 5/5/2026

Pedagogical Promise and Peril of AI: A Text Mining Analysis of ChatGPT Research Discussions in Programming Education

Text mining analysis of ChatGPT research publications in programming education, identifying four dominant themes in scholarly discourse.

Ax Chenyu Huang, Jianghao Lin, Zhengyang Tang, Bo Jiang, Ruoqing Jiang, Benyou Wang, Lai Wei 5/5/2026

AlphaInventory: Evolving White-Box Inventory Policies via Large Language Models with Deployment Guarantees

AlphaInventory applies LLM-based evolutionary search to optimize inventory policies in online, non-stationary environments with deployment guarantees.

Ax Zuyao You, Zhesong Yu, Mingyu Liu, Bilei Zhu, Yuan Wan, Zuxuan Wu 5/5/2026

GaMMA: Towards Joint Global-Temporal Music Understanding in Large Multimodal Models

Large multimodal model for music understanding combining audio encoders with mixture-of-experts design for time-series and non-time-series music tasks.

Ax Fazle Rabbi, Lin Ling, Song Wang, Jinqiu Yang 5/5/2026

Social Bias in LLM-Generated Code: Benchmark and Mitigation

Benchmark study of social bias in LLM-generated code across 343 real-world tasks, extending prior Solar work with SocialBias-Bench evaluation framework.

Ax Aninda Ray 5/5/2026

Agent Capsules: Quality-Gated Granularity Control for Multi-Agent LLM Pipelines

Agent Capsules runtime optimizes multi-agent LLM pipelines by merging agents adaptively while maintaining quality.

Ax Pankaj Gupta, Kartik Bose 5/5/2026

RadLite: Multi-Task LoRA Fine-Tuning of Small Language Models for CPU-Deployable Radiology AI

RadLite uses LoRA fine-tuning on small language models for radiology tasks deployable on consumer CPUs.

Ax Zhixiong Zhao, Zukang Xu, Dawei Yang 5/5/2026

BWLA: Breaking the Barrier of W1AX Post-Training Quantization for LLMs

BWLA method achieves 1-bit weight and activation quantization for LLMs, enabling efficient deployment.

Ax Alfredo Metere 5/5/2026

Skills as Verifiable Artifacts: A Trust Schema and a Biconditional Correctness Criterion for Human-in-the-Loop Agent Runtimes

Proposes trust schema and verification framework for agent skills as deployable artifacts in LLM agent runtimes.

Ax Shouyu Yin, Zhao Tian, Junjie Chen, Shikai Guo 5/5/2026

Improving LLM Code Generation via Requirement-Aware Curriculum Reinforcement Learning

Improves LLM code generation for complex requirements using requirement-aware curriculum reinforcement learning.

Ax Xin Du, Kumiko Tanaka-Ishii 5/5/2026

Escaping Mode Collapse in LLM Generation via Geometric Regulation

Addresses mode collapse in LLM text generation through geometric regulation using dynamical systems perspective.

Ax Kenneth J. K. Ong 5/5/2026

Impact of Task Phrasing on Presumptions in Large Language Models

Studies how task phrasing affects LLM presumptions and adaptation using iterated prisoner's dilemma experiments.

Ax P. Rosciszewski, A. Krzywaniak, S. Iserte, K. Rojek, P. Gepner 5/5/2026

Adaptation of AI-accelerated CFD Simulations to the IPU platform

Evaluates Intelligence Processing Units for AI-accelerated CFD simulations using TensorFlow and Poplar SDK.

Ax Lu Dai, Liang Sun, Fanpu Cao, Ziyang Rao, Cehao Yang, Hao Liu, Hui Xiong 5/5/2026

LLM-Oriented Information Retrieval: A Denoising-First Perspective

Perspective on LLM-oriented information retrieval from denoising-first angle, addressing attention budgets and hallucination vulnerabilities in RAG.

Ax Zhanwei Wang, Huiling Yang, Min Sheng, Khaled B. Letaief, Kaibin Huang 5/5/2026

Space Network of Experts: Architecture and Expert Placement

Architecture for deploying large LLMs across satellite networks leveraging solar energy with optimized expert placement strategies.

Ax Abdurrahman Javat, Allan Kazakov 5/5/2026

Silicon Showdown: Performance, Efficiency, and Ecosystem Barriers in Consumer-Grade LLM Inference

Empirical analysis comparing Nvidia and Apple Silicon ecosystems for local LLM inference on consumer hardware with 70B+ models.

Ax Dongxin Guo, Jikun Wu, Siu Ming Yiu 5/5/2026

SAGA: Workflow-Atomic Scheduling for AI Agent Inference on GPU Clusters

SAGA: Workflow-atomic scheduling system for GPU clusters treating entire AI agent workflows as first-class units, reducing latency 3-8x.

Ax Ziwen Zhao, Menglin Yang 5/5/2026

Hierarchical Abstract Tree for Cross-Document Retrieval-Augmented Generation

Hierarchical abstract tree method for cross-document multi-hop retrieval-augmented generation addressing clustering and distribution challenges.

Ax Junda Ying, Yuxuan Wang, Bowen Yang, Peijie Zhou, Lei Zhang 5/5/2026

Beyond Continuity: Simulation-free Reconstruction of Discrete Branching Dynamics from Single-cell Snapshots

Method for reconstructing discrete cellular trajectories from snapshots using unbalanced optimal transport accounting for birth-death dynamics.

Ax Michito Takeshita, Takuro Kawada, Takumi Ohashi, Shunsuke Kitada, Hitoshi Iyatomi 5/5/2026

A11y-Compressor: A Framework for Enhancing the Efficiency of GUI Agent Observations through Visual Context Reconstruction and Redundancy Reduction

A11y-Compressor: Framework compressing accessibility trees for GUI agents by reconstructing visual context and reducing redundancy.

Ax James Mooney, Zae Myung Kim, Young-Jun Lee, Dongyeop Kang 5/5/2026

Structure Liberates: How Constrained Sensemaking Produces More Novel Research Output

SCISENSE: Framework operationalizing scientific ideation as structured cognitive stages with 100K-scale citation-conditioned dataset.

Ax Aharon Azulay, Jan Dubi\'nski, Zhuoyun Li, Atharv Mittal, Yossi Gandelsman 5/5/2026

Jailbreaking Vision-Language Models Through the Visual Modality

Four jailbreak attacks against vision-language models exploiting visual modality through symbol encoding, substitution, and manipulation.

Ax Yao Ni, Jeremie Houssineau, Yew Soon Ong, Piotr Koniusz 5/5/2026

Possibilistic Predictive Uncertainty for Deep Learning

Possibilistic approach to epistemic uncertainty modeling in deep neural networks balancing Bayesian rigor with computational efficiency.

Ax Massimo Rondelli, Francesco Pivi, Maurizio Gabbrielli 5/5/2026

BlenderRAG: High-Fidelity 3D Object Generation via Retrieval-Augmented Code Synthesis

BlenderRAG: Retrieval-augmented generation system for generating Blender code from natural language using 500 curated multimodal examples.

Ax Jiali Cui, Zhiqiang Lao, Heather Yu 5/5/2026

Learning Multimodal Energy-Based Model with Multimodal Variational Auto-Encoder via MCMC Revision

Energy-based models combined with multimodal VAEs via MCMC for learning complex dependencies in multimodal data.

Ax Zhijie Cai, Haolong Chen, Guangxu Zhu 5/5/2026

AdaMeZO: Adam-style Zeroth-Order Optimizer for LLM Fine-tuning Without Maintaining the Moments

AdaMeZO: Adam-style zeroth-order optimizer for LLM fine-tuning that reduces GPU memory by using only forward passes, improving convergence over prior MeZO method.

Ax Andrzej Ruszczynski, Tiangang Zhang 5/5/2026

Reinforcement Learning with Markov Risk Measures and Multipattern Risk Approximation

Q-learning method with multipattern risk-averse Markov decision processes and mini-batch risk measures with regret bounds.

Ax Jiaming Zhang, Yujie Yang, Yao Lyu, Shengbo Eben Li, Liping Zhang 5/5/2026

Augmented Lagrangian Multiplier Network for State-wise Safety in Reinforcement Learning

Augmented Lagrangian multiplier network for enforcing state-wise safety constraints in reinforcement learning with improved training stability.

Ax Ziyu Zheng, Yaming Yang, Zhe Wang, Ziyu Guan, Wei Zhao 5/5/2026

Empowering Heterogeneous Graph Foundation Models via Decoupled Relation Alignment

Decoupled relation alignment approach extends graph foundation models to multi-domain heterogeneous graphs while preserving type-specific semantics.

Ax Zihao Ding, Beining Wu, Jun Huang 5/5/2026

EASE: Federated Multimodal Unlearning via Entanglement-Aware Anchor Closure

EASE method enables federated unlearning in multimodal models by decoupling entangled knowledge across image-text modalities and client gradients.

Ax Xihao Chen, Yangyang Guo, Roger Zimmermann 5/5/2026

Make Your LVLM KV Cache More Lightweight

LightKV reduces KV cache memory overhead in large vision-language models by exploiting token redundancy during inference prefill stage.

Ax Alfredo Madrid-Garc\'ia, Miguel Rujas 5/5/2026

When RAG Chatbots Expose Their Backend: An Anonymized Case Study of Privacy and Security Risks in Patient-Facing Medical AI

Security assessment of patient-facing RAG medical chatbot exposing privacy and backend vulnerabilities, highlighting governance gaps in medical AI.

Ax Ziyang Huang, Yi Cao, Ali K. Shargh, Jing Luo, Ruidong Mei, Mohd Zaki, Zhan Liu, Wyatt Bunstine, William Jurayj, Somdatta Goswami, Tyrel McQueen, Michael Shields, Jaafar El-Awady, Paulette Clancy, Benjamin Van Durme, Nicholas Andrews, William Walden, Daniel Khashabi 5/5/2026