Isolater - Feed

Ax Karthik Ragunath Ananda Kumar, Subrahmanyam Arunachalam 3/18/2026

Learning to Present: Inverse Specification Rewards for Agentic Slide Generation

RL environment where LLM agents learn to generate professional presentations through research, planning, and tool use with multi-component reward system.

Ax Rui Ge, Yichao Fu, Yuyang Qian, Junda Su, Yiming Zhao, Peng Zhao, Hao Zhang 3/18/2026

Internalizing Agency from Reflective Experience

Method for training LLM agents to leverage rich environment feedback through reflective experience and post-training, improving long-horizon planning.

Ax Tianyu Xie, Jinfa Huang, Yuexiao Ma, Rongfang Luo, Yan Yang, Wang Chen, Yuhui Zeng, Ruize Fang, Yixuan Zou, Xiawu Zheng, Jiebo Luo, Rongrong Ji 3/18/2026

SocialOmni: Benchmarking Audio-Visual Social Interactivity in Omni Models

Benchmark evaluating audio-visual social interactivity capabilities of omni-modal LLMs in dynamic dialogue settings.

Ax Chenyu Ge 3/18/2026

SAC-NeRF: Adaptive Ray Sampling for Neural Radiance Fields via Soft Actor-Critic Reinforcement Learning

RL framework using Soft Actor-Critic to learn adaptive ray sampling policies for efficient neural radiance field rendering.

Ax Suyash Mishra, Srikanth Patil, Satyanarayan Pati, Sagar Sahu, Baddu Narendra 3/18/2026

Finder: A Multimodal AI-Powered Search Framework for Pharmaceutical Data Retrieval

Multimodal AI search framework combining vector search, hybrid retrieval, and reasoning for pharmaceutical data across text, images, audio, and video.

Ax Yu Li, Yuchen Zheng, Giles Hamilton-Fletcher, Marco Mezzavilla, Yao Wang, Sundeep Rangan, Maurizio Porfiri, Zhou Yu, John-Ross Rizzo 3/18/2026

Exploring the Use of VLMs for Navigation Assistance for People with Blindness and Low Vision

Evaluation of VLMs (GPT-4V, Gemini, Claude, Llava) for navigation assistance tasks for people with vision impairments.

Ax Guangchen Lan 3/18/2026

Alternating Reinforcement Learning with Contextual Rubric Rewards

Framework extending RLHF using multi-dimensional rubric-based rewards instead of scalar signals for RL training.

Ax Zeyu Zhang, Xiangxiang Dai, Ziyi Han, Xutong Liu, John C. S. Lui 3/18/2026

Steering Frozen LLMs: Adaptive Social Alignment via Online Prompt Routing

Inference-time governance approach for LLMs using adaptive prompt routing to enable social alignment without retraining.

Ax Keivan Alizadeh, Parshin Shojaee, Minsik Cho, Mehrdad Farajtabar 3/18/2026

Recursive Language Models Meet Uncertainty: The Surprising Effectiveness of Self-Reflective Program Search for Long Context

Study on recursive language models with self-reflective program search for long-context handling, addressing information extraction challenges.

Ax Ruixi Lin 3/18/2026

Discovering the Hidden Role of Gini Index In Prompt-based Classification

Analysis of Gini Index role in prompt-based classification for detecting and optimizing class accuracy disparities in long-tailed datasets.

Ax Liu Hung Ming 3/18/2026

Beyond Reward Suppression: Reshaping Steganographic Communication Protocols in MARL via Dynamic Representational Circuit Breaking

Defense mechanism against steganographic collusion in multi-agent reinforcement learning using dynamic representational circuit breaking.

Ax Peiyu Yang, Naveed Akhtar, Jiantong Jiang, Ajmal Mian 3/18/2026

Attribution-Guided Model Rectification of Unreliable Neural Network Behaviors

Model rectification framework using attribution-guided rank-one editing to fix unreliable neural network behaviors on corrupted samples.

Ax Lansiaux Edouard, Leman Margaux 3/18/2026

OrthoAI v2: From Single-Agent Segmentation to Dual-Agent Treatment Planning for Clear Aligners

Open-source pipeline extending single-agent AI orthodontic treatment planning to dual-agent framework with improved tooth segmentation and landmarks.

Ax Kyle Dumont, Nicholas Herbert, Hayder Tirmazi, Shrikanth Upadhayaya 3/18/2026

DRCY: Agentic Hardware Design Reviews

AI agent system for hardware design reviews using LLMs to verify semantic correctness of component connections against datasheets.

Ax Alexandre Cristov\~ao Maiorano 3/18/2026

Automated Self-Testing as a Quality Gate: Evidence-Driven Release Management for LLM Applications

Framework for LLM application release management using automated self-testing with evidence-based quality gates across five dimensions.

Ax Yongzhong Xu 3/18/2026

Spectral Edge Dynamics of Training Trajectories: Signal--Noise Geometry Across Scales

Analysis of transformer training dynamics using Spectral Edge Dynamics to measure coherent optimization directions versus stochastic noise.

Ax Lingyun Zhang, Yu Xie, Ping Chen 3/18/2026

IdentityGuard: Context-Aware Restriction and Provenance for Personalized Synthesis

Context-aware safety framework for personalized text-to-image models that prevents misuse without broad concept erasure.

Ax Pengcheng Li, Jie Zhang, Tianwei Zhang, Han Qiu, Zhang kejun, Weiming Zhang, Nenghai Yu, Wenbo Zhou 3/18/2026

State-Dependent Safety Failures in Multi-Turn Language Model Interaction

Analysis of multi-turn safety failures in LLMs through state-space perspective, showing structured contextual evolution enables jailbreaks.

Ax Bingzhou Li, Tao Huang 3/18/2026

DASH: Dynamic Audio-Driven Semantic Chunking for Efficient Omnimodal Token Compression

Token compression method for omnimodal LLMs using dynamic audio-driven semantic chunking to reduce inference costs for audio-visual processing.

Ax Weihao Zhang, Yitong Zhou, Huanyu Qu, Hongyi Li 3/18/2026

Loosely-Structured Software: Engineering Context, Structure, and Evolution Entropy in Runtime-Rewired Multi-Agent Systems

Study on engineering challenges in LLM-based multi-agent systems, addressing context pressure, coordination errors, and system drift at scale.

Ax Ruyi Zhang, Heng Gao, Songlei Jian, Yusong Tan, Haifang Zhou 3/18/2026

BadLLM-TG: A Backdoor Defender powered by LLM Trigger Generator

Defense framework against backdoor attacks in LLMs using trigger generation and inversion to locate and mitigate malicious triggers.

Ax Lars Krupp, Daniel Gei{\ss}ler, Francisco M. Calatrava-Nicolas, Vishal Banwari, Paul Lukowicz, Jakob Karolus 3/18/2026

This Is Taking Too Long -- Investigating Time as a Proxy for Energy Consumption of LLMs

Research on using inference time as a proxy to estimate LLM energy consumption, addressing opacity in API-based model access and environmental impact.

Ax Yulin Peng, Haowen Hou, Xinxin Zhu, Ying Tiffany He, F. Richard Yu 3/18/2026

SEMAG: Self-Evolutionary Multi-Agent Code Generation

SEMAG: self-evolutionary multi-agent code generation framework that decomposes programming tasks into planning, coding, debugging stages with adaptive workflow selection.

Ax Ye Wang, Zixuan Wu, Lifeng Shen, Jiang Xie, Xiaoling Wang, Hong Yu, Guoyin Wang 3/18/2026

Mastering the Minority: An Uncertainty-guided Multi-Expert Framework for Challenging-tailed Sequence Learning

Uncertainty-guided multi-expert framework for imbalanced sequence learning addressing poor expert specialization and prediction conflicts in long-tailed data.

Ax AI Scientists, Xinyi Lin, Danqing Yin, Ying Guo 3/18/2026

LLM-Driven Discovery of High-Entropy Catalysts via Retrieval-Augmented Generation

Retrieval-augmented generation framework using GPT-4 to accelerate CO2 reduction catalyst discovery by exploring chemical spaces and interpreting results.

Ax Artem Sakhno, Ivan Sergeev, Alexey Shestov, Omar Zoloev, Elizaveta Kovtun, Gleb Gusev, Andrey Savchenko, Maksim Makarenko 3/18/2026

Embedding-Aware Feature Discovery: Bridging Latent Representations and Interpretable Features in Event Sequences

Method bridging learned embeddings and handcrafted features in event sequences for financial systems, addressing interpretability and latency constraints in production ML.

Ax Mateusz Dziemian, Maxwell Lin, Xiaohan Fu, Micha Nowak, Nick Winter, Eliot Jones, Andy Zou, Lama Ahmad, Kamalika Chaudhuri, Sahana Chennabasappa, Xander Davies, Lauren Deason, Benjamin L. Edelman, Tanner Emek, Ivan Evtimov, Jim Gust, Maia Hamin, Kat He, Klaudia Krawiecka, Riccardo Patana, Neil Perry, Troy Peterson, Xiangyu Qi, Javier Rando, Zifan Wang, Zihan Wang, Spencer Whitman, Eric Winsor, Arman Zharmagambetov, Matt Fredrikson, Zico Kolter 3/18/2026

How Vulnerable Are AI Agents to Indirect Prompt Injections? Insights from a Large-Scale Public Competition

Large-scale competition analysis revealing LLM agents' vulnerability to indirect prompt injection attacks through adversarial instructions in external content sources.

Ax Lit Sin Tan, Junzhe Chen, Xiaolong Fu, Lichen Ma, Junshi Huang, Jianzhong Shi, Yan Li, Lijie Wen 3/18/2026

Meta-TTRL: A Metacognitive Framework for Self-Improving Test-Time Reinforcement Learning in Unified Multimodal Models

Meta-TTRL: metacognitive test-time reinforcement learning framework for unified multimodal models enabling knowledge accumulation across similar prompts in text-to-image generation.

Ax MiroMind Team, S. Bai, L. Bing, L. Lei, R. Li, X. Li, X. Lin, E. Min, L. Su, B. Wang, L. Wang, L. Wang, S. Wang, X. Wang, Y. Zhang, Z. Zhang, G. Chen, L. Chen, Z. Cheng, Y. Deng, Z. Huang, D. Ng, J. Ni, Q. Ren, X. Tang, B. L. Wang, H. Wang, N. Wang, C. Wei, Q. Wu, J. Xia, Y. Xiao, H. Xu, X. Xu, C. Xue, Z. Yang, Z. Yang, F. Ye, H. Ye, J. Yu, C. Zhang, W. Zhang, H. Zhao, P. Zhu 3/18/2026

MiroThinker-1.7 & H1: Towards Heavy-Duty Research Agents via Verification

MiroThinker-1.7 and H1: research agents with enhanced verification and multi-step reasoning via structured planning and contextual reasoning for long-horizon tasks.

Ax Yihao Zhang, Zeming Wei, Xiaokun Luan, Chengcan Wu, Zhixin Zhang, Jiangrong Wu, Haolin Wu, Huanran Chen, Jun Sun, Meng Sun 3/18/2026

ClawWorm: Self-Propagating Attacks Across LLM Agent Ecosystems

ClawWorm: first documented self-propagating attack across LLM agent ecosystems, demonstrating security vulnerabilities in OpenClaw platform with 40,000+ active instances.

Ax Omkar Patil, Ondrej Biza, Thomas Weng, Karl Schmeckpeper, Wil Thomason, Xiaohan Zhang, Robin Walters, Nakul Gopalan, Sebastian Castro, Eric Rosen 3/18/2026

You've Got a Golden Ticket: Improving Generative Robot Policies With A Single Noise Vector

Technique improving pretrained diffusion/flow-matching robot policies by replacing sampled noise with optimized constant vectors for better downstream reward performance.

Ax Jacob Levy, Tyler Westenbroek, Kevin Huang, Fernando Palafox, Patrick Yin, Shayegan Omidshafiei, Dong-Ki Kim, Abhishek Gupta, David Fridovich-Keil 3/18/2026

Simulation Distillation: Pretraining World Models in Simulation for Rapid Real-World Adaptation

Simulation Distillation method enabling sim-to-real transfer in robotics by pretraining world models in simulation for rapid real-world adaptation with low data.

Ax Yihong Guo, Dongqiangzi Ye, Sijia Chen, Anqi Liu, Xianming Liu 3/18/2026

CorrectionPlanner: Self-Correction Planner with Reinforcement Learning in Autonomous Driving

CorrectionPlanner: autonomous driving planner using reinforcement learning with explicit self-correction mechanism in propose-evaluate-correct loop for unsafe action handling.

Ax Yara Alakeel, Chatrine Qwaider, Hanan Aldarmaki, Sawsan Alqahtani 3/18/2026

Morphemes Without Borders: Evaluating Root-Pattern Morphology in Arabic Tokenizers and LLMs

Evaluation of how LLMs and tokenizers handle Arabic root-pattern morphology, testing whether models capture genuine morphological structure or rely on surface memorization.

Ax Hippolyte Verninas, Caner Korkmaz, Stefanos Zafeiriou, Tolga Birdal, Simone Foti 3/18/2026

Parallelised Differentiable Straightest Geodesics for 3D Meshes

Differentiable framework for computing geodesics on 3D meshes with parallelization support to improve machine learning on non-Euclidean geometric domains.

Ax Hao Wu, Yongheng Zhang, Yuan Gao, Fan Xu, Fan Zhang, Ruobing Xie, Ruijian Gou, Yuxuan Liang, Xiaomeng Huang, Xian Wu 3/18/2026

OMNIFLOW: A Physics-Grounded Multimodal Agent for Generalized Scientific Reasoning

OMNIFLOW: multimodal agent combining LLMs with physics-grounded reasoning to handle spatiotemporal PDE dynamics without domain-specific fine-tuning, reducing non-physical hallucinations.

Ax Samira Abedini, Sina Mavali, Lea Sch\"onherr, Martin Pawelczyk, Rebekka Burkholz 3/18/2026

Don't Trust Stubborn Neighbors: A Security Framework for Agentic Networks

Security framework for LLM-based multi-agent systems addressing manipulation risks from malicious agents in interactive agentic networks through communication channel exploitation.

Ax Thackshanaramana B 3/18/2026

Hypothesis Class Determines Explanation: Why Accurate Models Disagree on Feature Attribution

Large-scale empirical study demonstrating prediction-equivalent models produce substantially different feature attributions across 24 datasets, challenging explainability assumptions.

Ax Nazia Riasat 3/18/2026

When Stability Fails: Hidden Failure Modes Of LLMS in Data-Constrained Scientific Decision-Making

Study showing LLM stability across repeated runs does not guarantee agreement with statistical ground truth in data-constrained scientific decision-making workflows.

Ax Jeremy J Samuelson 3/18/2026

Informationally Compressive Anonymization: Non-Degrading Sensitive Input Protection for Privacy-Preserving Supervised Machine Learning

Informationally Compressive Anonymization (ICA) method for privacy-preserving ML that protects sensitive data without the performance degradation of differential privacy or homomorphic encryption.

Ax Gabrielle Benabdallah 3/18/2026