Ax Ilya E. Monosov 23d ago

A toy framework for single and multi-agent human-AI curiosity ecosystems

Toy framework modeling curiosity as ecosystem with adjustable weights for uncertainty reduction, costs, and delayed returns.

Ax Yijun Zhang, Fan Xu, Jiaxin Ding, Yule Xie, Shiqing Gao, Xin Ding, Haoxiang Zhang, Luoyi Fu, Xinbing Wang 23d ago

Information Gain-based Rollout Policy Optimization: An Adaptive Tree-Structured Rollout Approach for Multi-Turn LLM Agents

Information gain-based adaptive tree-structured rollout optimization for multi-turn LLM agents on long-horizon tasks.

Ax Ziting Wang, Yin Li, Zuhao Yang, Xiuchang Li, Jiale Bai, Gao Cong 23d ago

Demonstrating TOFFEE: A Learned System for Synthesizing Data Agent Trajectories at Scale

TOFFEE system for synthesizing high-quality data agent trajectories at scale for heterogeneous enterprise environments.

Ax Heting Mao 23d ago

From Application-Layer Simulation to Native Meta-Architecture: Structural Tension as an Endogenous Driver for Heterogeneous AI Evolution

Theoretical framework for embedding application-layer cognitive protocols into native LLM meta-architecture via structural mechanisms.

Ax Yanping Chen, Weijie Shi, Wen Yang, Jiajie Xu 23d ago

Task Decomposition-Guided Reranking for Adaptive Agent Skill Retrieval

Task decomposition-guided reranking method for improved skill selection in agent systems with large skill libraries.

Ax He Liu, Changtao Miao, Xinjie Yang, Tianle Song, Yin Wu, Junchi Chen, Bintao He, Xinyuan Zhang, Bo Zhang, Shi Yan, Wei Lu, Wei Wang, Danyang Xu, Jiansheng Cai, Zhe Li 23d ago

DT-Guard: Intent-Driven Reasoning-Active Training for Reasoning-Free LLM Safety Guardrail

LLM safety guardrail using intent-driven reasoning to balance efficiency and robustness against complex safety risks.

Ax Franz Motzkus, Sebastian Bernhard 23d ago

Driving the Wrong Way: Leveraging Interpretability in End2End Autonomous Driving Models

Post-hoc interpretability module using dictionary learning to decompose autonomous driving model behavior into semantic concepts.

Ax Xiachong Lin, Du Yin, Arian Prabowo, Hao Xue, Wen Hu, Imran Razzak, Matthew Amos, Sam Behrens, Flora D. Salim 23d ago

TopoBrick: Agentic Topology Sampling of Exogenous Variables for Zero-Shot Building IoT Forecasting

Training-free framework using agentic topology sampling and knowledge graphs for zero-shot IoT forecasting in building sensor networks.

Ax Marcelo Arenas, Pablo Barcel\'o, Diego Bustamante, Jose Caraball, Mar\'ia Alejandra Schild, Bernardo Subercaseaux 23d ago

ExplAIner: A Declarative Query Language for Explaining Classification Models

ExplAIner declarative query language for specifying and analyzing model explanation methods uniformly across classification models.

Ax Yufan Wang, Anit Kumar Sahu, Yan Fei Ng, Daniel Kang, Shayan Vassef, Soorya Ram Shimgekar, Koustuv Saha, Piyum Zonooz, Navin Kumar, Chee Leong Cheng, Li Yan Khor 23d ago

Finding H. pylori in the Fine Print: Evidence-Linked Multi-Agent Case Finding from Gastric Biopsy Reports

Multi-agent LLM system extracts H. pylori infection evidence from medical reports across heterogeneous structured and free-text fields.

Ax Jihao Liu, Guoxiong Gao, Zeming Sun, Bin Wu, Shurui Liu, Jiedong Jiang, Haocheng Ju, Leheng Chen, Ronnie Cheng, Xiping Zhang, Bin Dong 23d ago

Danus: Orchestrating Mathematical Reasoning Agents with Fact-Graph Memory

Danus system orchestrates LLM-based mathematical reasoning agents using fact-graph memory for research-level problem solving.

Ax Marcos Eduardo Cruz Victorio, Karl Mason 23d ago

Multi-Agent Deep Reinforcement Learning for Multi Objective Battery Management in Dairy Farms

Multi-agent deep reinforcement learning system for battery management optimization in dairy farm renewable energy integration.

Ax Kai Ruan, Zihe Huang, Ziqi Zhou, Qianshan Wei, Xuan Wang, Hao Sun 23d ago

Doomed from the Start: Early Abort of LLM Agent Episodes via a Recall-Controlled Probe Cascade

Method to predict LLM agent failures early using probe cascades on hidden activations to avoid wasted compute in multi-step tasks.

Ax Qian Sun, Yong-Ming Tian, Jia-Wei Huang, Cheng Feng, Shao-Qun Zhang 23d ago

RMISC: A Large-scale Real-world Multivariate Corpus for Time Series Foundation Models

Large-scale multivariate time series dataset for training foundation models on real-world data rather than synthetic datasets.

Ax Chase McDonald, Nathan Tsang, Wesley N. Kerr 23d ago

FootsiesGym: A Fighting Game Benchmark for Two-Player Zero-Sum Imperfect-Information Games

FootsiesGym: open-source benchmark environment for two-player zero-sum imperfect-information games with vectorized simulator for efficient training.

Ax Anna C\'ordoba, Adam Puente Tercero, Nerea Angulo Hijo, Mar Linares Tercero, Julia Barrientos, Ainhoa Miranda, Jes\'us Olivera 23d ago

FreqDepthKV: Frequency-Guided Depth Sharing for Robust KV Cache Compression in Long-Context LLM Inference

FreqDepthKV: frequency-guided cache compression factorizing KV states into shared low-frequency and sparse high-frequency components for long-context inference.

Ax Han-Jun Ko, Jr-Jen Chen, Haobo Yuan, Hsin-Ying Lee, Tiancheng Shen, Ming-Hsuan Yang, Yu-Chiang Frank Wang 23d ago

Bridging Physical Reasoning and Task Generalization via Visual Action Outcome Reasoning Alignment

VAORA: reward design for vision-language models to improve physical reasoning and align internal reasoning with actions in interactive tasks.

Ax Anna Cordoba, Adam Puente Tercero, Nerea Angulo Hijo, Mar Linares Tercero, Julia Barrientos, Ainhoa Miranda, Jesus Olivera 23d ago

DepthWeave-KV: Token-Adaptive Cross-Layer Residual Factorization for Long-Context KV Cache Compression

DepthWeave-KV: token-adaptive KV cache compression across transformer layers using residual factorization for long-context inference.

Ax Aparna Madva, Sharath Srivatsa, Srinath Srinivasa, Tulika Saha 23d ago

Rethinking Indic AI from a Lens of Cultural Heritage Preservation

Examines AI's impact on linguistic and cultural preservation in Indian subcontinent, viewing AI as double-edged for inclusion vs. homogenization.

Ax Igor Santos-Grueiro 23d ago

Lingering Authority: Revocable Resource-and-Effect Capabilities for Coding Agents

PORTICO: reference monitor for revocable capabilities in coding agents, compiling task contracts into initial capabilities and closure predicates.

Ax James Rhodes, George Kang 23d ago

Proof of Execution: Runtime Verification for Governed AI Agent Actions

Runtime verification framework for AI agent actions: formalizes authorization, tamper-evidence, and deterministic reconstruction of agent trajectories.

Ax Nikita Agrawal, Ruben Mayer 23d ago

Benchmarking KV-Cache Optimizations across Task Quality and System Performance for Long-Context Serving

Benchmark comparing KV-cache optimization techniques (quantization, pruning, merging) across models, tasks, and serving stacks for long-context LLM inference.

Ax Fan Huang 23d ago

Catalyst Papers in Artificial Intelligence Research: A Landscape on ICLR from 2017 to 2025

Analyzes ICLR papers 2017-2025 to identify trajectory-changing methodological contributions using public reviewer scores and decisions.

Ax Vasudha Varadarajan, Akhila Yerukola, Mona T. Diab, Maarten Sap 23d ago

CCBENCH: Assessing LLM Cultural Competence via Implicitly Signaled Norms using Health Queries

CCBENCH: benchmark evaluating LLM cultural competency by assessing adaptation to implicitly signaled cultural norms in health queries.

Ax Yukyeong Song, Seoyeon Choi, Jinhee Kim, Yeongje Kim, Lauren Weisberg, Jewoong Moon 23d ago

A Guiding Framework for K-12 Teachers in Creating AI-powered Learning Technologies through Vibe Coding

GAIDE framework enabling K-12 teachers to create AI-powered learning tech via vibe coding with LLMs, supporting teachers as designers.

Ax Muntasir Hoq, Griffin Pitts, Zhangqi Duan, Arun Balajiee Lekshmi Narayanan, Mohammad Hassany, Andrew Lan, Peter Brusilovsky, Bita Akram 23d ago

Automated Recommendation of Programming Learning Content Using Pattern-based Knowledge Components

Uses pattern-based knowledge components to automatically recommend programming learning resources, reducing need for expert curation.

Ax Dexter Hadley 23d ago

CANONIC: Governance Is Compilation

CANONIC: system that applies compiler-like governance to LLM-generated content, using formal grammars to admit/reject artifacts into evidence ledgers at scale.

Ax Leandro D. Carneiro, Andre L. S. Meirelles, Juliano de A. Gomes, Rafael C. A. Cabral 23d ago

CHARLIE: An On-Premise Multi-Agent Retrieval-Augmented Generation System for Evidential Reasoning in Forensic Science

CHARLIE: on-premise multi-agent RAG system for structured evidential reasoning in digital forensics with traceability and compliance requirements.

Ax Xue Li, Yiming Gai 23d ago

Modality Relevance is not Modality Utility: Post-hoc Selective Modality Escalation for Cost-Aware Multimodal RAG

Multimodal RAG system using post-hoc selective modality escalation to balance cost and utility by adaptively invoking vision-language models.

Ax Lorenzo Molfetta, Giacomo Frisoni, Nicol\`o Monaldini, Gianluca Moro 23d ago

PORTS: Preference-Optimized Retrievers for Tool Selection with Large Language Models

PORTS: preference-optimized retrieval method for tool selection in LLM agents. Aligns retrievers with tool-calling LLMs through joint training.

Ax Nishan Pantha, Pranath Reddy Kumbam, Sajil Awale, Pushwitha Krishnappa, Muthukumaran Ramasubramanian, Nidhi Jha, Emily Foshee, Ankur Kumar, Rachel Slank, Ashkbiz Danehkar, Rahul Ramachandran 23d ago

Scientific Code Search at Scale: A Multi-Domain Dataset and Benchmark

Multi-domain scientific code search benchmark with 5,264 curated repositories across scientific computing domains to evaluate code discovery tools.

Ax Haiwen Yi, Xinyuan Song 23d ago

Learning to Control LLM Agent Harnesses with Offline Reinforcement Learning

Offline RL approach to learn control policies for LLM agent execution harnesses. Formalizes harness operation as finite-horizon MDP with frozen LLM executor.

Ax Edwin H. Wintermute, Harmon Bhasin, Christina M. Agapakis, Dianzhuo Wang, Evan Seeyave, Arjun Banerjee, Daniel Fulop, Matthew C. Watson, Adam J. Meyer, Sandrine Boissel, Jens H. Kuhn, Rishi Jain, Noah D. Taylor, Helena Shomar, Patrick M. Boyle, Kenny Workman 23d ago

Evaluating calibrated refusal and safe usefulness in dual-use biology settings

BioSecBench-Refusal: benchmark for evaluating AI agent safety in biology tasks, pairing 61 routine tasks with 46 red-team scenarios for biosecurity assessment.

Ax Hairui Zhu, Yiying Yang, Tengjin Weng, Ziyu Lu, Xiao Yao, Xiaoyang Ye, Lin Ma, Wenhao Jiang 23d ago

CanvasAgent: Enabling Complex Image Creation and Editing via Visual Tool Orchestration

CanvasAgent: multi-modal agent orchestrating visual tools for complex image creation and editing through synthesis, segmentation, and composition.

Ax Bo Huang, Fengxiang Li, Hao Xu, Haoyang Huang, Hongyi Fu, Jinhua Hao, Kun Yuan, Minglei Zhang, Pengcheng Xu, Shiyang Liu, Wenhao Zhuang, Yuze Shi, Zongxian Feng, Chao Wang, Cheng He, Chongling Rao, Deyu Cao, Fan Yang, Gang Xiong, Haochen Liu, Jiabao Li, Jian Liang, Jinghui Jia, Jingwen Chang, Jun Du, Junyu Shi, Min Li, Mingqi Wu, Qiang Gao, Shangpeng Yan, Shaotong Qi, Shu Xu, Shuo Zhou, Tiankuo Xu, Tong Zheng, Weilun Zhao, Xiancheng Meng, Xianda Sun, Xiaoyu Jiang, Xunhao Jia, Yao Xia, Yimeng Xu, Yinghan Cui, Yingpeng Chen, Yiwen Ning, Yong Wang, Yuxuan Sun, Zhongsheng Liu, Ming Sun, Cheng Luo, Chen Yang, Han Li, Kun Gai 23d ago

KAT-Coder-V2.5 Technical Report

KAT-Coder-V2.5: agentic coding model trained autonomously in executable repositories. Introduces AutoBuilder for sandbox reconstruction and end-to-end agentic post-training framework.

Ax Guanyu Cai, Ruiming Tian, Lang Yang, Zhouhong Ren, Jinliang Yuan, Lingkun Li, Jiliang Wang 23d ago

Is Your NPU Ready for LLMs? Dissecting the Hidden Efficiency Bottlenecks in Mobile LLM Inference

Comprehensive measurement study of mobile LLM inference across frameworks (llama.cpp, GENIE) and hardware backends (CPU, GPU, NPU). Introduces PowerBench profiling tool to identify efficiency bottlenecks.

Ax Lars Benedikt Kaesberg 23d ago

Decision Protocols in Multi-Agent Large Language Model Conversations

Study of decision protocols in multi-agent LLM conversations for task performance improvement through specialized agent distribution and discussion mechanisms.

Ax V\'aclav Jane\v{c}ek, Thomas Melham 23d ago

Privilege and confidentiality in generative AI workflows

Analysis of confidentiality and legal privilege risks in generative AI systems across training, context windows, and RAG-based knowledge databases.

Ax Konstantin Berlin 23d ago

Full-range Binary Classifier Calibration for Stable Model Updates in Production

Calibration method for binary classifiers in adversarial environments maintaining consistent false-positive rates across predictions during model retraining.

Ax Zhaoyu Bai, Jiaqi Cai 23d ago

PatchOptic for Shared-State LLM Workflows with Projected Views and Verified Structured Updates

PatchOptic system for managing shared state in agentic LLM workflows using projected views and verified structured updates instead of grep-like searches.

Ax Paul K. Mandal, Pavan Reddy, Tristan Malatynski 23d ago

Statistical Adversaries: Natural Backdoor-like Features in Vision Datasets

Analysis of naturally occurring statistical signals in vision datasets that behave like backdoor triggers, studying Imagenet patterns linked to labels.

Ax Sai Varun Kodathala 23d ago

aiAuthZ: Off-Host, Identity-Bound Authorization for AI Agents

aiAuthZ security framework for identity-bound authorization in AI agents, evaluating LLM vulnerability to tool-call forgery attacks across 15 models.

Ax Muhammad Zain Amin, Kibele Sebnem Yildirim 23d ago

Self-Review Reinforcement Learning (SRRL) with Cross-Episode Memory and Policy Distillation

Self-Review Reinforcement Learning method for LLMs using cross-episode memory and policy distillation to handle sparse/delayed environmental feedback.

Ax Yibo Hu, Jiaming Qu 23d ago

Most LLM Conformity Needs No Speaker: Measuring the Speaker-Free Floor in Peer-Pressure Benchmarks

Study showing LLM conformity to peer responses is largely confounded by repeated answers independent of speaker presence in benchmark tasks.

Ax Haonan Huang 23d ago

The yes-no bias of large language models reflects answer order and wording, not shifts in moral judgment

Analysis of LLM yes-no bias in binary judgments, isolating effects of answer order and wording from moral judgment shifts using psychometric methods.

Ax Sadia Kamal, Arefa Patwary, Anthony Marchiafava, Atriya Sen, Sagnik Ray Choudhury 23d ago

Prompt Robustness Is Task-Dependent: Comparing Objective and Belief-Style Questions in LLM Evaluation

Evaluation study comparing prompt robustness between objective and subjective LLM tasks, showing different sensitivity patterns across model families.

Ax Gregor Kobsik, Tim Elsner, Leif Kobbelt 23d ago

Harnessing Generative Image Models for Training-Free Primitive Shape Abstraction

Training-free approach using generative image models for 3D primitive shape abstraction without task-specific fine-tuning.

Ax Abhash Shrestha, Subigya Gautam, Anu Sapkota, Sanju Tiwari, Tek Raj Chhetri 23d ago

Whose fairness? Structural concentration in AI bias research

Analysis of structural concentration in AI bias research community, examining whose fairness definitions and debiasing frameworks are produced.

Ax Archie Chaudhury 23d ago

ResonatorLM: Causal Resonant Field Mixing for Efficient Long-Context Language Modelin

ResonatorLM architecture for efficient long-context language modeling using causal resonant field mixing as alternative to transformers.

Ax Giulia Lanzillotta, Mandana Samiei, Doina Precup, Razvan Pascanu, Claire Vernade 23d ago

To Retain or to Adapt? Generalizing Continual Learning

Continual learning research challenging retention-centered assumptions and prioritizing real-time adaptation in non-stationary environments.