All Articles
19262 articles total
Learning Coordinated Preference for Multi-Objective Multi-Agent Reinforcement Learning
The PCMA framework is proposed for multi-objective multi-agent reinforcement learning to coordinate agent-specific preferences and improve team performance.
ClinHallu: A Benchmark for Diagnosing Stage-Wise Hallucinations in Medical MLLM Reasoning
ClinHallu is a benchmark for diagnosing stage-wise hallucinations in medical multimodal LLMs, providing a structured reasoning trace for better diagnosis.
Learning optimal policies from event logs through reinforcement learning: a comparison of deep and MDP-based approaches
A comparison of deep RL and MDP-based approaches for learning optimal behavioral policies from historical event logs in process mining.
ANSR-DT: A Neuro-Symbolic Framework for Adaptive and Explainable Digital Twins
ANSR-DT is a neuro-symbolic framework for industrial digital twins that combines CNN-LSTM, Prolog-based reasoning, and PPO for adaptive and explainable decisions.
LLM-Powered AI Agent Systems and Their Applications in Industry
A comprehensive review of the evolution, architectures, and industrial applications of LLM-powered AI agent systems, including current challenges and solutions.
An O(x)Caml book that runs
A discussion or announcement regarding an OCaml book that is interactive or executable.
When Errors Become Narratives: A Longitudinal Taxonomy of Silent Failures in a Production LLM Agent Runtime
A study on 'silent failures' in production LLM agent runtimes, identifying a 'fail-plausible' pattern where LLMs hallucinate narratives instead of reporting errors.
AudioDER: A Deduplication-Enhanced Reasoning Dataset for Post-Training Large Audio-Language Models
Introduction of AudioDER, a deduplicated reasoning-oriented dataset designed to improve the post-training of Large Audio-Language Models (LALMs).
Regulating the Machine Contributor: Governance and Policy Alignment in Open Source
An analysis of AI-driven contributions to open-source projects, proposing a governance framework to align autonomous agents with existing community policies.
A Comparative Study of Deep Learning Architectures for Multi-Horizon Behavioural Forecasting for Mobile Health
A comparative study of deep learning architectures for behavioral forecasting in mobile health, finding that PatchTST and TimesFM are particularly effective.
Expert-Driven Survival Machines: Improving Stratification and Interpretability in Multiple Clinical Cohorts
Proposal of AdaCSM, a Mixture-of-Experts (MoE) framework for adaptive deep clustering in clinical survival prediction to improve patient stratification.
Moonlight in Latent Space: Chirality and Structural Correspondence Between Beethoven's Op. 27 No. 2 and Machine Learning Mechanisms
A computational analysis exploring structural correspondences between Beethoven's Moonlight Sonata and various machine learning architectures.
When Good Verifiers Go Bad: Self-Improving VLMs Can Regress on New Tasks
Research showing that self-improving VLMs can regress when using task-specific verifiers that are inaccurate for the target task, despite decreasing training loss.
From Self-Supervised Speech Models to Mixture-of-Experts for Robust Anti-Spoofing
A method for converting self-supervised speech models into Mixture-of-Experts (MoE) architectures to enhance robustness in anti-spoofing detection.
Listening with Attention: Entropy-Guided Explainability for Transformer-Based Audio Models
Introduction of LEAF-X, an entropy-guided XAI framework that provides more faithful and sparse explanations for Transformer-based ASR models like Whisper.
Banned Book Library in a Wi-Fi Smart Light Bulb
A creative project involving the installation of a banned book library within a Wi-Fi smart light bulb.
San Francisco Weighs PG&E Takeover Amid Soaring Utility Costs
San Francisco is considering taking over PG&E due to rising utility costs.
Securing the Future of IoMT in the Post-Quantum Era: An Edge-Native Federated Learning Approach
A study proposing a Kubernetes-based framework for securing Internet of Medical Things (IoMT) devices using Post-Quantum Cryptography and Federated Learning.
From Shield to Target: Denial-of-Service Attacks on LLM-Based Agent Guardrails
Research demonstrating how LLM-based agent guardrails can be targeted by denial-of-service attacks through crafted reasoning loops.
TRACE: Trajectory-Routed Causal Memory for Delayed-Evidence Visuomotor Imitation
Introduction of TRACE, a memory framework for visuomotor imitation in robotics that uses path signatures to handle delayed evidence.