AI/ML arXiv cs.AI

Server-side Anti-cheat in FPS games for Aimbot detection using Deep learning and Machine learning

Development of YAACS, a server-side aimbot detection system for FPS games using Stacked LSTM and deep learning to minimize false positives.

AI/ML arXiv cs.AI

Nemotron-Labs-3-Puzzle-75B-A9B: Compressing Hybrid MoE LLMs

Introduction of Puzzle-75B-A9B, a compressed variant of Nemotron-3-Super that significantly increases server throughput and concurrency using a hybrid MoE pruning approach.

AI/ML arXiv cs.AI

Decentralized Aggregation of LLM Predictions via Wagering Mechanisms

Proposal of WALLA, a decentralized wagering mechanism for aggregating LLM predictions that ensures incentive compatibility and advantage-aligned weights.

AI/ML arXiv cs.AI

MechMath Agent Team: LLM Driven Agents for Mathematical Research

The MechMath Agent Team (MMAT) is a multi-agent LLM system designed to co-pilot mathematical research and produce formally certified proofs.

AI/ML arXiv cs.AI

LLM-as-a-Tutor: Policy-Aware Prompt Adaptation for Non-Verifiable RL

LLM-as-a-Tutor is a framework for non-verifiable RL that adapts prompt difficulty in real-time based on the evolving capabilities of the policy.

AI/ML arXiv cs.AI

Agent Step Value: State-Transition Measurement with State-Grounded LLM Evaluators

Introduction of Agent Step Value (ASV), a framework for measuring the impact of individual agent actions on state transitions using LLM evaluators.

AI/ML arXiv cs.AI

ResearchStudio-Idea: An Evidence-Grounded Research-Ideation Skill Suite from ML Conference Outcomes

ResearchStudio-Idea is a skill suite for ML research ideation that uses evidence grounding and a corpus of conference papers to generate traceable research proposals.

AI/ML arXiv cs.AI

PLACEMEM: Toward a Compute-Aware Memory Plane for Lifelong Agents

The authors propose PLACEMEM, a compute-aware memory plane for lifelong agents that uses versioned capsules to manage persistence, evolution, and correction of agent memories.

AI/ML arXiv cs.AI

Forethought: Verifiable Reasoning from Neurosymbolic Primitive Programming

Forethought is a neurosymbolic reasoning system that treats reasoning as explicit, verifiable programs, allowing small models to match the capabilities of frontier models.

AI/ML arXiv cs.AI

Language models guide symbolic equation discovery by controlling search

The paper introduces LLM-PySR, a framework where language models control the search process of symbolic regression to discover scientific equations more effectively.

AI/ML arXiv cs.AI

A Clustering-Based Framework for Identifying Suspicious Trading Patterns in Capital Market

A study implements an unsupervised fraud-detection toolkit using K-Means++ clustering to identify suspicious trading patterns in capital markets.

AI/ML arXiv cs.AI

Agentic IoT: Architectures, Applications, and Challenges Toward the Internet of Agents

This paper outlines the concept of Agentic IoT, moving from passive data collection to distributed cognitive agent ecosystems across the device-to-cloud continuum.

AI/ML arXiv cs.AI

Unsupervised Features Mining via Activation Geometry

The Mining via Activation Geometry (MAG) framework extracts unsupervised reasoning features from LLM activations to enable vector steering and improved data selection.

AI/ML arXiv cs.AI

Biological Motifs for Agentic Control

This research maps biological control motifs to agentic software design patterns using a typed interface correspondence to improve reliability and security in LLM agents.

AI/ML arXiv cs.AI

Progress- and Reliability-Oriented Group Policy Optimization for Agentic Reinforcement Learning

ProGPO is introduced as a learned-critic-free method for context-consistent step-level reinforcement learning in agentic tasks.

AI/ML arXiv cs.AI

Shortcut Learning in Legal Judgment Prediction: Empirical Evidence from the UK Employment Tribunal

An empirical study reveals that legal judgment prediction models often rely on 'shortcut learning' from outcome-revealing cues in post-hoc judicial texts.

Cybersecurity arXiv cs.AI

Agentic SABRE: An Uncertainty-Aware Neuro-Symbolic Multi-Agent Framework for Adaptive Ransomware Detection

Agentic SABRE is a neuro-symbolic multi-agent framework for adaptive ransomware detection that combines semantic evidence with behavioral telemetry and uncertainty quantification.

AI/ML arXiv cs.AI

Agent Reinforcement Learning via Pivotal-Aware Self-Feedback Retry

PivoARL is a new self-feedback retry framework for LLM agents that identifies pivotal errors to reduce redundant interactions and improve learning efficiency.

AI/ML arXiv cs.AI

Explainable Reinforcement Learning for Adaptive Traffic Signal Control

Researchers propose an explainable RL framework for traffic signal control that uses entity-centric architecture and attention networks for transparent decision-making.

AI/ML arXiv cs.AI

Can Conversational Temporal Dynamics Improve Depression Detection in Dyads? A Preliminary Investigation in Multi-Modality Perspectives

A study demonstrates that conversational temporal dynamics (turn-pair timing) can serve as a lightweight and interpretable modality for improving depression detection in dyads.