All Articles
16570 articles total
Binding Drift in Multi-Step Tool-Augmented Agents
Analysis of 'binding drift' in tool-augmented agents, showing how entities can shift during multi-step workflows and proposing a re-verifier fix.
Cost Accounting for Reactive Computational Graphs: Exhaustive Sweeps, Sequential Mutation, and the Backward-Locality Gap
A cost accounting analysis of reactive computational graphs for neural network interventions, validated using the NeuroDSL engine in Julia.
Codeberg: ToU extension to prohibit LLM-extrusions
Codeberg updates its Terms of Use to prohibit the use of its platform for LLM training data extraction (scraping).
Glow emerges from stealth at $1.2B valuation to challenge endpoint security in the AI era
Glow launches with a $1.2B valuation to provide endpoint security tailored for AI agents and developer tools in enterprise environments.
FedCC: A Low-Resource Federated Adaptation of Foundation Models for Robust Corpus Callosum localization in Fetal Ultrasound Images
FedCC is a federated learning framework using DINOv2 and LoRA for privacy-preserving corpus callosum localization in fetal ultrasound images.
Compressing What Matters: Neuron Importance Meets Data-Aware Low Rank Approximation for Language Model Compression
Researchers propose a new method for LLM compression combining neuron importance and data-aware low-rank approximation with dynamic rate allocation.
Edge-Efficient Transformer for End-to-End RF Spectrum Monitoring
E-SpecFormer is an edge-efficient Transformer for RF spectrum monitoring featuring a new Softmax-free attention mechanism called LiTAN.
Preference-Conditioned Multi-Objective Reinforcement Learning for Runtime-Tunable Transit Signal Priority
A preference-conditioned multi-objective RL controller is introduced for runtime-tunable transit signal priority to balance bus delay and traffic flow.
BearingNAS: Obtaining In-Sensor Intelligent Fault Diagnosis Systems for Bearings Using a Laptop
BearingNAS provides a hardware-aware NAS framework for deploying intelligent fault diagnosis systems directly onto low-resource sensor dies.
Towards Principled Continual Anomaly Detection: A Systematic Framework and Benchmark Scenarios
A systematic framework for reproducible benchmark scenario design is proposed for continual anomaly detection in tabular cybersecurity datasets.
SechKAN: Kolmogorov-Arnold Networks with Hyperbolic Secant Functions
SechKAN introduces a Kolmogorov-Arnold Network architecture using hyperbolic secant functions for improved function fitting and image classification.
Dual-domain fused LSTM modeling for efficient time-dependent reliability analysis
The DDF-LSTM model uses a dual-domain fused architecture to efficiently analyze time-dependent reliability and failure probabilities of engineering systems.
Codeberg bans vibe coded projects
Codeberg has reportedly banned projects that are 'vibe coded', though the specific criteria for this ban are discussed in the comments.
MechAInistic: An LLM-guided Multi-Agent System for Reasoning over Genome-Scale Constraint-Based Metabolic Models
MechAInistic is a multi-agent LLM system designed to transform natural-language biological questions into executable workflows for metabolic model reasoning.
Assistant or Actor? Student Trust, Control, and Delegation Regret When Using a General-Purpose AI Agent
Researchers study 'delegation regret' in AI agents, finding that users demand more control for irreversible and externally visible actions regardless of the stakes.
The Economics of Autonomy: Real-Time Risk Indexing for Insurable AI-Driven 6G Systems
The GIRAF framework introduces Governance-as-Code for real-time risk quantification and trust modulation in agentic 6G wireless systems.
Market Strategy Evaluation for Prosumers in Local Electricity Markets
A study evaluates agent-based simulation platforms for prosumers in local electricity markets, showing that market-adaptive pricing strategies increase community financial gain.
Domain Design for the Cops and Robbers Problem
This research casts the graph-theory 'Cops and Robbers' problem as a non-deterministic planning problem and uses state-of-the-art planners to determine if a graph is k-copwin.
A Calculus of Discernment: Decision-Relevant Insight, Sequence Value, and Forgetting as Higher-Order Learning
The paper proposes APOHA, a theory where 'forgetting' is treated as a higher-order learning operator to maximize decision-relevance and reduce cumulative regret.
FALCON-Discover: Discovering Concentrated False-Confidence Regions for Calibration
FALCON-Discover is a model-agnostic framework that identifies local regions of false-confidence in ML predictions to improve calibration.