All Articles
17680 articles total
Symbolic Mechanistic Data Attribution: Tracing Training Influence to Learned Behavioral Policies
Introduces Symbolic Mechanistic Data Attribution (SMDA), a tool to trace high-level model behaviors back to specific training examples using SAE features.
Anomaly Factory 3D: A Modular Framework for Diverse Pseudo-Anomaly Synthesis in Unsupervised 3D Anomaly Detection
Presents Anomaly Factory 3D (AF3AD), a modular framework for synthesizing pseudo-anomalies to improve unsupervised 3D anomaly detection.
A Multi-Dataset Benchmark for Evaluating LLM Agents in Microservice Failure Diagnosis
Introduces AIOps2025 and RCA100, benchmarks for evaluating LLM agents in diagnosing microservice failures based on reasoning processes.
Behavior Uncloning: Distilling Mode Redirection into Policy Weights without Inference-Time Steering
Proposes MoRE (Mode Redirection), a method to distill desired behavior modes into policy weights to remove unsafe behaviors without inference-time overhead.
Trump drops restrictions on Anthropic’s Mythos and Fable models
Anthropic is restoring access to its Fable models following the removal of certain restrictions.
Wayve launches $85M employee tender offer at $8.5B valuation
AI startup Wayve launches an $85M employee tender offer at an $8.5B valuation to attract and retain talent.
Statistically Indistinguishable, Operationally Distinct: A Formal Barrier for Tabular Foundation Models
Researchers propose the Operational Turing Test (OTT) to demonstrate that tabular foundation models cannot reason about running systems without access to governing rules.
Priced Motion Through Optimal Faces: A Normal-Fan Geometry for Non-Stationary Adversarial MDPs
The paper introduces a normal-fan geometry for non-stationary adversarial MDPs to better analyze the cost of non-stationarity in reward sequences.
Unified Complex-valued Neural Network: A Magnitude-Phase Computational Model for Event-Driven Neuromorphic Learning
A new Unified Complex-valued Neuron (UCN) model is introduced to integrate continuous activation and phase-driven event generation for neuromorphic learning.
BTI-Net: Bidirectional Decoder-Level Task Interaction via Uncertainty-Aware Gating for Multi-Task Medical Image Analysis
BTI-Net is introduced for multi-task medical image analysis, using bidirectional decoder-level interaction and uncertainty-aware gating to improve segmentation and classification.
A Deep Multiscale Neural Network for Accurate Neurological Disorder Detection from MRI Scans and Real-Time Web Deployment
The Enhanced Neurological Disorder Detection Network (End-Net) is proposed for multi-class MRI classification using multiscale features and WGAN-GP for class imbalance.
LLM Semantic Signaling Game and Mechanism Design: Systematic Blindness, Awareness Shaping, and Mindset Dynamics
Researchers develop a semantic signaling game framework to analyze strategic communication, deception, and awareness shaping in LLM-mediated interactions.
When Stopping Fails: Rethinking Minimal Risk Conditions through Human-Interactive Autonomous Driving for Safe Transportation Systems
An analysis of autonomous vehicle (AV) failure incidents suggests that current 'stopping' fallback behaviors are insufficient and need to be replaced with human-interactive autonomy.
Knowing in Advance When an Evolutionary Outer Loop Will Not Help: A Pre-Registered Cheap-Baseline Screening Rule
A pre-registered screening rule is introduced to determine if an evolutionary outer loop for neural network parameters is worth the computational expense compared to a cheap single-shot alternative.
Efficient Spatio-Temporal Grounding with Multimodal Large Models via Second-Level Tracking and RL Verification
A new pipeline for spatio-temporal grounding in long videos using second-level tracking and RL verification to balance efficiency and localization quality.
How to Leverage Synthetic Speech for LLM-Based ASR Systems?
Research on leveraging synthetic speech for ASR training, demonstrating that room impulse responses (RIRs) can narrow the gap between synthetic and real audio data.
The strength of clinical evidence is recoverable from language model representations but not from their stated grades
Study finding that LLMs possess recoverable evidence-strength signals in their representations but fail to accurately state those grades explicitly.
Metric Aggregation Divergence: A Hidden Validity Threat in Agent-Based Policy Optimization and a Contractual Remedy
Identification of 'Metric Aggregation Divergence' in agent-based policy optimization and the introduction of 'metric contracts' to ensure pipeline consistency.
Flow Matching in Feature Space for Stochastic World Modeling
Introduction of FlowWM, a stochastic world model that performs flow matching within high-dimensional pretrained feature spaces for better perception and robustness.
Fairness Attacks on Recommender Systems
A novel reinforcement learning-based attack method designed to exacerbate unfairness and bias within recommender systems.