All Articles
16950 articles total
Adapting Generalist Vehicle Models for High-Speed MPC Across Terrains
OptCar is introduced as a method to adapt generalist vehicle models for high-speed MPC across various terrains using limited real-world data.
Privacy Preserving Recommender Systems Balancing Personalization with Privacy
A framework combining federated learning and differential privacy to create personalized recommender systems that protect user privacy.
Efficient Text-to-Audio Generation via Pruning
Research demonstrates that pruning the U-Net backbone of AudioLDM can reduce parameters by up to 83% while maintaining audio generation quality.
The Refusal Residue: When Probes Catch Alignment Faking and When They Don't
An investigation into 'alignment faking' in LLMs, introducing a measurement framework to detect when models fake compliance.
Evaluation Ability Does Not Imply Optimization Utility: LLM-as-a-Judge Signals in Closed-Loop Table Recognition
Study finds that LLM-as-a-judge signals are often weak and insufficient for closed-loop table recognition optimization.
Learning Engagement Assistant (LEA): Cross-Course Scalability and Classroom Evaluation of an Agentic AI Tutoring System
Evaluation of the Learning Engagement Assistant (LEA), an agentic AI tutoring system, testing its scalability and real-world classroom performance.
Ultrahuman’s former hardware VP raises $5.5M for devices that control AI agents, not just record you
Former Ultrahuman hardware VP raises $5.5M to develop Aina, a device designed to control AI agents rather than just recording data.
Adaptive Filtering of the KV Cache: Diagnosing and Correcting Structural-Role Bias in LLM Inference
Researchers propose a role-conditional allocation method for KV cache eviction in LLMs to correct structural-role bias, improving accuracy for schema-dense inputs like JSON.
Classifying daily activities needs posture, reconstructing them needs motion
A study on daily activity recognition finds that body posture is sufficient for classification, while temporal motion dynamics are essential for movement reconstruction.
Audited Selective Verification for Risk-Controlled N-1 Thermal Contingency Screening under Deployment Shift
Introduces Audited Selective Verification, a risk-budgeted screening layer for energy management systems that reduces full power-flow studies by 29-75% while maintaining safety bounds.
Continuously Evolving Deepfake Detection: An Architecture and Public-Benchmark Evaluation of a Dynamic Detection System
BitMind Forensics (BMF) is a dynamic deepfake detection system trained via adversarial competition that outperforms static open-source and some commercial detectors.
EMAGN: Efficient Multi-Attention Graph Network via Learned Clustering for Scalable Traffic Forecasting
EMAGN proposes an efficient multi-attention graph network that linearizes spatial attention via learned clustering to scale traffic forecasting on limited GPU memory.
Reassessing Muon for Matrix Factorization
A critical reassessment of the Muon optimizer finds it does not consistently outperform AdamW in low-rank matrix factorization, suggesting its benefits may be scale-dependent.
Discourse-Aware Policy Analysis with Argumentation: A Hybrid LLM-Symbolic Framework for Disaster Governance
Apaf is a hybrid LLM-symbolic framework that uses bipolar argumentation to analyze policy discourse and tensions in disaster governance documents.
Deconstructing Actor-Critic: A Large-scale Empirical Study of Design Components for Practitioners
A large-scale empirical study of actor-critic design components for RL provides guidance on robust configurations for real-world control systems like water treatment plants.
Faithful Autoformalization of Natural Language Assertions
Monty is an autoformalization framework that uses conformance and validity scores to synthesize faithful executable assertions from natural language specifications in Java.
OnePlus halts operations in USA and Europe
OnePlus is reportedly halting its business operations in the USA and European markets.
A Beautiful Theory Falls to Ugly Data
A discussion or article exploring the gap between theoretical models and actual empirical data.
Meta now alerts parents if their teen discussed suicide or self-harm with its AI chatbot
Meta is implementing alerts for parents when their teenage children discuss self-harm or suicide with its AI chatbots.
COMPUTER COPS: Inside the big business of selling AI to the police
An investigation into the commercialization and deployment of AI tools within American policing and their impact on legal processes.
