All Articles
16920 articles total
Human4K: A Large-Scale 4K Multi-View Mocap Dataset for Whole-Body 3D Human Reconstruction
Introduction of Human4K, a large-scale 4K multi-view dataset designed for high-precision 3D human reconstruction and motion capture.
Beyond Color Geometry: Evaluating Human-Like Color Representations in Vision Models
Research evaluating how vision models represent color compared to human perception, finding that Masked Autoencoders show the strongest alignment with human-like grounding.
Barnamala: Parameter-Efficient Handwritten Devanagari Recognition at Benchmark Saturation
Development of Barnamala, a highly compact convolutional network for Devanagari recognition that achieves state-of-the-art accuracy with significantly fewer parameters.
Social Simulations: from Agent-Based Modeling to Digital Twins
A conceptual overview of the evolution of social simulations, moving from agent-based modeling to AI-enhanced digital twins.
Groc-PO: Grounded Context Preference Optimization for Truthful Multimodal LLMs
Proposed Groc-PO, a grounded preference optimization framework to reduce hallucinations and improve truthfulness in multimodal LLMs.
How Agents Ask for Permission: User Permissions for AI Agents, from Interfaces to Enforcement
A survey of user-level permission systems for AI agents, analyzing 21 proposals to address security risks like prompt injection and unauthorized actions.
San Francisco mayor pushes for tougher rules after the Waymo traffic fiasco
San Francisco Mayor Daniel Lurie is calling for stricter regulations on robotaxi operators like Waymo following a major gridlock incident.
Agile perceptive multi-skill locomotion for quadrupedal robots in the wild
Researchers introduce APT-RL, a framework using Action Pretrained Transformers and Reinforcement Learning to enable quadrupedal robots to navigate complex terrains at high speeds.
From Prediction to Collaboration: Interactive Symbolic Music Analysis
A new unified framework for symbolic Roman-numeral music analysis combines predictive performance with support for interactive refinement and constrained completion.
Memory as a Controlled Process: Learned Adaptive Memory Management for LLM Agents
MemCon is a new framework that treats LLM agent memory as a Markov Decision Process, adaptively managing retrieval and consolidation to improve task success and reduce token cost.
Protective Capacity Hallucination: When Large Language Models Claim Nonexistent Capabilities
A study identifies 'Protective Capacity Hallucination' (PCH), where LLMs claim to perform real-world protective actions they are incapable of executing.
Semantic Anchoring for Robotic Action Representations
Researchers propose a plug-and-play method to anchor robotic action representations to a semantic manifold, improving generalization in Vision-Language-Action (VLA) models.
The SIGReg Objective as Variational Free Energy: A Theoretical Active-Inference Account of JEPA World Models
A theoretical paper establishes a correspondence between SIGReg-based JEPA world models and Active Inference variational free energy, with proofs verified in Lean 4.
From Language to Navigation Goals: A Vision-Language Approach for Semantic Navigation of Mobile Robots Using RGB-D Perception
A new language-driven navigation framework uses ROS 2 and RGB-D perception to allow mobile robots to interpret natural language requests and navigate to targets.
OvisOCR2 Technical Report
OvisOCR2 is a new 0.8B end-to-end document parsing model that converts page images to Markdown, achieving state-of-the-art results on document benchmarks.
Consensus as Privileged Context for Label-Free Self-Distillation
CANON is a label-free self-distillation method that uses consensus signals as dense token-level supervision to improve LLM reasoning accuracy.
My car's OTA update broke Android Auto, and it's a indictment of modern software
A discussion on how over-the-air (OTA) updates in modern cars can break critical software like Android Auto, reflecting poor software quality and testing in the automotive industry.
The Little Book of Reinforcement Learning
A resource introduction to 'The Little Book of Reinforcement Learning', providing a structured approach to learning RL.
Show HN: Clx – Compile Lua to Native Executables Through C++20
Clx is a tool that compiles Lua scripts into native executables via C++20, improving performance and distribution.
Show HN: ReasonGate- An explainable gate that blocks LLM prompt injection
ReasonGate is an explainable gate mechanism designed to block LLM prompt injection attacks.