All Articles
17939 articles total
Vero: An Open RL Recipe for General Visual Reasoning
Introduction of Vero, a family of open-weight VLMs and a 600K-sample dataset designed to improve general visual reasoning tasks.
Automated Standardization of Legacy Biomedical Metadata Using an Ontology-Constrained LLM Agent
A system using LLM agents and real-time ontology queries to automate the standardization of legacy biomedical metadata.
FM-Agent: Scaling Formal Methods to Large Systems via LLM-Based Hoare-Style Reasoning
FM-Agent leverages LLMs to automate compositional reasoning and specification generation for large-scale software systems to find bugs.
DF3DV-1K: A Large-Scale Dataset and Benchmark for Distractor-Free Novel View Synthesis
Introduction of DF3DV-1K, a large-scale dataset for benchmarking distractor-free novel view synthesis in radiance fields.
Mitigating Simplicity Bias in OOD Detection through Object Co-occurrence Analysis
The OCO framework improves out-of-distribution detection by analyzing object co-occurrence patterns in images.
CADBench: A Multimodal Benchmark for AI-Assisted CAD Program Generation
CADBench is a multimodal benchmark for evaluating AI-assisted generation of editable CAD programs from images and 3D observations.
Superhuman Safe and Agile Racing through Multi-Agent Reinforcement Learning
Multi-agent reinforcement learning is used to achieve superhuman, safe, and agile high-speed quadrotor racing.
Any2Any: Efficient Cross-Embodiment Transfer for Humanoid Whole-Body Tracking
Any2Any provides an efficient transfer paradigm for humanoid whole-body tracking models across different robot embodiments.
I solved my mystery fatigue with AI
A personal account of using AI to diagnose and solve chronic fatigue.
DeFrame: Debiasing Large Language Models Against Framing Effects
Introduces DeFrame, a method to debias LLMs against framing effects where semantically equivalent prompts produce different fairness outcomes.
LoRDO: Distributed Low-Rank Optimization with Infrequent Communication
Presents LoRDO, a framework for distributed low-rank optimization that reduces communication overhead by 10x during foundation model training.
Flickering Multi-Armed Bandits
Introduces Flickering Multi-Armed Bandits (FMAB) to model sequential decision-making in environments with changing action availability.
Reinforcement-aware Knowledge Distillation for LLM Reasoning
Proposes RL-aware Knowledge Distillation (RLAD) to efficiently distill reasoning capabilities from large LLMs into smaller students using reinforcement learning.
Latent Gaussian Splatting for 4D Panoptic Occupancy Tracking
Presents Latent Gaussian Splatting (LaGS) for 4D panoptic occupancy tracking in dynamic robotic environments.
The MAMA-MIA Challenge: Advancing Generalizability and Fairness in Breast MRI Tumor Segmentation and Treatment Response Prediction
Reports on the MAMA-MIA Challenge, a benchmark for breast MRI tumor segmentation and treatment response prediction using AI.
ZeSTA: Zero-Shot TTS Augmentation with Domain-Conditioned Training for Data-Efficient Personalized Speech Synthesis
Introduces ZeSTA, a domain-conditioned training framework for data-efficient personalized speech synthesis using zero-shot TTS augmentation.
Class-Incremental Motion Forecasting
Proposes an end-to-end framework for class-incremental motion forecasting in autonomous vehicles to adapt to new object classes without forgetting.
The Autonomy Tax: Defense Training Breaks LLM Agents
Analyzes the 'Autonomy Tax,' showing how defense training against prompt injections often degrades the competence of multi-step LLM agents.
How to feed a dictator
A discussion thread on Hacker News titled 'How to feed a dictator'.
A Love Story
A Hacker News post titled 'A Love Story'.