AI研究論文デイリー

TokenFormer: Rethinking Transformer Scaling with Tokenized Model Parameters

A Large Recurrent Action Model: xLSTM enables Fast Inference for Robotics Tasks

Decoding Reading Goals from Eye Movements

On Memorization of Large Language Models in Logical Reasoning

Toxicity of the Commons: Curating Open-Source Pre-Training Data

CORAL: Benchmarking Multi-turn Conversational Retrieval-Augmentation Generation

AutoMIR: Effective Zero-Shot Medical Information Retrieval without Relevance Labels

Stealing User Prompts from Mixture of Experts

Can Models Help Us Create Better Models? Evaluating LLMs as Data Scientists

ReferEverything: Towards Segmenting Everything We Can Speak of in Videos

SlowFast-VGen: Slow-Fast Learning for Action-Driven Long Video Generation