CRISP: Enhanced Relative Impact Analysis of Scholarly Papers
Discover CRISP, a novel method using LLMs to rank and analyze the relative impact of scholarly publications with improved accuracy and efficiency.
Federated Pretraining for Multimodal Large Language Models
Discover Fed-CMP, a novel federated pretraining method enhancing multimodal large language models while preserving data privacy and improving performance.
Optimize Diffusion Models by Adjusting Stroke Size
Learn how adjusting stroke size in diffusion models improves performance in low signal-to-noise conditions for clearer, more accurate results.
Challenges in Spatial Reasoning of Advanced LLM Models
Explore the limitations of frontier LLMs in spatial reasoning and the impact of integrating imagery modules on 3D mental rotation tasks.
TED: Training-Free Knowledge Distillation for Multimodal AI
Discover TED, a training-free knowledge distillation method that boosts multimodal reasoning performance without costly model updates or large datasets.