Category: Computing
-
SpecEyes: Accelerating Agentic Multimodal LLMs via Speculative Perception and Planning
Discover how the SpecEyes framework accelerates agentic multimodal large language models (MLLMs) by enhancing speed and accuracy in AI applications.
-
One View Is Enough! Monocular Training for In-the-Wild Novel View Generation
Discover how OVIE transforms monocular novel-view synthesis by training on single images, achieving faster and more efficient results than traditional methods.
-
VISion On Request: Enhanced VLLM efficiency with sparse, dynamically selected, vision-language interactions
Explore how VISOR enhances the efficiency of Large Vision-Language Models, preserving visual information while reducing computational costs for complex tasks.
-
Exploring Novel Data Storage Approaches for Large-Scale Numerical Weather Prediction
Explore the performance of DAOS and Ceph object storage systems for HPC and AI applications, highlighting their advantages over traditional file systems.
-
When to Trust the Cheap Check: Weak and Strong Verification for Reasoning
Explore the balance between weak and strong verification in large language models, enhancing output trustworthiness while managing resources effectively.
-
MARS: Margin-Aware Reward-Modeling with Self-Refinement
Discover how MARS enhances reward modeling in AI by targeting ambiguous preference pairs for improved training efficiency. Learn more!
-
Reverso: Efficient Time Series Foundation Models for Zero-shot Forecasting
Discover efficient time series foundation models for zero-shot forecasting, highlighting the Reverso family that balances performance with reduced size.