Accelerating Foundation Models with Privileged Information
Discover how PIQL framework uses privileged information to speed up training and improve accuracy of foundation models in machine learning.
Adaptive Response Length for Efficient Diffusion LLM Inference
Optimize Diffusion LLMs with Predict-then-Diffuse for adaptive response length, boosting efficiency and output quality in compute-budgeted AI inference.
MotionCache: Fast Autoregressive Video Generation
Discover MotionCache, a motion-aware caching method that boosts autoregressive video generation speed while preserving quality and reducing errors.
SIEVES Boosts Visual AI Accuracy with Selective Prediction
Discover how SIEVES improves visual-language AI models by scoring visual evidence for better selective prediction and robust out-of-distribution performanc...
Quotient-Space Diffusion Models for Symmetry-Aware AI
Discover how quotient-space diffusion models enhance generative AI by addressing symmetry, boosting molecular structure generation and scientific research.