Enhancing Reinforcement Learning with Contrastive Rewards
Discover ConSPO, a novel contrastive framework improving reinforcement learning with verifiable rewards for better AI reasoning and performance.
AdaFocus: Efficient Long Video Understanding with Adaptive Sampling
Discover AdaFocus, an AI method using adaptive sampling and zero-cache look-back for efficient, accurate long video understanding and multimedia reasoning.
Seg-Agent: Training-Free Language-Guided Image Segmentation
Discover Seg-Agent, a training-free framework for language-guided image segmentation using multimodal reasoning for accurate, real-time results.
Optimal AI Workflow Release with Always-Valid Inference
Discover when to release AI workflows using always-valid inference for reliable, adaptive generate-verify systems in dynamic AI environments.
Expressivity Limits of Probabilistic Circuits vs Large Language Models
Explore the expressivity boundaries of Probabilistic Circuits compared to Large Language Models and their impact on natural language processing.