Boosting Generative Model Performance: A Trajectory Diversity Approach

This research explores methods to improve the performance of Generative Models through trajectory diversification, specifically focusing on the GRPO (Generative Reinforcement Policy Optimization) framework. The novelty likely lies in the specific 'Expand and Prune' strategy for enhancing the exploration capabilities within the generative process.