[Preprint] AIPO: Improving Training Objective for Iterative Preference Optimization - View it on GitHub