RePO: Understanding Preference Learning Through ReLU-Based Optimization | Read Paper on Bytez