Fetching the paper…

Understanding Simplicity Bias towards Compositional Mappings via Learning Dynamics · Around