Changing the Training Data Distribution to Reduce Simplicity Bias Improves In-distribution Generalization | Read Paper on Bytez