Mixup: Beyond empirical risk minimization

🤖 Plain-English Summary

Large deep neural networks are powerful, but exhibit undesirable behaviors such as memorization and sensitivity to adversarial examples. Our experiments on the ImageNet-2012, CIFAR-10, CIFAR-100, Google commands and UCI datasets show that mixup improves the generalization of advanced neural network architectures.

🔑 Key Findings

• In this work, we propose mixup, a simple learning principle to alleviate these issues.
• In essence, mixup trains a neural network on convex combinations of pairs of examples and their labels.
• By doing so, mixup regularizes the neural network to favor simple linear behavior in-between training examples.

🔗 Related Resources