Learning to Reweight Examples for Robust Deep Learning

TLDR

Deep neural networks are powerful for complex supervised tasks but are prone to overfitting training biases and label noise, and while example reweighting algorithms can mitigate these issues, they typically require careful tuning of additional hyperparameters. The authors propose a novel meta‑learning algorithm that learns to assign weights to training examples based on their gradient directions. The method performs a meta‑gradient descent step on the example weights—initialized at zero—to minimize loss on a clean, unbiased validation set. This approach is easily implemented on any deep network, requires no extra hyperparameter tuning, and achieves strong performance on class‑imbalance and corrupted‑label problems with only a small amount of clean validation data.

Abstract

Deep neural networks have been shown to be very powerful modeling tools for many supervised learning tasks involving complex input patterns. However, they can also easily overfit to training set biases and label noises. In addition to various regularizers, example reweighting algorithms are popular solutions to these problems, but they require careful tuning of additional hyperparameters, such as example mining schedules and regularization hyperparameters. In contrast to past reweighting methods, which typically consist of functions of the cost value of each example, in this work we propose a novel meta-learning algorithm that learns to assign weights to training examples based on their gradient directions. To determine the example weights, our method performs a meta gradient descent step on the current mini-batch example weights (which are initialized from zero) to minimize the loss on a clean unbiased validation set. Our proposed method can be easily implemented on any type of deep network, does not require any additional hyperparameter tuning, and achieves impressive performance on class imbalance and corrupted label problems where only a small amount of clean validation data is available.