Domain Generalization via Gradient Surgery

TL;DR

This paper introduces a gradient surgery method to resolve conflicting gradients during training, improving the generalization of deep learning models across unseen domains in image classification tasks.

Contribution

It proposes a novel gradient agreement strategy based on gradient surgery to mitigate conflicting gradients in domain generalization, enhancing model performance on unseen domains.

Findings

Improved accuracy on multi-domain image classification datasets.

Effective reduction of gradient conflicts during training.

Enhanced generalization to unseen target domains.

Abstract

In real-life applications, machine learning models often face scenarios where there is a change in data distribution between training and test domains. When the aim is to make predictions on distributions different from those seen at training, we incur in a domain generalization problem. Methods to address this issue learn a model using data from multiple source domains, and then apply this model to the unseen target domain. Our hypothesis is that when training with multiple domains, conflicting gradients within each mini-batch contain information specific to the individual domains which is irrelevant to the others, including the test domain. If left untouched, such disagreement may degrade generalization performance. In this work, we characterize the conflicting gradients emerging in domain shift scenarios and devise novel gradient agreement strategies based on gradient surgery to alleviate their effect. We validate our approach in image classification tasks with three multi-domain datasets, showing the value of the proposed agreement strategy in enhancing the generalization capability of deep learning models in domain shift scenarios.