Domain Generalization via Gradient Surgery

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. Introduction Deep learning models have shown remarkable results in diverse application areas such as image understanding [13, 29] , speech recognition [10, 19] and natural language processing [25, 27] . Such models are typically trained under the standard supervised learning paradigm, assuming that training and test data come from the same distribution. However, in real life, training and test conditions may differ by several factors, such as a change in data acquisition device or target population. This makes models perform poorly when applied to test data whose distribution differs from the training data and, therefore, limits their implementation in such real scenarios. The goal is then to develop deep learning models that generalize outside the training distribution, under domain shift conditions.