DecAug: Out-of-Distribution Generalization via Decomposed Feature Representation and Semantic Augmentation

While deep learning demonstrates its strong ability to handle independent and identically distributed (IID) data, it often suffers from out-of-distribution (OoD) generalization, where the test data come another distribution (w.r.t. training one). Designing a general OoD generalization framework for wide range of applications is challenging, mainly due different kinds shifts in real world, such as shift across domains or extrapolation correlation. Most previous approaches can only solve one specific shift, leading unsatisfactory performance when applied various benchmarks. In this work, we propose DecAug, novel decomposed feature representation semantic augmentation approach generalization. Specifically, DecAug disentangles category-related context-related features by orthogonalizing two gradients intermediate features) losses predicting category context labels, contain causal information target object, while cause between data. Furthermore, perform gradient-based on improve robustness learned representations. Experimental results show that outperforms other state-of-the-art methods datasets, which among very few deal with types challenges.