In this paper, our goal is to design a simple learning paradigm for long-tail visual recognition, which not only improves the robustness of the feature extractor but also alleviates the bias of the classifier towards head classes while reducing the training skills and overhead. We propose an efficient one-stage training strategy for long-tailed visual recognition called Global and Local Mixture Consistency cumulative learning (GLMC). Our core ideas are twofold: (1) a global and local mixture consistency loss improves the robustness of the feature extractor. Specifically, we generate two augmented batches by the global MixUp and local CutMix from the same batch data, respectively, and then use cosine similarity to minimize the difference. (2) A cumulative head-tail soft label reweighted loss mitigates the head class bias problem. We use empirical class frequencies to reweight the mixed label of the head-tail class for long-tailed data and then balance the conventional loss and the rebalanced loss with a coefficient accumulated by epochs. Our approach achieves state-of-the-art accuracy on CIFAR10-LT, CIFAR100-LT, and ImageNet-LT datasets. Additional experiments on balanced ImageNet and CIFAR demonstrate that GLMC can significantly improve the generalization of backbones. Code is made publicly available at https://github.com/ynu-yangpeng/GLMC

Federated learning (FL) is a rising distributed machine learning area, which aims to train a high-performing global model with data collected from a number of local clients. Many FL applications receive data over time in the form of data streams. Streaming data are likely to suffer concept drift. It can significantly harm a model’s predictive ability. However, no study has characterized concept drift in FL or investigated how it can affect the global and local models’ performance. This paper aims to provide such understanding by 1) categorizing concept drift in temporal and spatial dimensions with ten features and 2) investigating the impact of the features in depth. We find that: the temporal features degrade FL models to a different extend and do not affect model convergence after the new data concept becomes stable; the spatial features cause data heterogeneity and affect both accuracy and convergence speed.