Federated learning (FL) is a distributed learning paradigm that enables multiple clients to learn a powerful global model by aggregating local training. However, the performance of the global model is often hampered by non-i.i.d. distribution among the clients, requiring extensive efforts to mitigate inter-client data heterogeneity. Going beyond inter-client data heterogeneity, we note that intra-client heterogeneity can also be observed on complex real-world data and seriously deteriorate FL performance. In this paper, we present a novel FL algorithm, i.e., FedIns, to handle intra-client data heterogeneity by enabling instance-adaptive inference in the FL framework. Instead of huge instance-adaptive models, we resort to a parameter-efficient fine-tuning method, i.e., scale and shift deep features (SSF), upon a pre-trained model. Specifically, we first train an SSF pool for each client, and aggregate these SSF pools on the server side, thus still maintaining a low communication cost. To enable instance-adaptive inference, for a given instance, we dynamically find the best-matched SSF subsets from the pool and aggregate them to generate an adaptive SSF specified for the instance, thereby reducing the intra-client as well as the inter-client heterogeneity. Extensive experiments show that our FedIns outperforms state-of-the-art FL algorithms, e.g., a 6.64\% improvement against the top-performing method with less than 15\% communication cost on Tiny-ImageNet. Our code and models will be publicly released.

Federated learning (FL) facilitates edge devices to cooperatively train a global shared model while maintaining the training data locally and privately. However, a common but impractical assumption in FL is that the participating edge devices possess the same required resources and share identical global model architecture. In this study, we propose a novel FL method called Federated Intermediate Layers Learning (FedIN), supporting heterogeneous models without utilizing any public dataset. The training models in FedIN are divided into three parts, including an extractor, the intermediate layers, and a classifier. The model architectures of the extractor and classifier are the same in all devices to maintain the consistency of the intermediate layer features, while the architectures of the intermediate layers can vary for heterogeneous devices according to their resource capacities. To exploit the knowledge from features, we propose IN training, training the intermediate layers in line with the features from other clients. Additionally, we formulate and solve a convex optimization problem to mitigate the gradient divergence problem induced by the conflicts between the IN training and the local training. The experiment results show that FedIN achieves the best performance in the heterogeneous model environment compared with the state-of-the-art algorithms. Furthermore, our ablation study demonstrates the effectiveness of IN training and the solution to the convex optimization problem.