FedDSR: Federated Deep Supervision and Regularization Towards Autonomous Driving
Wei-Bin Kou, Guangxu Zhu, Bingyang Cheng, Chen Zhang, Yik-Chung Wu, Jianping Wang
TL;DR
This paper addresses the generalization and convergence challenges of federated learning in autonomous driving under non-IID data by introducing FedDSR, a framework that injects mutual-information-based intermediate supervision and negative-entropy regularization at multiple architecture-agnostic points in deep networks. It presents a formal formulation, practical training algorithm, and a convergence analysis showing O(1/√T) rate with drift terms tied to local updates and data heterogeneity. Extensive experiments on semantic segmentation across Cityscapes, CamVid, and SynthiaSF—with models like DeepLabv3+, SeaFormer, and TopFormer—demonstrate FedDSR’s ability to improve mIoU by up to 8.93% and reduce training rounds by up to 28.57% compared with FedAvg and other baselines, including beneficial interactions with several FL methods. Ablation studies confirm that a moderate number of intermediate points, appropriate spacing, and early placement of points yield the best performance, underscoring FedDSR’s practical design guidelines for federated AD systems.
Abstract
Federated Learning (FL) enables collaborative training of autonomous driving (AD) models across distributed vehicles while preserving data privacy. However, FL encounters critical challenges such as poor generalization and slow convergence due to non-independent and identically distributed (non-IID) data from diverse driving environments. To overcome these obstacles, we introduce Federated Deep Supervision and Regularization (FedDSR), a paradigm that incorporates multi-access intermediate layer supervision and regularization within federated AD system. Specifically, FedDSR comprises following integral strategies: (I) to select multiple intermediate layers based on predefined architecture-agnostic standards. (II) to compute mutual information (MI) and negative entropy (NE) on those selected layers to serve as intermediate loss and regularizer. These terms are integrated into the output-layer loss to form a unified optimization objective, enabling comprehensive optimization across the network hierarchy. (III) to aggregate models from vehicles trained based on aforementioned rules of (I) and (II) to generate the global model on central server. By guiding and penalizing the learning of feature representations at intermediate stages, FedDSR enhances the model generalization and accelerates model convergence for federated AD. We then take the semantic segmentation task as an example to assess FedDSR and apply FedDSR to multiple model architectures and FL algorithms. Extensive experiments demonstrate that FedDSR achieves up to 8.93% improvement in mIoU and 28.57% reduction in training rounds, compared to other FL baselines, making it highly suitable for practical deployment in federated AD ecosystems.