Shared encoder improves multi-task learning accuracy in vehicle networks
Encoder-Sharing Hierarchical Federated Multi-Task Learning for VANETs
Machine LearningArtificial IntelligenceDistributed, Parallel, and Cluster ComputingEmerging Technologies
Summary
Vehicles often need to learn different but related tasks, making it hard to train one model for all. The authors propose a way for cars to share a common part of their learning model (the encoder) while keeping their own specific parts (decoders) private. This helps the vehicles learn better features together without sharing raw data or task details. Tests on digit and traffic sign datasets showed their method improved accuracy and reduced the number of communication rounds needed. This approach is useful where vehicles have different tasks but want to collaborate safely and efficiently.
What this means in practice
- •For automotive software developers: Develop collaborative vehicle perception systems that support different tasks by sharing a common feature encoder while keeping task-specific models local.
- •For smart city planners: Improve distributed sensing networks in urban environments by enabling heterogeneous vehicles to share learned knowledge without sharing raw data.
Authors
M. Saeid HaghighiFard, Sinem Coleri
Abstract
Most federated learning frameworks for vehicular ad hoc networks assume that all vehicles collaboratively train a single model for a common task. This assumption limits their applicability to practical vehicular environments, where vehicles may perform heterogeneous but related perception tasks with different output spaces. This paper proposes encoder-sharing hierarchical multi-task federated learning (EN-HMTFL), which integrates cluster-based hierarchical federated learning with a globally shared encoder and vehicle-local decoders. EN-HMTFL enables vehicles performing different tasks to collaboratively learn a transferable feature representation while preserving their task-specific models locally. Only the encoder is exchanged and aggregated through the hierarchy, whereas raw data and local decoder parameters remain at the vehicles. The proposed framework is evaluated on the MNIST and GTSRB datasets in different vehicular scenarios. Across the evaluated scenarios, EN-HMTFL improves accuracy by up to 24.0% relative to the compared representation-sharing benchmark. In scenarios where EN-HMTFL converges earlier, the reduction reaches up to 69 communication rounds (28.8%).