On-Device Federated Continual Learning on RISC-V-based Ultra-Low-Power SoC for Intelligent Nano-Drone Swarms
Lars Kröger, Cristian Cioflan, Victor Kartsch, Luca Benini
TL;DR
The paper tackles the challenge of enabling privacy-preserving, on-device continual learning in ultra-low-power nano-drones under severe memory and compute constraints. It introduces a regularization-based On-Device Federated Continual Learning (ODFCL) scheme built on FedProx, splitting a pretrained DSICNet into frozen features and a trainable head, and executing distributed training across a 10-core RISC-V SoC (GAP9Shield). Activation-based regularization and global-local model synchronization mitigate catastrophic forgetting while enabling on-device updates and aggregation. The results show a 24% improvement over naive fine-tuning, 46% accuracy on a 10-class face-recognition task, and practical timing/power metrics that demonstrate the approach’s feasibility for energy-constrained edge sensor networks and nano-drone swarms.
Abstract
RISC-V-based architectures are paving the way for efficient On-Device Learning (ODL) in smart edge devices. When applied across multiple nodes, ODL enables the creation of intelligent sensor networks that preserve data privacy. However, developing ODL-capable, battery-operated embedded platforms presents significant challenges due to constrained computational resources and limited device lifetime, besides intrinsic learning issues such as catastrophic forgetting. We face these challenges by proposing a regularization-based On-Device Federated Continual Learning algorithm tailored for multiple nano-drones performing face recognition tasks. We demonstrate our approach on a RISC-V-based 10-core ultra-low-power SoC, optimizing the ODL computational requirements. We improve the classification accuracy by 24% over naive fine-tuning, requiring 178 ms per local epoch and 10.5 s per global epoch, demonstrating the effectiveness of the architecture for this task.