12 mJ per Class On-Device Online Few-Shot Class-Incremental Learning
This work addresses the challenge of efficient few-shot class-incremental learning for ultra-low-power edge devices, which is incremental as it builds on existing FSCIL methods with hardware-specific optimizations.
The paper tackles the problem of enabling machine learning systems to learn new classes with few examples on battery-powered, memory-constrained edge devices, achieving state-of-the-art results with 68.62% average accuracy on the FSCIL CIFAR100 benchmark and demonstrating online learning with 12 mJ per new class on a 60 mW microcontroller.
Few-Shot Class-Incremental Learning (FSCIL) enables machine learning systems to expand their inference capabilities to new classes using only a few labeled examples, without forgetting the previously learned classes. Classical backpropagation-based learning and its variants are often unsuitable for battery-powered, memory-constrained systems at the extreme edge. In this work, we introduce Online Few-Shot Class-Incremental Learning (O-FSCIL), based on a lightweight model consisting of a pretrained and metalearned feature extractor and an expandable explicit memory storing the class prototypes. The architecture is pretrained with a novel feature orthogonality regularization and metalearned with a multi-margin loss. For learning a new class, our approach extends the explicit memory with novel class prototypes, while the remaining architecture is kept frozen. This allows learning previously unseen classes based on only a few examples with one single pass (hence online). O-FSCIL obtains an average accuracy of 68.62% on the FSCIL CIFAR100 benchmark, achieving state-of-the-art results. Tailored for ultra-low-power platforms, we implement O-FSCIL on the 60 mW GAP9 microcontroller, demonstrating online learning capabilities within just 12 mJ per new class.