MetaLDC: Meta Learning of Low-Dimensional Computing Classifiers for Fast On-Device Adaption
This addresses the challenge of enabling intelligent edge devices to quickly adapt to new tasks with minimal computational resources, representing an incremental improvement in efficient on-device learning.
The paper tackles the problem of fast model adaptation for unseen tasks on resource-constrained edge devices by proposing MetaLDC, a meta-learning approach for low-dimensional computing classifiers, which achieves higher accuracy, robustness against bit errors, and cost-efficient hardware computation compared to state-of-the-art baselines.
Fast model updates for unseen tasks on intelligent edge devices are crucial but also challenging due to the limited computational power. In this paper,we propose MetaLDC, which meta-trains braininspired ultra-efficient low-dimensional computing classifiers to enable fast adaptation on tiny devices with minimal computational costs. Concretely, during the meta-training stage, MetaLDC meta trains a representation offline by explicitly taking into account that the final (binary) class layer will be fine-tuned for fast adaptation for unseen tasks on tiny devices; during the meta-testing stage, MetaLDC uses closed-form gradients of the loss function to enable fast adaptation of the class layer. Unlike traditional neural networks, MetaLDC is designed based on the emerging LDC framework to enable ultra-efficient on-device inference. Our experiments have demonstrated that compared to SOTA baselines, MetaLDC achieves higher accuracy, robustness against random bit errors, as well as cost-efficient hardware computation.