MetaLDC: Meta Learning of Low-Dimensional Computing Classifiers for Fast On-Device Adaptation

Watch a 15-minute research symposium presentation exploring MetaLDC, a novel approach for enabling fast model updates on edge devices through meta-learning of low-dimensional computing classifiers. Learn how PhD student Yejia LIU from UC Riverside addresses the challenges of limited computational power in intelligent edge devices by combining brain-inspired ultra-efficient low-dimensional computing with meta-training techniques. Discover the framework's two-stage process: offline meta-training of representations and fast adaptation using closed-form gradients during meta-testing. Examine experimental results demonstrating MetaLDC's superior performance in accuracy, robustness against random bit errors, and computational efficiency compared to state-of-the-art baselines. Gain insights into vector symbolic architecture, hyper-dimensional computing, and low-dimensional classifiers while understanding their practical applications in edge AI development.