Hypervector Design for Efficient Hyperdimensional Computing on Edge Devices
Explore hypervector design for efficient hyperdimensional computing on edge devices in this 21-minute conference talk from the tinyML Research Symposium 2021. Delve into a novel technique that minimizes hypervector dimension while maintaining accuracy and improving classifier robustness. Learn how this approach formulates hypervector design as a multi-objective optimization problem, achieving a 128x reduction in hypervector dimension and significant improvements in model size, inference time, and energy consumption. Discover the trade-offs between accuracy and robustness to noise, and gain insights into Pareto front solutions as a design parameter in hypervector design. Understand the potential applications in wearable health monitoring and neural processing, as presented by PhD student Toygun Basaklar from the University of Wisconsin-Madison.