Performance Analysis of Hyperdimensional Computing for Character Recognition

The Internet of Things (IoT) is an increasingly expanding network of devices. The whole premise of IoT is to provide comfort and convenience, requiring functions such as recognition. This is done through data - first collected by nodes, and then processed to provide meaningful information, commonly using machine learning. However, conventional machine learning techniques such as deep-learning require complicated operations, which may be computationally heavy for small devices. This study explores the capabilities of hyperdimensional computing (HDC), as a less complex architecture for data classification. The effect of dimensionality (and therefore energy) in the performance of HDC, done through a character recognition application, was explored. Results show that a dimensionality of 4,000-bit one-shot learning system yields an average accuracy close to that of its 12,000-bit counterpart at 0% distortion, and an average accuracy of 89.94% with 14.29% distortion. This study provides insights towards optimizations of HDC applications.