Integrating event-based dynamic vision sensors with sparse hyperdimensional computing

Abstract

We propose to embed features extracted from event-driven dynamic vision sensors to binary sparse representations in hyperdimensional (HD) space for regression. This embedding compresses events generated across 346×260 differential pixels to a sparse 8160bit vector by applying random activation functions. The sparse representation not only simplifies inference, but also enables online learning with the same memory footprint. Specifically, it allows efficient updates by retaining binary vector components over the course of online learning that cannot be otherwise achieved with dense representations demanding multibit vector components. We demonstrate online learning capability: using estimates and confidences of an initial model trained with only 25% of data, our method continuously updates the model for the remaining 75% of data, resulting in a close match with accuracy obtained with an oracle model on ground truth labels. When mapped on an 8-core accelerator, our method also achieves lower error, latency, and energy compared to other sparse/dense alternatives. Furthermore, it is 9.84× more energy-efficient and 6.25× faster than an optimized 9-layer perceptron with comparable accuracy. CCS Concepts• Computer systems organization → Embedded software; • Computing methodologies → Machine learning.

A Primer on Hyperdimensional Computing for iEEG Seizure Detection

“…Another recent innovative approach is to combine HD computing with event-driven inputs such as dynamic vision sensors ( 84 ). Hersche et al ( 85 ) showed how to embed features extracted from such spiking sensors into binary sparse representations to reduce the complexity of downstream tasks such as regression.…”

Energy Efficient In-Memory Hyperdimensional Encoding for Spatio-Temporal Signal Processing

“…Sequential processing of the data record allows time sharing the sense amplifier array and the downstream binder module, allowing us to save a significant amount of energy and area in the peripherals. The binder module consists of an array of D XOR gates daisy-chained with D registers, which collectively implement (6).…”

“…HDC has been deployed in a diverse set of application domains, for instance in solving Raven's progressive matrices [2], analogical reasoning [3], natural language processing [4], robotics [5], [6], text classification [7]- [10], activity recognition [11], DNA sequencing [12], and biosignal processing [13]- [18] (see [19] for an overview).…”

An Extension to Basis-Hypervectors for Learning from Circular Data in Hyperdimensional Computing

“…HDC has already proven to be useful in several applications, including both learning problems, such as classification [12] and regression [14], and classical problems, such as consistent hashing [13]. Regardless of the application, the most fundamental step in HDC is mapping objects in the input space to the hyperspace, a process called encoding.…”