Header

UZH-Logo

Maintenance Infos

Closing the Accuracy Gap in an Event-Based Visual Recognition Task


Rückauer, Bodo; Känzig, Nicolas; Liu, Shih-Chii; Delbruck, Tobi; Sandamirskaya, Yulia (2019). Closing the Accuracy Gap in an Event-Based Visual Recognition Task. arXiv.org 1906.08859, University of Zurich.

Abstract

Mobile and embedded applications require neural networks-based pattern recognition systems to perform well under a tight computational budget. In contrast to commonly used synchronous, frame-based vision systems and CNNs, asynchronous, spiking neural networks driven by event-based visual input respond with low latency to sparse, salient features in the input, leading to high efficiency at run-time. The discrete nature of the event-based data streams makes direct training of asynchronous neural networks challenging. This paper studies asynchronous spiking neural networks, obtained by conversion from a conventional CNN trained on frame-based data. As an example, we consider a CNN trained to steer a robot to follow a moving target. We identify possible pitfalls of the conversion and demonstrate how the proposed solutions bring the classification accuracy of the asynchronous network to only 3\% below the performance of the original synchronous CNN, while requiring 12x fewer computations. While being applied to a simple task, this work is an important step towards low-power, fast, and embedded neural networks-based vision solutions for robotic applications.

Abstract

Mobile and embedded applications require neural networks-based pattern recognition systems to perform well under a tight computational budget. In contrast to commonly used synchronous, frame-based vision systems and CNNs, asynchronous, spiking neural networks driven by event-based visual input respond with low latency to sparse, salient features in the input, leading to high efficiency at run-time. The discrete nature of the event-based data streams makes direct training of asynchronous neural networks challenging. This paper studies asynchronous spiking neural networks, obtained by conversion from a conventional CNN trained on frame-based data. As an example, we consider a CNN trained to steer a robot to follow a moving target. We identify possible pitfalls of the conversion and demonstrate how the proposed solutions bring the classification accuracy of the asynchronous network to only 3\% below the performance of the original synchronous CNN, while requiring 12x fewer computations. While being applied to a simple task, this work is an important step towards low-power, fast, and embedded neural networks-based vision solutions for robotic applications.

Statistics

Downloads

1 download since deposited on 14 Feb 2020
1 download since 12 months
Detailed statistics

Additional indexing

Item Type:Working Paper
Communities & Collections:07 Faculty of Science > Institute of Neuroinformatics
Dewey Decimal Classification:570 Life sciences; biology
Language:English
Date:2019
Deposited On:14 Feb 2020 11:12
Last Modified:14 Feb 2020 11:15
Series Name:arXiv.org
ISSN:2331-8422
OA Status:Green
Free access at:Official URL. An embargo period may apply.
Official URL:https://arxiv.org/abs/1906.08859

Download

Green Open Access

Download PDF  'Closing the Accuracy Gap in an Event-Based Visual Recognition Task'.
Preview
Content: Published Version
Filetype: PDF
Size: 1MB