Moving Deep Learning into Web Browser: How Far Can We Go?
share
Recently, several JavaScript-based deep learning frameworks have emerged, making it possible to perform deep learning tasks directly on browsers. However, there are debates on the necessity and efficiency of deep learning on browsers. On one hand, advocators think that given the cross-platform feature, the browser is an ideal platform to realize client-side machine learning, which can preserve the data privacy, increase the user personalization, as well as lower the backend workload. On the other hand, objectors think that deep learning on browsers is impractical due to the poor performance of JavaScript and constraints imposed by browsers. To make a first step towards consensus, in this paper, we conduct an empirical study of deep learning on browsers. We first survey 7 most popular JavaScript-based deep learning frameworks, investigating to what extent deep learning tasks have been supported so far. Then we investigate the performance of different frameworks when running different deep learning tasks. Finally, we dig out the performance gap between deep learning on browsers and on native platform by comparing the performance of TensorFlow.js and TensorFlow in Python. Our findings could help to improve the efficiency of deep learning on browsers.
READ FULL TEXT
