System Demo for Transfer Learning across Vision and Text using Domain Specific CNN Accelerator for On-Device NLP Applications

TL;DR

This paper demonstrates the implementation of transfer learning for text classification and sentiment analysis on mobile devices using power-efficient CNN-DSA chips, achieving state-of-the-art results with low power consumption.

Contribution

It introduces a novel application of CNN-DSA chips for on-device NLP tasks using transfer learning from vision models, with optimized network compression techniques.

Findings

Achieved power consumption below 300mW on mobile devices.

Successfully classified English sentences into 14 ontologies.

Performed sentiment analysis on Chinese reviews.

Abstract

Power-efficient CNN Domain Specific Accelerator (CNN-DSA) chips are currently available for wide use in mobile devices. These chips are mainly used in computer vision applications. However, the recent work of Super Characters method for text classification and sentiment analysis tasks using two-dimensional CNN models has also achieved state-of-the-art results through the method of transfer learning from vision to text. In this paper, we implemented the text classification and sentiment analysis applications on mobile devices using CNN-DSA chips. Compact network representations using one-bit and three-bits precision for coefficients and five-bits for activations are used in the CNN-DSA chip with power consumption less than 300mW. For edge devices under memory and compute constraints, the network is further compressed by approximating the external Fully Connected (FC) layers within the CNN-DSA chip. At the workshop, we have two system demonstrations for NLP tasks. The first demo classifies the input English Wikipedia sentence into one of the 14 ontologies. The second demo classifies the Chinese online-shopping review into positive or negative.