Compressing sentiment analysis CNN models for efficient hardware processing

Abstract

Convolutional neural networks (CNNs) were created for image classification tasks. Shortly after their creation, they were applied to other domains, including natural language processing (NLP). Nowadays, solutions based on artificial intelligence appear on mobile devices and embedded systems, which places constraints on memory and power consumption, among others. Due to CNN memory and computing requirements, it is necessary to compress them in order to be mapped to the hardware. This paper presents the results of the compression of efficient CNNs for sentiment analysis. The main steps involve pruning and quantization. The process of mapping the compressed network to an FPGA and the results of this implementation are described. The conducted simulations showed that the 5-bit width is enough to ensure no drop in accuracy when compared to the floating-point version of the network. Additionally, the memory footprint was significantly reduced (between 85 and 93% as compared to the original model).