Abstract
Embedding the ability of sentiment analysis in smart devices is especially challenging because sentiment analysis relies on deep neural networks, in particular, convolutional neural networks. The paper presents a novel hardware-friendly detector of image polarity, enhanced with the ability of saliency detection. The approach stems from a hardware-oriented design process, which trades off prediction accuracy and computational resources. The eventual solution combines lightweight deep-learning architectures and post-training quantization. Experimental results on standard benchmarks confirmed that the design strategy can infer automatically the salient parts and the polarity of an image with high accuracy. Saliency-based solutions in the literature prove impractical due to their considerable computational costs; the paper shows that the novel design strategy can deploy and perform successfully on a variety of commercial smartphones, yielding real-time performances.
Highlights
The availability of ever-increasing computational power and the diffusion of distributed computing allow to apply deeplearning paradigms to complex problems
Deep learning lies at the core of a variety of applications supported by smart devices, but power consumption and hardware constraints tend to limit the deployment of those learning models
Sentiment analysis is a most interesting, yet challenging application relying on deep learning [1, 2], since it aims to extract the emotional information conveyed by media contents
Summary
The availability of ever-increasing computational power and the diffusion of distributed computing allow to apply deeplearning paradigms to complex problems. Convolutional neural networks (CNNs), for example, represent a key tool to deal with image/video processing domains, but convey a notable effort in the architecture design and bring about a considerable computational cost. This makes the real-time implementation of CNNs on embedded systems a very challenging task. That problem calls for a multidisciplinary approach, involving cognitive models [3], computational resources [4], Natural Language Processing [5, 6], and multimodal analysis [7]. To tackle the so-called subjective perception problem [8], i.e., different users perceive the same image in different ways, designers often envision custom solutions, involving both algorithms and hardware implementations
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have