Abstract
As one of most fascinating machine learning techniques, deep neural network(DNN) has demonstrated excellent performance in various intelligent tasks suchas image classification. DNN achieves such performance, to a large extent, byperforming expensive training over huge volumes of training data. To reduce thedata storage and transfer overhead in smart resource-limited Internet-of-Thing(IoT) systems, effective data compression is a "must-have" feature beforetransferring real-time produced dataset for training or classification. Whilethere have been many well-known image compression approaches (such as JPEG), wefor the first time find that a human-visual based image compression approachsuch as JPEG compression is not an optimized solution for DNN systems,especially with high compression ratios. To this end, we develop an imagecompression framework tailored for DNN applications, named "DeepN-JPEG", toembrace the nature of deep cascaded information process mechanism of DNNarchitecture. Extensive experiments, based on "ImageNet" dataset with variousstate-of-the-art DNNs, show that "DeepN-JPEG" can achieve ~3.5x highercompression rate over the popular JPEG solution while maintaining the sameaccuracy level for image recognition, demonstrating its great potential ofstorage and power efficiency in DNN-based smart IoT system design.