Non-Redundant Feature Extraction in Mobile Edge Computing

Extracting discriminative features of data on Internet of Things (IoT) devices can reduce the amount of data uploaded by IoT devices to edge/cloud servers, thereby reducing the response time, which has attracted widespread attention from industry and academia. However, many existing related approaches have the following two limitations: (i) extracting a large number of redundant features, resulting in a waste of bandwidth resources, and (ii) extracting harmful features, resulting in low performance. To this end, this paper proposes a feature extractor generation algorithm for extracting non-redundant features, namely the Nor algorithm. The Nor algorithm analyzes existing discriminative feature extractors and finds that they contain a large number of redundant vectors. It further analyzes the locations where the redundant vectors appear and proposes how to remove them. The goal of this paper is to reduce the amount of data uploaded by IoT devices to the cloud/edge server by removing redundant vectors of the feature extractor, thereby reducing the data transmission time and the time it takes for the cloud/edge server to process features, and improving image recognition accuracy. Experimental results show that the Nor algorithm can effectively remove redundant vectors in the feature extractor and reduce the amount of feature data uploaded by IoT devices to the cloud/edge server. For example, in the Yale dataset, we can reduce the amount of feature data that needs to be uploaded to the edge server by 68.8% compared to the WAPL algorithm.