2021/08/20 by Roberto G. Pacheco, Pacheco, Roberto G., Fernanda D. V. R. Oliveira +3 · 1 citation
Computer Science · Medicine · #Advanced Neural Network Applications #Adversarial Robustness in Machine Learning #COVID-19 diagnosis using AI #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Networking and Internet Architecture (cs.NI)
paper · pdf · doi:10.48550/arxiv.2108.09343
openalex publication_date 2021/08/20 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
Edge offloading for deep neural networks (DNNs) can be adaptive to the\ninput's complexity by using early-exit DNNs. These DNNs have side branches\nthroughout their architecture, allowing the inference to end earlier in the\nedge. The branches estimate the accuracy for a given input. If this estimated\naccuracy reaches a threshold, the inference ends on the edge. Otherwise, the\nedge offloads the inference to the cloud to process the remaining DNN layers.\nHowever, DNNs for image classification deals with distorted images, which\nnegatively impact the branches' estimated accuracy. Consequently, the edge\noffloads more inferences to the cloud. This work introduces expert side\nbranches trained on a particular distortion type to improve robustness against\nimage distortion. The edge detects the distortion type and selects appropriate\nexpert branches to perform the inference. This approach increases the estimated\naccuracy on the edge, improving the offloading decisions. We validate our\nproposal in a realistic scenario, in which the edge offloads DNN inference to\nAmazon EC2 instances.\n