Abstract:Convolutional neural networks (CNNs) have demonstrated encouraging results in image classification tasks. However, the prohibitive computational cost of CNNs hinders the deployment of CNNs onto resource-constrained embedded devices. To address this issue, we propose EdgeCompress, a comprehensive compression framework to reduce the computational overhead of CNNs. In EdgeCompress, we first introduce dynamic image cropping (DIC), where we design a lightweight foreground predictor to accurately crop the most informative foreground object of input images for inference, which avoids redundant computation on background regions. Subsequently, we present compound shrinking (CS) to collaboratively compress the three dimensions (depth, width, and resolution) of CNNs according to their contribution to accuracy and model computation. DIC and CS together constitute a multidimensional CNN compression framework, which is able to comprehensively reduce the computational redundancy in both input images and neural network architectures, thereby improving the inference efficiency of CNNs. Further, we present a dynamic inference framework to efficiently process input images with different recognition difficulties, where we cascade multiple models with different complexities from our compression framework and dynamically adopt different models for different input images, which further compresses the computational redundancy and improves the inference efficiency of CNNs, facilitating the deployment of advanced CNNs onto embedded hardware. Experiments on ImageNet-1K demonstrate that EdgeCompress reduces the computation of ResNet-50 by 48.8% while improving the top-1 accuracy by 0.8%. Meanwhile, we improve the accuracy by 4.1% with similar computation compared to HRank. The state-of-the-art compression framework. The source code and models are available at https://github.com/ntuliuteam/edge-compress .

An Edge-oriented Deep Learning Model Security Assessment Framework

Understanding Sensor Data Using Deep Learning Methods on Resource-Constrained Edge Devices.

What Happens on the Edge, Stays on the Edge: Toward Compressive Deep Learning

On-Demand Deep Model Compression for Mobile Devices

Security Analysis and Enhancement of Model Compressed Deep Learning Systems under Adversarial Attacks

A Robust Deep-Neural-Network-Based Compressed Model for Mobile Device Assisted by Edge Server

Deep Model Compression for Mobile Platforms: A Survey

EdgeCompress: Coupling Multidimensional Model Compression and Dynamic Inference for EdgeAI

Safety and Performance, Why Not Both? Bi-Objective Optimized Model Compression against Heterogeneous Attacks Toward AI Software Deployment

A Convenient Deep Learning Model Attack and Defense Evaluation Analysis Platform.

Context-Aware Deep Model Compression for Edge Cloud Computing

EdgeShield: A Universal and Efficient Edge Computing Framework for Robust AI

A Comprehensive Evaluation Framework for Deep Model Robustness

Safety and Performance, Why not Both? Bi-Objective Optimized Model Compression toward AI Software Deployment

A Survey of Security Protection Methods for Deep Learning Model

Bringing AI To Edge: From Deep Learning's Perspective

EdgePro: Edge Deep Learning Model Protection Via Neuron Authorization

Deep Learning for Secure Mobile Edge Computing

EdgeKE: An On-Demand Deep Learning IoT System for Cognitive Big Data on Industrial Edge Devices

Edge enhanced deep learning system for IoT edge device security analytics