Abstract:In the emerging Internet of Things (IoT) paradigm, mobile cloud inference serves as an efficient application framework that relieves the computation and storage burden on resource-constrained mobile devices offloading the workload to cloud servers. However, mobile cloud inference encounters computation, communication, and privacy challenges to ensure efficient system inference and protect the privacy of mobile users' collected information. To address the deployment of deep neural networks (DNN) with large capacity, we propose splitting computing (SC) where the entire model is divided into two parts, to be executed mobile and cloud ends respectively. However, the transmission of intermediate data poses a bottleneck system performance. This paper initially demonstrates the privacy issue arising from the machine analysis oriented intermediate feature. We conduct a preliminary experiment to intuitively reveal the latent potential for enhancing the privacy-preserving ability of the initial feature. Motivated by this, we propose a framework for privacy-preserving intermediate feature compression, which addresses the limitations in both compression and privacy that arise in the original extracted feature data. Specifically, we propose a method that jointly enhances privacy and encoding efficiency, achieved through the collaboration of the encoding feature privacy enhancement module and the privacy feature ordering enhancement module. Additionally, we develop gradient-reversal optimization strategy based on information theory to ensure the utmost concealment of core privacy information throughout the entire codec process. We evaluate the proposed method on two DNN models using two datasets, demonstrating its ability to achieve superior analysis accuracy and higher privacy preservation than HEVC. Furthermore, we provide an application case of a wireless sensor network to validate the effectiveness of the proposed method in a real-world scenario.

Toward Intelligent Sensing: Intermediate Deep Feature Compression

Intermediate Deep Feature Compression: the Next Battlefield of Intelligent Sensing

Lossy Intermediate Deep Learning Feature Compression and Evaluation

Sensitivity-Aware Bit Allocation for Intermediate Deep Feature Compression.

Towards Analysis-Friendly Face Representation with Scalable Feature and Texture Compression

2C-Net: integrate image compression and classification via deep neural network

Private Compression for Intermediate Feature in Iot-Supported Mobile Cloud Inference

Agnostic Feature Compression with Semantic Guided Channel Importance Analysis

Low-Rate Feature Compression for Collaborative Intelligence: Reducing Redundancy in Spatial and Statistical Levels

Texture-guided Coding for Deep Features

End-to-end Compression Towards Machine Vision: Network Architecture Design and Optimization

Deep Image Compression Toward Machine Vision: A Unified Optimization Framework

An End-to-End Channel-Adaptive Feature Compression Approach in Device-Edge Co-Inference Systems

Deep Image Compression Towards Machine Vision: A Unified Optimization Framework

Toward Scalable Image Feature Compression: A Content-Adaptive and Diffusion-Based Approach

A feature compression method based on similarity matching

Lightweight Compression of Intermediate Neural Network Features for Collaborative Intelligence

Privacy-preserved Intermediate Feature Compression for Cyber-Physical Systems.

Feature Compression for Cloud-Edge Multimodal 3D Object Detection

Towards Image Understanding from Deep Compression without Decoding

Deep Neural Network Interlayer Feature Map Compression Based on Least-Squares Fitting.