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

The size of deep learning models in artificial intelligence (AI) software is increasing rapidly, hindering the large-scale deployment on resource-restricted devices (e.g., smartphones). To mitigate this issue, AI software compression plays a crucial role, which aims to compress model size while keeping high performance. However, the intrinsic defects in a big model may be inherited by the compressed one. Such defects may be easily leveraged by adversaries, since a compressed model is usually deployed in a large number of devices without adequate protection. In this article, we aim to address the safe model compression problem from the perspective of safety-performance co-optimization. Specifically, inspired by the test-driven development (TDD) paradigm in software engineering, we propose a test-driven sparse training framework called SafeCompress. By simulating the attack mechanism as safety testing, SafeCompress can automatically compress a big model to a small one following the dynamic sparse training paradigm. Then, considering two kinds of representative and heterogeneous attack mechanisms, i.e., black-box membership inference attack and white-box membership inference attack, we develop two concrete instances called BMIA-SafeCompress and WMIA-SafeCompress. Further, we implement another instance called MMIA-SafeCompress by extending SafeCompress to defend against the occasion when adversaries conduct black-box and white-box membership inference attacks simultaneously. We conduct extensive experiments on five datasets for both computer vision and natural language processing tasks. The results show the effectiveness and generalizability of our framework. We also discuss how to adapt SafeCompress to other attacks besides membership inference attack, demonstrating the flexibility of SafeCompress.

I. Introduction

In the last decade, artificial intelligence (AI) software, especially software based on deep neural networks (DNN), has attracted much attention and made a significant influence [1]. Currently, AI software, with DNN as representatives, is recognized as an emerging type of software artifact (sometimes known as “software 2.0” [2]). Notably, the size of DNN-based AI software has increased rapidly in recent years (mostly because of a trained deep neural network model). For instance, a state-of-the-art model of computer vision contains more than 15 billion parameters [3]. A recent natural language model, GPT-3, is even bigger, surpassing 175 billion parameters; this situation requires nearly 1TB of space to store only the model [4]. Such a big model hinders realistic applications such as autonomous driving when the software is required to be deployed in resource-restricted devices such as wearable devices or edge nodes. To this end, a new branch is derived from the traditional area of software compression [5], [6], called AI software compression (especially DNN model compression¹

In the rest of the article, without incurring ambiguity, we use ‘model compression’ to indicate ‘DNN model compression in AI software’ for brevity.

), and has attracted a lot of research interest.