Introduction

In recent years, deep learning technology has developed rapidly and has attracted significant research interests from both academia and industry due to its excellent performance in various fields. Due to the strong data dependence, deep neural networks (DNNs) are easily subjected to data poisoning attacks, which undermine the availability and integrity of neural network models. Additionally, the unexplainable nature of deep neural networks increases the difficulty of defending against these attacks, posing a challenge to the security of learning-based algorithms. In recent years, a new type of data poisoning attack has emerged, dubbed, “backdoor attacks.” In backdoor attacks, an attacker can train the model with poisoned data to obtain a model that performs well on a normal input but behaves wrongly with crafted triggers. In this chapter, we first survey the security and privacy issues of deep learning (DL). Then, we summarize the motivation and challenges of the existing backdoor attacks in learning-based algorithms. After that, we briefly overview three key issues introduced in this monograph. At last, we present the aims and organization of the monograph.