Machine learning is one of the most prevailing techniques in computer science, and it has been widely applied in image processing, natural language processing, pattern recognition, cybersecurity, and other fields. Regardless of successful applications of machine learning algorithms in many scenarios, e.g., facial recognition, malware detection, automatic driving, and intrusion detection, these algorithms and corresponding training data are vulnerable to a variety of security threats, inducing a significant performance decrease. Hence, it is vital to call for further attention regarding security threats and corresponding defensive techniques of machine learning, which motivates a comprehensive survey in this paper. Until now, researchers from academia and industry have found out many security threats against a variety of learning algorithms, including naive Bayes, logistic regression, decision tree, support vector machine (SVM), principle component analysis, clustering, and prevailing deep neural networks. Thus, we revisit existing security threats and give a systematic survey on them from two aspects, the training phase and the testing/inferring phase. After that, we categorize current defensive techniques of machine learning into four groups: security assessment mechanisms, countermeasures in the training phase, those in the testing or inferring phase, data security, and privacy. Finally, we provide five notable trends in the research on security threats and defensive techniques of machine learning, which are worth doing in-depth studies in future.

As the extension of cloud computing and a foundation of IoT, fog computing is experiencing fast prosperity because of its potential to mitigate some troublesome issues such as network congestion, latency, and the local autonomy. However, privacy issues and the subsequent inefficiency are dragging down the performances of fog computing. The majority of existing works hardly consider a reasonable balance between them while suffering from poisoning attacks. To address the aforementioned issues, we propose a novel blockchain-enabled federated learning (FL-Block) scheme to close the gap. FL-Block allows local learning updates of end devices exchanges with blockchain-based global learning model, which is verified by miners. Built upon this, the FL-Block enables autonomous machine learning without any centralized authority to maintain the global model and coordinate by using a PoW consensus mechanism of blockchain. Furthermore, we analyze the latency performance of FL-Block and further derive the optimal block generation rate by taking communication, consensus delays, and computation cost into consideration. Extensive evaluation results show the superior performances of FL-Block from the aspects of privacy protection, efficiency, and resistance to poisoning attack.

Blockchain technology has been deemed to be an ideal choice for strengthening existing computing systems in varied manners. As one of the network-enabled technologies, cloud computing has been broadly adopted in the industry through numerous cloud service models. Fusing blockchain technology with existing cloud systems has a great potential in both functionality/ performance enhancement and security/ privacy improvement. The question remains on how blockchain technology inserts into current deployed cloud solutions and enables the reengineering of cloud datacenter. This survey addresses this issue and investigates recent efforts in the technical fusion of blockchain and clouds. Three technical dimensions roughly are covered in this work. First, we concern the service model and review an emerging cloud-relevant blockchain service model, Blockchain-as-a-Service (BaaS); second, security is considered a key technical dimension in this work and both access control and searchable encryption schemes are assessed; finally, we examine the performance of cloud datacenter with supports/ participance of blockchain from hardware and software perspectives. Main findings of this survey will be theoretical supports for future reference of blockchain-enabled reengineering of cloud datacenter.

The prosperity of machine learning has been accompanied by increasing attacks on the training process. Among them, poisoning attacks have become an emerging threat during model training. Poisoning attacks have profound impacts on the target models, e.g., making them unable to converge or manipulating their prediction results. Moreover, the rapid development of recent distributed learning frameworks, especially federated learning, has further stimulated the development of poisoning attacks. Defending against poisoning attacks is challenging and urgent. However, the systematic review from a unified perspective remains blank. This survey provides an in-depth and up-to-date overview of poisoning attacks and corresponding countermeasures in both centralized and federated learning. We firstly categorize attack methods based on their goals. Secondly, we offer detailed analysis of the differences and connections among the attack techniques. Furthermore, we present countermeasures in different learning framework and highlight their advantages and disadvantages. Finally, we discuss the reasons for the feasibility of poisoning attacks and address the potential research directions from attacks and defenses perspectives, separately.

Edge computing is a key enabling technology that meets continuously increasing requirements for the intelligent Internet of Things (IoT) applications. To cope with the increasing privacy leakages of machine learning while benefiting from unbalanced data distributions, federated learning has been wildly adopted as a novel intelligent edge computing framework with localized training mechanism. However, recent studies found that the federated learning framework exhibits inherent vulnerabilities on active attacks, and poisoning attack is one of the most powerful and secluded attacks where the functionalities of the global model could be damaged through attackers well-crafted local updates. In this paper, we give a comprehensive exploration of the poisoning attack mechanisms in the context of federated learning. We first present a poison data generation method, named Data_Gen, based on the generative adversarial networks (GAN). This method mainly relies upon the iteratively updated global model parameters to regenerate samples of interested victims. Secondly, we further propose a novel generative poisoning attack model, named PoisonGAN, against the federated learning framework. This model utilizes the designed Data_Gen method to efficiently reduce the attack assumptions, and make attacks feasible in practice. We finally evaluate our data generation and attack models by implementing two-types of typical poisoning attack strategies, label-flipping and backdoor, on a federated learning prototype. The experimental results demonstrate that these two attack models are effective in federated learning.

Traffic analysis is the process of monitoring network activities, discovering specific patterns, and gleaning valuable information from network traffic. It can be applied in various fields such as network assert probing and anomaly detection. With the advent of network traffic encryption, however, traffic analysis becomes an arduous task. Due to the invisibility of packet payload, traditional traffic analysis methods relying on capturing valuable information from plaintext payload are likely to lose efficacy. Machine learning has been emerging as a powerful tool to extract informative features without getting access to payload, and thus is widely employed in encrypted traffic analysis. In this paper, we present a comprehensive survey on recent achievements in machine learning-powered encrypted traffic analysis. To begin with, we review the literature in this area and summarize the analysis goals that serve as the basis for literature classification. Then, we abstract the workflow of encrypted traffic analysis with machine learning tools, including traffic collection, traffic representation, traffic analysis method, and performance evaluation. For the surveyed studies, the requirements of classification granularity and information timeliness may vary a lot for different analysis goals. Hence, in terms of the goal of traffic analysis, we present a comprehensive review on existing studies according to four categories: network asset identification, network characterization, privacy leakage detection, and anomaly detection. Finally, we discuss the challenges and directions for future research on encrypted traffic analysis.