Search

Machine learning models are established with a variety of data collected from individual users who are concerned about their privacy. Various cloud service providers (e.g., Amazon, Google, Alibaba) commercialize their models by selling them or providing limited access via APIs. However, existing studies have shown that such cloud-based models are susceptible to training data inference attacks (i.e., membership inference attacks, attribute inference attacks, and model inversion attacks) since the trained models remember information about the training data. In this article, we perform an exhaustive investigation on the current training data inference attacks against cloud-based models. According to the attack scenario, we divide the existing inference attacks into two categories: centralized and collaborative learning scenarios. We first give a comprehensive review of attacks in each category and then give a qualitative comparison of the existing attacks. Further, we quantitatively compare the performance of different attack strategies by experiments. Last but not least, we pinpoint some open issues and potential future directions that need to be investigated in this area. [ABSTRACT FROM AUTHOR]

Wireless systems continue to go toward higher carrier frequencies, including terahertz bands, to take advantage of higher bandwidth channels. At the same time, antenna arrays remain important with continued increases in array elements. However, the power consumption of RF and digital circuits can increase proportionally to both the amount of signal bandwidth and the number of antennas. The use of one-bit analog-to-digital converters at the receiver is a cost- and power-efficient solution for wideband and/or massive antenna wireless systems. The nonlinearity of one-bit received signals brings challenges in physical-layer (PHY) design at the receiver. At the same time, the binary nature of these signals opens new opportunities for artificial intelligence (AI)-based PHY design. This article covers recent progress in incorporating AI into the design of classical PHY techniques and emerging studies on establishing AI-inspired frameworks that fundamentally replace classical model-driven techniques with data-driven AI techniques. It concludes with a discussion, including practical challenges and future research directions. [ABSTRACT FROM AUTHOR]

Model-free techniques, such as machine learning (ML), have recently attracted much interest toward the physical layer design (e.g., symbol detection, channel estimation, and beamforming). Most of these ML techniques employ centralized learning (CLK) schemes and assume the availability of datasets at a parameter server (PS), demanding the transmission of data from edge devices, such as mobile phones, to the PS. Exploiting the data generated at the edge, federated learning (FL) has been proposed recently as a distributed learning scheme, in which each device computes the model parameters and sends them to the PS for model aggregation, while the datasets are kept intact at the edge. Thus, FL is more communication-efficient and privacy-preserving than CL and applicable to the wireless communication scenarios, wherein the data are generated at the edge devices. This article presents the recent advances in FL-based training for physical layer design problems. Compared to CL, the effectiveness of FL is presented in terms of communication overhead with a slight performance loss in the learning accuracy. The design challenges, such as model, data, and hardware complexity, are also discussed in detail along with possible solutions. [ABSTRACT FROM AUTHOR]

Random access channel (RACH) procedures in modern wireless communications are generally based on multi-channel slotted-ALOHA, which can be optimized by flexibly organizing devices' transmission and retransmission. However, due to the lack of information about the traffic generation statistics and the occurrence of random collision, optimizing RACH in an exact manner is generally challenging. In this article, we first summarize the general structure of optimization for different RACH schemes, and then review existing RACH optimization methods based on machine learning (ML) and non-ML techniques. We demonstrate that the ML-based methods can better optimize RACH schemes compared to non-ML-based methods due to their capability in solving high-complexity long-term optimization problems. To further improve the RACH performance, we introduce a decoupled learning strategy (DLS) for access control optimization, which individually executes two sub-tasks: traffic prediction and access control configuration. In detail, the traffic prediction relies on an online supervised learning method adopting recurrent neural networks that can accurately capture traffic statistics over consecutive frames, while the access control configuration uses either a non-ML-based controller or a cooperatively trained deep reinforcement learning (DRL)-based controller selected depending on the complexity of different random access schemes. Numerical results show that the DLS optimizer considerably outperforms conventional DRL optimizers in terms of higher training efficiency and better access performance. [ABSTRACT FROM AUTHOR]

Machine learning models have achieved state-of-the-art performance in various fields, from image classification to speech recognition. However, such models are trained with a large amount of sensitive training data, and are typically computationally expensive to build. As a result, many cloud providers (e.g., Google) have launched machine-learning-as-a-service, which helps clients benefit from the sophisticated cloud-based machine learning models via accessing public APIs. Such a business paradigm significantly expedites and simplifies the development circles. Unfortunately, the commercial value of such cloud-based machine learning models motivates attackers to conduct model extraction attacks for free use or as a springboard to conduct other attacks (e.g., craft adversarial examples in black-box settings). In this article, we conduct a thorough investigation of existing approaches to model extraction attacks and defenses on cloud-based models. We classify the state-of-the-art attack schemes into two categories based on whether the attacker aims to steal the property (i.e., parameters, hyperparameters, and architecture) or the functionality of the model. We also categorize defending schemes into two groups based on whether the scheme relies on output disturbance or query observation. We not only present a detailed survey of each method, but also demonstrate the comparison of both attack and defense approaches via experiments. We highlight several future directions in both model extraction attacks and its defenses, which shed light on possible avenues for further studies. [ABSTRACT FROM AUTHOR]

Machine learning has recently drawn much attention for a variety of applications, thanks to its good performance in identification, recognition, and regression problems. One such important application is V2V communication propagation channel research. In this article, the challenges and opportunities of machine-learning-based data processing techniques for evaluation of V2V channel measurements are presented. This article reviews some state-of-the-art applications including identification of channel line-of-sight situations, tracking of MPCs, and MPC clustering. The data obtained with these methods form, inter alia, the basis for accurate channel models. Furthermore, some challenges of machine-learning-based data processing for V2V channel research are discussed as basis for future studies. [ABSTRACT FROM AUTHOR]

ML has been increasingly adopted in wireless communications, with popular techniques, such as supervised, unsupervised, and reinforcement learning, applied to traffic classification, channel encoding/ decoding, and cognitive radio. This article discusses a different class of ML technique, stochastic online learning, and its promising applications to MEC. Based on stochastic gradient descent, stochastic online learning learns from the changes of dynamic systems (i.e., the gradient of the Lagrange multipliers) rather than training data, decouples tasks between time slots and edge devices, and asymptotically minimizes the time-averaged operational cost of MEC in a fully distributed fashion with the increase of the learning time. By taking the widely adopted big data analytic framework MapReduce as an example, numerical studies show that the network throughput can increase by eight times through adopting stochastic online learning as compared to existing offline implementations. [ABSTRACT FROM AUTHOR]

With the popularity of smart devices, multi-access edge computing (MEC) has become the mainstream of dealing with big data in heterogeneous Internet of Things (H-IoT). MEC makes full use of the computing power of edge nodes, which greatly reduces the computing pressure of data centers, and brings great convenience to the storage and processing of big data. However, it is easy to become the object of hacker attacks due to the lack of centralized management of distributed nodes. Once these nodes are compromised, a series of privacy issues can happen. In this article, we first overview the architecture of MEC for H-IoT. The MEC covers three-level advanced functional entities, including moblie edge (ME) system-level, ME host-level and ME network- level. Second, we draw our attention to the privacy issues in the MEC, especially in data aggregation and data mining. In addition, we consider machine learning privacy preserving as a case study in the application of MEC. Simulation results are shown to reveal the feasibility of the proposed method. Finally, we propose open issues for future work. [ABSTRACT FROM AUTHOR]

Over the last decade, a significant amount of effort has been invested on architecting agile and adaptive management solutions in support of autonomic, self-managing networks. Autonomic networking calls for automated decisions for management actions. This can be realized through a set of pre-defined network management policies engineered from human expert knowledge. However, engineering sufficiently accurate knowledge considering the high complexity of today's networking environment is a difficult task. This has been a particularly limiting factor in the practical deployment of autonomic systems. ML is a powerful technique for extracting knowledge from data. However, there has been little evidence of its application in realizing practical management solutions for autonomic networks. Recent advances in network softwarization and programmability through SDN and NFV, the proliferation of new sources of data, and the availability of lowcost and seemingly infinite storage and compute resource from the cloud are paving the way for the adoption of ML to realize cognitive network management in support of autonomic networking. This article is intended to stimulate thought and foster discussion on how to defeat the bottlenecks that are limiting the wide deployment of autonomic systems, and the role that ML can play in this regard. [ABSTRACT FROM PUBLISHER]