Artificial intelligence (AI) is an extensive scientific discipline which enables computer systems to solve problems by emulating complex biological processes such as learning, reasoning and self-correction. This paper presents a comprehensive review of the application of AI techniques for improving performance of optical communication systems and networks. The use of AI-based techniques is first studied in applications related to optical transmission, ranging from the characterization and operation of network components to performance monitoring, mitigation of nonlinearities, and quality of transmission estimation. Then, applications related to optical network control and management are also reviewed, including topics like optical network planning and operation in both transport and access networks. Finally, the paper also presents a summary of opportunities and challenges in optical networking where AI is expected to play a key role in the near future

With the development of the Internet of Things, it has been widely studied and deployed in industrial manufacturing, intelligent transportation, and healthcare systems. The time-series feature of the IoT makes the data density and the data dimension higher, where anomaly detection is important to ensure hardware and software security. However, the traditional anomaly detection algorithm has difficulty meeting this demand, not only in complexity but also accuracy. Sometimes the anomaly can be well reconstructed, resulting in a low reconstruction error. In this paper, we propose a memory-augmented autoencoder approach for detecting anomalies in IoT data, which aims to use reconstruction errors to determine data anomalies. First, a memory mechanism is introduced to suppress the generalization ability of the model, and a memory-augmented autoencoder TSMAE is designed for time-series data anomaly detection. Second, by adding penalties and derivable rectifier functions to loss to make the addressing vector sparse, memory modules are encouraged to extract typical normal patterns, thus inhibiting model generalization ability. Finally, through experiments on ECG and Wafer datasets, the validity of TSMAE is verified, and the rationality of hyperparameter setting is discussed through visualizing the memory module addressing vector.

The industrial Internet of things (IIoT), a new computing mode in Industry 4.0, is deployed to connect IoT devices and use communication technology to respond to control commands and handle industrial data. IIoT is typically employed to improve the efficiency of computing and sensing and can be used in many scenarios, such as intelligent manufacturing and video surveillance. To build an IIoT system, we need a collection of software to manage and monitor each system component when there are large-scale devices. Application programming interface (API) is an effective way to invoke public services provided by different platforms. Developers can invoke different APIs to operate IoT devices without knowing the implementation process. We can design a workflow to configure how and when to invoke target APIs. Thus, APIs are a powerful tool for rapidly developing industrial systems. However, the increasing number of APIs exacerbates the problem of finding suitable APIs. Current related recommendation methods have defects. For example, most existing methods focus on the relation between users and APIs but neglect the valuable relations among the users or APIs themselves. To address these problems, this article studies implicit knowledge in IIoT by using collaborative learning techniques. Considering the increased dimensions and dynamics of IoT devices, we explore the possible relationships between users and between APIs. We enhance the matrix factorization (MF) model with the mined implicit knowledge that are implicit relationships on both sides. We build an ensemble model by using all implicit knowledge. We conduct experiments on a collected real-world dataset and simulate industrial system scenarios. The experimental results verify the effectiveness and superiority of the proposed models.

We propose a cognitive Quality of Transmission (QoT) estimator for classifying lightpaths into high or low quality categories in impairment-aware wavelength-routed optical networks. The technique is based on Case-Based Reasoning (CBR), an artificial intelligence technique which solves new problems by exploiting previous experiences, which are stored on a knowledge base. We also show that by including learning and forgetting techniques, the underlying knowledge base can be optimized, thus leading to a significant reduction on the computing time for on-line operation. The performance of the cognitive estimator is evaluated in a long haul and in an ultra-long haul network, and we demonstrate that it achieves more than 98% successful classifications, and that it is up to four orders of magnitude faster when compared with a non-cognitive QoT estimator, the Q-Tool.

Optics InfoBase is the Optical Society's online library for flagship journals, partnered and copublished journals, and recent proceedings from OSA conferences.

In this paper, a proportional-integral-derivative (PID) controller integrated with a neural network (NN) is proposed to ensure quality of service (QoS) bandwidth requirements in passive optical networks (PONs). To the best of our knowledge, this is the first time an approach that implements a NN to tune a PID to deal with QoS in PONs is used. In contrast to other tuning techniques such as Ziegler–Nichols or genetic algorithms (GA), our proposal allows a real-time adjustment of the tuning parameters according to the network conditions. Thus, the new algorithm provides an online control of the tuning process unlike the ZN and GA techniques, whose tuning parameters are calculated offline. The algorithm, called neural network service level PID (NN-SPID), guarantees minimum bandwidth levels to users depending on their service level agreement, and it is compared with a tuning technique based on genetic algorithms (GA-SPID). The simulation study demonstrates that NN-SPID continuously adapts the tuning parameters, achieving lower fluctuations than GA-SPID in the allocation process. As a consequence, it provides a more stable response than GA-SPID since it needs to launch the GA to obtain new tuning values. Furthermore, NN-SPID guarantees the minimum bandwidth levels faster than GA-SPID. Finally, NN-SPID is more robust than GA-SPID under real-time changes of the guaranteed bandwidth levels, as GA-SPID shows high fluctuations in the allocated bandwidth, especially just after any change is made.