Machine Learning From Distributed, Streaming Data

The field of machine learning has undergone radical transformations during the last decade. These transformations, which have been fueled by our ability to collect and generate tremendous volumes of training data and leverage massive amounts of low-cost computing power, have led to an explosion in research activity in the field by academic and industrial researchers. Unlike many other disciplines, advances in machine learning research are being rapidly adopted by industry and are beginning to disrupt fields ranging from health care [1], journalism [2], and the retail industry [3] to wireless communications [4], supply-chain management [5], and the automotive industry [6].

In many of the up-and-coming applications of machine learning in these and other fields, such as connected and/or autonomous vehicles, smart grids, edge-caching wireless networks, cloud computing, and urban policing, data are increasingly distributed and are also often streaming. Training predictive models in this distributed, streaming setting requires a rethinking of off-the-shelf machine learning solutions. A number of academic and industrial researchers have recognized the need for this in the last few years; the resulting solutions leverage algorithmic and analytical tools from a number of research areas that cut across multiple disciplines [7]–[8][9][10].

Many of these tools, such as stochastic approximation [11], [12], online learning [13], [14], distributed optimization [15], [16], and decentralized computing [17], have been the mainstay of signal processing researchers for more than a few decades. IEEE Signal Processing Magazine (SPM), therefore, is one of the best forums for archiving the latest advances in machine learning from data that are distributed, streaming, or both distributed and streaming and for discussing many of the open challenges that remain to be solved for the broad adoption of machine learning tools across a large number of industries that are expected to routinely deal with large volumes of distributed and/or streaming data sets.