Recognition is an important process of situation analysis in military command and control systems. This memorandum focuses on recognition tasks that must rely on data that evolve in time. This type of problem is known as spatiotemporal pattern recognition. The document provides an objective review of supervised learning approaches for spatiotemporal data. The list is non-exhaustive but provides a variety of popular methods presented in the literature. The spatiotemporal pattern recognition methods presented in this review are sliding windows, hidden Markov models and conventional recurrent neural networks, including different training methods such as backpropagation through time and real-time recurrent learning. It also covers Long-Short Term Memories and the reservoir computing approach, where Echo states networks and Liquid state machines are presented as the main reservoir methods. The review explains and discusses the concepts and algorithms behind the different techniques.
