Autonomous Learning Systems: From Data Streams to Knowledge in Real-time

Forewords

Adrian Stoica

Efficient and robust performance in imperfectly known, nonstationary, environments – and this characterizes the vast majority of real-world applications – requires systems that can improve themselves, transcending their initial design, continuously optimizing their parameters, models, and methods. These improvements come predominantly from learning – about the environment, about the ageing self, about the interactions with, and within, the environment, and from the ability to put this learning to use. Batch learning – or at least repeated updating learning from most recent batches, is sufficient only for a limited number of applications. For other applications learning needs to be incremental, to sample level, a learn-or-perish, or at least learn-or-pay (a hefty price) situation. In particular, real-time learning is most critical for bots, virtual or real, agents of the cyberphysical systems that need the agility to swiftly react to virus attacks, or physical robots exposed to hazards while performing search and rescue in disaster areas, or dealing with what is, for now, a largely unpredictable partner: the human. The fast advancement in autonomous systems makes the subject of real-time autonomous learning critically important, and yet the literature addressing this topic is extremely scarce.

Dr Angelov's pioneering book addresses this problem at its core, focusing on real-time, online learning from streaming data on a sample-by-sample basis. It offers ...