Machine learning (ML) is fundamental to defining how a machine can learn from datasets and present this information in the way the user desires. This brings two separate paradigms together--computing and statistics. We have experienced the evolution of ML in various forms by seeing and utilizing the processing and storage power of machines. The common objective, however, revolves around issues related to managing datasets, retrieving information through analyses, presenting information in a particular manner, gathering knowledge and intelligence, and so on. Various approaches to such issues brought us tools like artificial neural networks (ANN), artificial intelligence (AI), ML, and natural language processing (NLP). Statistical methods evolved as computing power grew, keeping the fundamental challenges intact to ensure the authenticity of data sources, confidence (P-values, correlation and covariance errors, and so on) in manipulating data, and the determination of possible inferences to support decision-making processes.

Locating data sources and collecting data in computing have been very advantageous because of the evolutions in device-centric data management principles. In contemporary terms, the Internet of things (IoT) has paved the way for data collection in various modes with or without human interventions (offline, online, and real-time). Having accepted this relationship between computing and statistics through ML, it is important to explore the usefulness of the outcomes. The outcomes have also evolved over time by introducing various artifacts in the areas of expert systems, decision support systems, data warehousing, data mining (of course with data marts), business intelligence, search engines, and so on. It is worth noting that ML has provided an autonomous environment for users to support the decisions through supervised, semi-supervised, unsupervised, and reinforcing learning environments, giving scope to transform this experience to lifelong learning that a machine can manage. This is perhaps possible with progressive transformation in computing, storage, and retrieval resources that machines can provide to meet the demand.

This book exposes readers to the area of lifelong machine learning (LML), or simply lifelong learning (LL). It discusses very succinctly the evolution of LL, though it does not adequately address the need to drop the word “machine” from LML. The book also adequately discusses the role of LML in supporting users in making more efficient and intelligent decisions. The book focuses on updating readers on (1) previous and current research on supervised methods, (2) unsupervised method lifelong models, (3) lifelong information extraction, (4) lifelong semi-supervised learning, and (5) lifelong reinforcement learning. During such discussions, the book takes care to discuss learning paradigms (like transfer learning and multitask learning) that are closely related to, but different from, LML. The book also lists various challenges that the LML paradigm faces.

A few basic questions, however, remain unaddressed. First, how has LML evolved to be termed LL (without “machine”)? Second, how could the LL paradigm support the decision-making processes for a user in a more intelligent manner than expert systems, data mining techniques, and business intelligence (BI) tools? Third, can LL, ML, and LML converge to remain in sync with learning perceived by humans? Finally, how is LL poised to address future challenges related to IoT and big data supported scenarios with optimization so that human-computer interfaces (HCI) could be improved upon? The book also provides scope for presenting and addressing real-life problems through benchmarking so that industry can learn from the results obtained.

These discussions notwithstanding, the book is very useful for researchers.