1
Cognitive Analytics in Continual Learning: A New Frontier in Machine Learning Research

Renuga Devi T.1, Muthukumar K.2*, Sujatha M.1† and Ezhilarasie R.1

1School of Computing, SASTRA Deemed University, Thanjavur, India

2School of Electrical & Electronics Engineering, SASTRA Deemed University, Thanjavur, India

Abstract

The cognitive system that started with automation has now set its benchmark to reach human-centric intelligence. The slow adoption of cognitive systems is most likely due to its meticulous training process. With cognitive computing as its backbone nowadays, any data can be converted into an asset anytime and anywhere. The complexity of data and its abandonment nature demand the coexistence of many technologies to provide deep insights in a domain. A generic artificial intelligence system built on deep learning and natural language processing evolves into a personalized business partner and a life companion that continuously learns. Combining tremendous power, humanity’s relationship with technology has undergone incredible shifts. The adaptation and embracement have led to a higher level of intelligence augmentation, mainly in decision support and engagement systems, penetrating its need in various fields, especially in the healthcare industry, business-to-business, industrial marketing, autonomous driving, financial services, manufacturing sectors, and as a human assistant in day-to-day activities. The expensive and complex process of using cognitive systems to get complete resolutions for specific business segments on historical static data and dynamic real-time data should be addressed with Hadoop, Spark, NoSQL, and other technologies that are part of cognitive systems besides NLP, AI, and ML. This chapter begins with an understanding of different analytics and the need of the hour, then gradually penetrates to give insights into cognitive systems, design principles, and key characteristics of the system, dwelling in the backbone of cognitive systems and its different learning approaches with some prominent use cases.

Keywords: Cognitive computing, machine learning algorithms, natural language processing, artificial intelligence, cognitive analytics

1.1 Introduction

The cognitive age is a continuous trend of massive technological development. The driving force behind this trend is the developing field of cognitive technology, which consists of profoundly disruptive systems that interpret unstructured data, reason to generate hypotheses, learn from experience, and organically interact with humans. With this technology, the capacity to generate insight from all types of data will be critical to success in the cognitive age.

Cognitive computing is likely most notable for upending the conventional IT view that a technology’s worth reduces with time; because cognitive systems improve as they learn, they actually grow more useful. This trait makes cognitive technology very valuable for business, and many early adopters are capitalizing on the competitive edge it provides. The cognitive era has arrived, not just because technology has matured, but also because the phenomena of big data necessitate it. The goal of cognitive computing is to be able to solve some uncertain real-world issues comparable to those addressed by the human brain [1].

Since its inception in the 1950s, cognitive science has grown at a rapid pace. Furthermore, as a key component of cognitive science, cognitive computing has a significant influence on artificial intelligence and information technology [2]. Computing systems in the past could gather, transport, and store unstructured data, but they could not interpret it. Cognitive computing systems are intended to foster a better “symbiotic relationship” between humans and technology by replicating human reasoning and problem-solving. Cognitive computing simulates the human brain using computerized models. It is accomplished by the combination of the Von Neumann paradigm and neuromorphic computing, which combines analytic, iterative processes with extremely sophisticated logical and reasoning operations in a very short period of time while utilizing very little power.

The excitement around AI equipment has been dubbed a “renaissance of equipment,” as vendors race to manufacture space-explicit or exceptional job-at-hand explicit designs that can fundamentally scale and increase computing productivity [3]. Cognitive systems are probabilistic in nature that hold the capability to adapt and sense the unpredictability and complexity of unstructured input. They analyze that information, organize it, and explain what it means, as well as the reasons for their judgments [4]. Cognitive computing refers to technological platforms that combine reasoning, machine learning, natural language processing, vision, voice, and human computer interaction that replicates the human brain operation and aid in decision-making. The progression of cognitive thought evolves from pure descriptivism through past prediction to prescriptiveness, reflecting a journey from understanding to anticipation and active guidance.

1.2 Evolution of Data Analytics

As we go forward, the graph in Figure 1.1 shows us the benefits that each type of analytics provides.

a) Descriptive Analytics

Acquiring and evaluating facts to explain what has happened. The majority of business reports are descriptive in nature, which is capable of providing historical data summary or explaining differences from one another. Insights from past data are provided in detail by descriptive analytics via data aggregation and data mining but fail to explain the reason behind the insights.

Figure 1.1 Benefits of analytics (source: https://swifterm.com/the-difference-between-descriptive-diagnostic-predictive-and-cognitive-analytics/).

b) Diagnostic Analytics

Diagnostic analytics addresses the reason behind the inference and discovers answers to why questions. The data are compared with past data to identify why the particular situation has happened. This method of data evaluation is useful to uncover data anomalies, determine the relationships within the data, and detect patterns and trends in product market analysis. Some of the diagnostic analytics used by various business firms include data discovery, alarms, drill-down, correlation, drill-up, and data mining. In-depth analysis by experienced demand planners provides assistance for better decision choices. Diagnostic analytics is a reactive process; it helps us only to anticipate the possibility of continuation of the current situation even when used with forecasting.

c) Predictive Analytics

Predictive analytics forms a part of business intelligence that uses predictive and descriptive factors of the available data to forecast and identify the possibility of the occurrence of an unknown pattern in the near future. Predictive analytics is a subset of business intelligence that analyzes and predicts the possibility of an unknown future result using descriptive and predictive factors from the past. It combines analytical techniques, data mining strategies, predictive models, and forecasting methods to assess the possibility of risk and linkages in the current data to perform future predictions. At this point, you are more interested in why something happened than in what happened. It offers proactive market responses.

d) Prescriptive Analytics

Prescriptive analytics combines descriptive, predictive, and diagnostic analysis to create the possibility to make things happen. Beginning with descriptive analysis, which informed us about what has happened, the next stage was to do a diagnostic about why it happened and the next was predictive analysis to predict when it would happen. As a consequence, prescriptive analysis uses business principles and mathematical models on the data to infer future decisions/actions from the current data. Business firms can implement prescriptive analytics in day-to-day transactions only when analytics-driven culture is followed for the entire organization. Larger firms such as Amazon and McDonald’s employ prescriptive analytics to increase revenue and customer experience by increasing their demand planning.

e) Cognitive Analytics

A software that takes all data and analytics and also learns on its own without explicit human direction is cognitive analytics. To achieve this self-learning, cognitive analytics combines advanced technologies like Natural Language Processing (NLP), artificial intelligence algorithms, machine learning and deep learning, semantics, data mining, and emotional intelligence [5]. Using these techniques, the cognitive application would become smarter and repair itself.

Figure 1.2 Conceptual view of cognitive computing [6].

1.3 Conceptual View of Cognitive Systems

Internal components of the cognitive analytics engine are depicted in Figure 1.2 by the large rectangle. To represent and reason with information, many knowledge representation structures are required. A variety of machine learning methods and inference engines are also required. Domain cognitive models encapsulate domain-specific cognitive processes to facilitate cognitive style problem solving. The learning and adaptation component increases system performance by learning from prior encounters with users. In contrast to all...