Computational Intelligence: An Introduction

Introduction

 

An extensive research yet continuous have been conducted over human intelligence over a rich period of time. In the 20th century, invention of computer systems had facilitated the progress for constructing and investigating systems that were exhibiting behaviour/attributes/features conventionally associated with intelligence. Later on, emerging science and technology or engineering disciplines were termed as the field of Artificial Intelligence. 

 

Though traditional AI converges to symbolic description and manipulation (as reasoning) in a top-down manner, it implies that the structure of a problem- environment, realm context, is  inspected earlier than an intelligent system is constructed depending upon this particular structure. 

 

(Also read: History of Artificial Intelligence)

 

Presently, AI has covered almost all aspects of approaches to human engineered intelligent systems to analyze systems, derive conclusions and determine required actions for a given structure.

 

Evidence shows that there are enormous groups of alternative approaches to recognize the intelligent behavior or feature of something, several properties of being non-symbolic are proposed operating in bottom-up manner where structure originates from an unorganized source instead of emerging from above. These groups are neural networks (deep learning), fuzzy systems, generating algorithms, evolutionary computation and were grouped under the platform of computational intelligence.

 

This discussion will give a brief over computational intelligence (CI), and different opinions about the young field of CI. We will also discuss the fundamental approaches used in this technology in detail.

 

 

Understanding Computation Intelligence

 

In layman terms, computational intelligence is the systematic study of designing of intelligent agents /systems. 

 

An agent can be anything that acts in particular environment and does something, an intelligent agent is a system that acts intelligently- 

 

• It behaves appropriately according to its circumstances and goals, 

• It has feasibility and can adapt in changing environments and objectives,  

• It acquires from experiences and makes selective decisions, providing adequate limitations and computations.

 

In particular, computational intelligence (CI) is the theory, design, application and development of computational paradigms that are biologically and linguistically inspired. 

 

In earlier times, these paradigms were based on three pillars of CI, i.e, Neural Network, Fuzzy Systems and Evolutionary Computation. However, with the development of advanced computation approaches, many nature motivated computing paradigms have evolved.

 

(Also read: When Artificial Intelligence (AI) And Neuroscience Meet)

 

Being an evolving field, computational intelligence involves different computing paradigms such as ambient intelligence, artificial life, artificial endocrine networks, social reasoning, cultural learning, artificial hormone networks, etc. 

 

Computational intelligence puts a major convergence in generating productive intelligent systems such as games and cognitive development systems. (From)

 

Also known as soft computing, CI is a form of computing model based on the methods through which humans learn. The technology uses multiple branches of science (math and logics) in order to develop machine learning algorithms.

 

As computers acquire from procedures depending on logic and science they become more matured and more intelligent, and the process is completely different from artificial intelligence over the perception that CI takes the growth/progress of a system into consideration where AI employs boolean values to attain certain learnings that CI doesn’t.  

 

Computational intelligence has the potential to solve real-life complex problems and to make intelligent-probabilistic decisions. Due to this, CI becomes immensely popular in cases of organising industry procedures, disease diagnostics, video games visualization, intelligent robots, smart chatbots, automated vehicles, translational systems such as Alexa and Siri that can understand human language and perform accordingly.

 

Expanding swiftly, CI has some amazing approaches such as evolutionary computation and learning theory nourishing it to be competitive for a broad bandwidth of real-life problems. We will learn about these approaches in the next section.

 

``... Computational intelligence is defined as a methodology involving computing (whether with a computer, wetware, etc.) that exhibits an ability to learn and/or deal with new situations such that the system is perceived to possess one or more attributes of reason, such as generalisation, discovery, association, and abstraction. The output of a computationally intelligent system often includes predictions and/or decisions. Put another way, computational intelligence comprises practical adaptation concepts, paradigms, algorithms, and implementations that enable or facilitate appropriate actions (intelligent behaviour) in complex and changing environments." A particular aspect of Bezdeks view

 

Approaches used in Computation Intelligence

---

Computational intelligence techniques, Source

---

• Neural Networks

 

Imitating the human brain as a main source of motivation, artificial neural networks are information-processing paradigms that are parallelly distributed networks having capability to learn and understand from example. 

 

For example, the neural network system learns through getting different images as part of examples, being fed regarding different information of those images and thereby experiencing differentiation amid several images by intelligent learning. 

 

Commonly used neural networks are feedforward neural networks, recurrent neural networks, convolutional neural networks, etc. 

 

(Must check: Applications of neural networks)

 

Advantages of employing artificial neural networks are following;

 

• An accurate portrayal of ANN can be used to model non-linearity of any degree and is very efficient in resolving nonlinear problems.

• ANN doesn’t demand to have priori knowledge of the systems/models.

• ANN has strong capability of self-learning and direct processing of data, henceforth, can handle situations when inadequate information is available or data is corrupted.

• ANNs are fast and robust making the learning ability swift to process.

 

• Fuzzy Systems

 

Inspired from human language, fuzzy systems address the linguistic ambiguity and hence solve unpredictable problems that are dependent on generalization of traditional logic systems that leads in performing approximate reasoning.

 

Computational intelligence allows systems to evolve and learn with imprecise and incomplete knowledge. In particular, CI has different phases of information & processing without having full understanding. For example, fuzzy logic can be designed via presenting multiple images to a computer system and make it capable to differentiate amid images/content.

 

This field of research emcompasses fuzzy sets and systems, fuzzy clustering and classification, fuzzy controllers, linguistic summarization, fuzzy neural networks, etc.

 

Fuzzy systems are aimed to overcome inaccuracy and uncertainty, they hold the proficiency to deal with large time delay, time variations and complexity and nonlinear processing.

 

Replicating the way a human being thinks logically to handle fuzzy situations/ information, fuzzy theory is a suitable approach to analyze complex large-scale systems qualitatively. 

 

For instance, typically, the association between a disease and symptom is complicated to explain via precise mathematical models due to the complexity of engineering practices, therefore applying fuzzy theory in diagnosis can be effective as the method closely related to human thinking behaviour and linguistic expression and holds the capability to express knowledge and understanding.

 

(Recommended reading: Branches of Artificial Intelligence)

 

• Evolutionary Computation

 

Following biological evolution, evolutionary computation deals with optimization based problems by virtue of producing, inspecting and modifying a mass of possible solutions. 

 

Considering exciting concepts from evolutionary theory along with natural selection and swarm intelligence, computational intelligence can be applied to computing systems. 

For example, analysts examine multiple animal species to understand how a species acts together in a group and individually. With no precise guidance, still a species maintains intelligence. 

 

Identically, a computing system could present ubiquitous computing intelligence and decision-making capabilities effectively without involvement of a human resource as intelligence. 

 

This system includes genetic algorithms, evolutionary programming, evolution strategies, genetic programming, differential evolution, evolutionary hardware, multi-purpose objective optimization and many more methods.

 

In computational intelligence, few biological evolutionary theories including reproduction or mutation are accounted for framing artificial intelligence more powerful than earlier to deal with real-life problems. 

 

Over the time, several evolution theories have been proposed and hence more number of evolutionary algorithms have been produced. For example, “ the survival of the fittest” originated from Darwin evolutionary theory and fitness function remains the part of genetic algorithms.

 

In computational intelligence, evolutionary computation is broadly employed for optimization problems in order to improve the present state and the progessive problems in order to anticipate future states.

 

(Suggest blog: Deep learning algorithms)

 

• Learning Theory

 

Indicating the philosophy and process of learning, learning theory signifies the way how a learner assembles information, processes and memorizes it to make effective decision-making over complex problems. 

 

Aimed to inspect/understand different learning techniques, learning theory focuses on multiple ways through which the process of learning can take place. 

 

For example, cognitive learning is part of learning theory. 

 

(Related reading: Types of Artificial Intelligence)

 

• Probabilistics Methods

 

Being non-constructive and non-deterministic processes, probabilistic methods are hugely deployed in order to provide the existence of an object.

 

For example, a group has multiple objects where each object has been assigned certain properties. Now if all the objects don’t have one specific property and then one random object is selected from the group. For this particular object, the probability of having that specific probability is certainly zero. 

 

Similarly, if the probability of any random object for any specific property is less than the one, there is some possibility of the existence of one or more objects not having that specific property. (From)

 

 

Conclusion

 

Computational intelligence is a novel and advanced tool for solving complex problems that are challenging to address with conventional techniques. 

 

In recent years, many new developments have been reported that lead to many applications in multiple areas from engineering, finance, social and biomedical sciences. In particular, computational intelligence has increasingly deployed in biomedical and human evolutionary biology to study the behaviour and complexity of biological systems.

 

We have learned that computational intelligence incorporates approaches fundamentally based on neural networks, fuzzy logics practices, evolutionary algorithms, support vector machine and similar approaches blending two or more techniques. Such methods can be useful in addressing multiple computational and diverse problems.