1
Fundamentals of Soft Computing

Soft computing is a vital tool used for performing several computing operations. It uses one or more computational models or techniques to generate optimum outcomes. To understand this concept, let us first clarify our idea about computation. In any computation operation, inputs are fed into the computing model for performing some operations based on which results are accordingly produced. In the context of computing, the input provided for computation is called an antecedent, and the output generated is called the consequence. Figure 1.1 illustrates the basics of any computing operation where computing is done using a control action (series of steps or actions). Here, in this example, the control action is stated as p?=?f(q), where "q" is the input, "p" is the output, and "f" is the mapping function, which can be any formal method or algorithm to solve a problem.

Hence, it can be concluded that computing is nothing but a mapping function that helps in solving a problem to produce an output based on the input provided. The control action for computing should be precise and definite so as to provide accurate solution for a given problem.

1.1 Introduction to Soft Computing

Many a time, it has been noticed that no fixed solution can be found for a computationally hard task. In such a case, a precisely stated analytical model may not work to produce precise results. For this, the soft computing approach can be used that does not require a fixed mathematical modeling for problem solving. In fact, the uniqueness and strength of soft computing lie in its superpower of fusing two or more soft computing computational models/techniques to generate optimum results.

The concept of soft computing was evolved by Prof. Lofti A. Zadeh (University of California, USA) in the year 1981. Soft computing, as described by Prof. Zadeh, is "a collection of methodologies that aim to exploit the tolerance for imprecision and uncertainty to achieve tractability, robustness, and low solution cost." Prof. Zadeh also emphasized that "soft computing is likely to play an increasingly important role in many application areas, including software engineering. The role model for soft computing is the human mind." Soft computing mimics the notable ability of the human mind to reason and make decisions in an environment of improbability and imprecision. The principal components of soft computing include fuzzy logic, neurocomputing, and probabilistic reasoning (PR).

Figure 1.1 Basic concepts of computing.

If you are wondering in which areas soft computing is being used in our day-to-day lives, the simplest and most common examples include kitchen appliances (rice cookers, microwaves, etc.) and home appliances (washing machines, refrigerators, etc.). Soft computing also finds its dominance in gaming (chess, poker, etc.), as well as in robotics work. Prominent research areas such as data compression, image/video recognition, speech processing, and handwriting recognition are some of the popular applications of soft computing.

1.2 Soft Computing versus Hard Computing

If we consider computing from the perspective of computer science, it is considered a certain task that can be accomplished using computers. Such computing may require certain software or hardware systems to accomplish the task(s) and to derive a certain outcome or output. To understand this easily, let us take a simple example of a self-driving car (named, say, Ziva). Now, the car Ziva is given the instructions to start moving (say, from point A) and to arrive at a destination point B. To accomplish this task, two possible cases can be considered, as discussed below:

Case A: The car Ziva uses a software program to make movement decisions. The path coordinates for movement decisions are already included in the software program with the help of which Ziva can take a predefined path to arrive at its destination. Now, suppose, while moving, Ziva encounters an obstacle in the path. In such a case, the software program can direct it to move to either to the right, or to the left, or to take a back turn. In this case, the self-driving car is not modeled to identify the nature and complexity of the obstacle to make a meaningful and proper decision. In this situation, the computation model used for the car is deterministic in nature, and the output is also concrete. Undoubtedly, there is less complexity in solving the problem, but the output is always fixed due to the rigidness of the computation method.

Case B: The car Ziva uses a software program to make movement decisions. However, in this case, the complexity of the program is more compared to the complexity of the program defined in Case A. This is so as the car is much more involved in complex decision-making. Ziva can now mimic the human brain in making decisions when any kind of obstacle is met in between its travel.

Ziva, first of all, assesses the type of the obstacle, then decides whether it can overcome the obstacle by any means, and finally, it keeps track of if any other alternate path can be chosen instead of overcoming the obstacle found in the same path. The decision to be taken by Ziva is not very crisp and precise, as there are many alternative solutions that can be followed to reach destination point B. For example, if the obstacle is a small stone, Ziva can easily climb up the stone and continue on the same path, as it will lead to a computationally less-expensive solution. However, if the obstacle is a big rock, Ziva may choose an alternative to choose another path to reach the destination point.

Case C: Now, let us consider Case C, in which the software program is written to let the self-driving car reach its destination by initially listing out all the possible paths available to reach from source A to destination B. For each path available, the cost of traveling the path is calculated and accordingly sorted to reach at the fastest time possible. Finally, the optimum path is chosen, considering the minimum cost as well as considering avoidance of any major obstacle. It can be realized that Case C appends both the cases of Case A and Case B to inherit approaches from both cases. It also adds some functionalities to tackle complex scenarios by choosing an optimum decision to finally reach destination point B.

The above three cases can be summarized (as listed in Figure 1.2) to check the points of differences among each of these cases. From each of the above three cases, it can be observed that the nature of computation in each of the three cases is not similar.

Notice that emphasis is given on reaching the destination point in the first case. As the result is precise and fixed, the computation of the Case A type is termed hard computing. Now, in the second case, the interest is to arrive at an approximate result, as a precise result is not guaranteed by this approach. The computation of the Case B type is termed soft computing. The third case inherits the properties of both Case A and Case B, and this part of computing is referred to as hybrid computing. Thus, computing in perspective of computer science can be broadly categorized, as shown in Figure 1.3.

Figure 1.2 Summarization of three varying cases of a self-driving car.

Figure 1.3 Classification of computing (in computer science).

The choice on which classification of computing should be used relies mainly on the nature of the problem to be solved. However, it is important that before choosing any of the computing techniques for problem solving, we should be clear about the main differences between hard computing and soft computing. Table 1.1 lists a few notable differences between hard computing and soft computing to deal with real-world problems.

The points of differences listed in Table 1.1 clear out the fact that soft computing methods are more suitable for solving real-world problems in which ideal models are not available. To name a few applications that may be solved using soft computing techniques include signal processing, robotics control, pattern recognition, business forecasting, speech processing, and many more. Recent research has given a lot of importance to the field of computational intelligence (CI). While traditional artificial intelligence (AI) follows the principle of hard computing, CI follows the principle of soft computing.

1.3 Characteristics of Soft Computing

As we understood that soft computing can deal with imprecision, partial truth, and uncertainty, its applications are varied, ranging from day-to-day applications to various applications related to science and engineering. Some of the dominant characteristics of soft computing are listed in Figure 1.4, and a brief discussion on each of these characteristics is given next:

Table 1.1 Important points of differences between soft computing and hard computing.