Man And Machine

Fuzzy systems are reducing the 
gap between the two, writes Biplab Das 

Imagine a speeding car, which suddenly comes across a red light. The driver presses the brake, but fails to manage a screeching halt and the car lands straight on the traffic island. Now imagine another car embedded with a smart device which alerts the driver to apply brakes as soon as the light turns red, averting a fatal crash. 

The system that prevents an accident is an example of a fuzzy system. Fuzzy systems are the ones which have the same reasoning rules as those of human beings. The reasoning system used in a fuzzy system is called fuzzy logic. A fuzzy system simply converts those reasoning to action at a rate much faster than a human being. 

The mechanism and application of fuzzy systems were explored in the Science City from February 4 to 6 in the International Conference on Fuzzy Systems. It was organised by the electronics and communication sciences (ECS) division of the Indian Statistical Institute (ISI) and Jadavpur University. 

"A fuzzy system is programmed in such a way that it stores thousands of common sense fuzzy rules," said Prof. James Bezdek from the computer science department of the University of West Florida and a star attraction of the conference. "Everytime a fuzzy system gets information from environment, it stores them in its sensory data section." 

In the next step, the information is recognised by the pattern recognition part of the fuzzy system and the system generates output. Bezdek explained the situation taking the example of the fast-moving car. In the car the 'red' traffic light and the need to stop 'immediately' generate two fuzzy sets. These sets then yield outputs in the decision making section. This is the defuzzification step. Relying on this final output, a fuzzy machine takes prompt action. 

"Fuzzy models are widely applied in washing machines, televisions, vacuum cleaners and train controllers," Bezdek said. In a washing machine the fuzzy system finds the size and texture of the wash load and uses a pulsing light sensor to measure the dirt in the wash water. 

Prof. Nikhil Ranjan Pal of ISI pointed out a major drawback of the fuzzy system. "A fuzzy system is not as flexible and efficient as a human brain," he said, "however, it is possible to overcome this difficulty with the application of artificial neural network (ANN)." 

"Implementation of ANN increases learning ability of a machine," said Prof. C. T. Lin, a control engineer from the National Chiao-Tung University, Taiwan. Lin talked about the neurofuzzy system, in which a single silicon chip receives information and yields output like a human neuron. When those silicon chips are interconnected they give rise to multilayer neurofuzzy system. 

Pal described a handy tool - self organizing feature mapping (SOFM) - to detect and correct the errors of a neurofuzzy system. This tool is extremely effective when a neurofuzzy system processes information in a wrong way. 

Genetic algorithm is another tool that increases the efficacy of a fuzzy system. The idea of genetic algorithm comes from the exchange of gene segments in chromosomes. Just as chromosomes exchange gene segments between one another, genetic algorithms exchange pieces of information to trigger a precise action in a fuzzy genetic or neurogenetic system, said Pal. The information is converted and stored in the form of binary digits by genetic algorithms. This is the concept of soft computing, Pal pointed out. 

Raghu Krishnapuram from the IBM India Research Lab of Indian Institute of Technology, New Delhi, spoke on the role of soft computing in e-commerce. "Many of the computational tasks in e-commerce involve parameters like trustworthiness of trading partners, brand loyalty of customers and product preferences of buyers, and these cannot be described precisely," he said. According to him, most of the customer attributes such as 'quality consciousness' and 'caloric consciousness' are inherently fuzzy. "Also customer segments are not predictable. Thus, fuzzy sets can play a major role in building a customer profile," he said. 

Jungwon Ryu and S. B. Cho from the department of computer science of Yoseni University, Korea, discussed how soft computing helped in detecting the expression of few genes in blood cancer. Out of 7129 genes, only 25 play important roles in expression of cancer genes in acute lymphoblastic leukaemia and acute myeloid leukaemia, said Cho. Soft computation makes detection of the cause of the disease easier, he said. Fuzzy system bolstered by neural network can increase the precision of aircraft landing, said Silviu Ionita and Emil Sofron at the electronics department of University of Petesti, Romania. 

Research is going on to make fuzzy systems advanced enough so that human language and gestures could be understood by a machine. Here, important cues come from human brain. Michio Sugeno of the Brain Science Institute, Japan, mentioned that it is possible to create a new system which understands a language. "With soft computing, it is possible to make a robot understand not only a language, but also recognise human gestures, voice and facial expressions," said Z. Zenn Bien, an electrical engineer collaborating with the Human-friendly Welfare Robot System Engineering Research Centre, Korea. 

Such advancement in machine intelligence raises a cliché question. Will robots dominate the earth? At least, burgeoning research in soft computing indicates that the gulf between man and machine is diminishing.

    February 11, 2002.