CS440 Computer Science Seminar

Introduction to Evolutionary Computing

Adaptation to environment

Traveling salesman problem• A salesperson must visit clients in different cities, and

then return home. What is the shortest tour through those cities, visiting each one once and only once?

• No known algorithms are able to generate the best answer in an amount time that grow only as a polynomial function of the number of elements (cities) in the problem.

• Belongs in the NP-hard class of problems, where NP stands for non-deterministic polynomial. For 100 cities, there are over 10155 different possible paths through all cities. The Universe is only 1018 seconds old!

Evolution Algorithm

Steps of evolutionary approach to discovering solutions

• Choosing the solution representation

• Devising a random variation operator

• Determining a rule for solution survival

• Initialization the population

Solving the traveling salesman problem: 1. Solution representation, 2. Devising random variation operator

Solving the traveling salesman problem: 3. Determining the rule for solution survival, 4. Initialize the population

• Rule for survival: survival of the fittest—the least total distance traveled.

• Initial population: in this case, chosen completely at random from the space of possible solutions.

The best result of the 1st generation for the 100-city traveling salesman problem

The best result of the 500th generation for the 100-city traveling salesman problem

The best result of the 1000th generation for the 100-city traveling salesman problem

The best result of the 4000th generation for the 100-city traveling salesman problem

Drug design using evolutionary algorithm

Evolutionary algorithm in high-level chess game

To probe further

• What is revolutionary computation, IEEE Spectrum, Feb. 2000

• How to solve It: Modern Heuristics, Zbigniew Michalewicz, Springer, 2000

• Evolution, Neural Networks, Games, and Intelligence, Kumar and Fogel, Proceedings of IEEE Vol. 87, no 9, pp. 1471-96, Sept. 1999