Kumar Chellapilla

Combining mutation operators in evolutionary programming

Traditional investigations with evolutionary programming for continuous parameter optimization problems have used a single mutation operator with a parametrized probability density function (PDF), typically a Gaussian. Using a variety of mutation operators that can be combined during evolution to generate PDFs of varying shapes could hold the potential for producing better solutions with less computational effort. In view of this, a linear combination of Gaussian and Cauchy mutations is proposed. Simulations indicate that both the adaptive and nonadaptive versions of this operator are capable of producing solutions that are statistically as good as, or better, than those produced when using Gaussian or Cauchy mutations alone.

Evolving an expert checkers playing program without using human expertise

An evolutionary algorithm has taught itself how to play the game of checkers without using features that would normally require human expertise. Using only the raw positions of pieces on the board and the piece differential, the evolutionary program optimized artificial neural networks to evaluate alternative positions in the game. Over the course of several hundred generations, the program taught itself to play at a level that is competitive with human experts (one level below human masters). This was verified by playing the best evolved neural network against 165 human players on an Internet gaming zone. The neural networks performance earned a rating that was better than 99.61% of all registered players at the Website. Control experiments between the best evolved neural network and a program that relies on material advantage indicate the superiority of the neural network both at equal levels of look ahead and CPU time. The results suggest that the principles of Darwinian evolution may he usefully applied to solving problems that have not yet been solved by human expertise.

Evolution, neural networks, games, and intelligence

Mathematical games provide a framework for studying intelligent behavior in models of real-world settings or restricted domains. The obstacle comes in choosing the appropriate representation and learning algorithm. Neural networks and evolutionary algorithms provide useful means for addressing these issues. This paper describes efforts to hybridize neural and evolutionary computation to learn appropriate strategies in zero- and nonzero-sum games, including the iterated prisoners dilemma, tic-tac-toe, and checkers. With respect to checkers, the evolutionary algorithm was able to discover a neural network that can be used to play at a near-expert level without injecting expert knowledge about how to play the game. The implications of evolutionary learning with respect to machine intelligence are also discussed. It is argued that evolution provides the framework for explaining naturally occurring intelligent entities and can be used to design machines that are also capable of intelligent behavior.

Genetic Programming 1998: Proceedings of the Third Annual Conference

Proceedings of the Annual Conferences on Genetic Programming. These proceedings present the most recent research in the field of genetic programming as well as recent research results in the fields of genetic algorithms, artificial life and evolution strategies, DNA computing, evolvable hardware, and genetic learning classifier systems.

Evolving computer programs without subtree crossover

An evolutionary programming procedure is used for optimizing computer programs in the form of symbolic expressions. Six tree mutation operators are proposed. Recombination operators such as crossover are not included. The viability and efficiency of the method is extensively investigated on a set of well-studied problems. The evidence indicates that the technique is not only viable but is indeed capable of evolving good computer programs. The results compare well with other evolutionary methods that rely on crossover to solve the same problems.

Multiple sequence alignment using evolutionary programming

Multiple sequence alignment can be used as a tool for the identification of common structure in an ordered string of nucleotides (in DNA or RNA) or amino acids (in proteins). Current multiple sequence alignment algorithms work well for sequences with high similarity but do not scale well when either the length or number of the sequences is large or if the similarity is low. The focus of the paper is to develop an evolutionary programming (EP) algorithm for multiple sequence alignment. An EP method with representation specific variation operators is proposed and tested on several data sets. Comparisons to other algorithms suggests that this algorithm is well suited to the multiple sequence alignment problem.

Inductive reasoning and bounded rationality reconsidered

Complex adaptive systems have historically been studied using simplifications that mandate deterministic interactions between agents or instead treat their interactions only with regard to their statistical expectation. This has led to an anticipation, even in the case of agents employing inductive reasoning in light of limited information, that such systems may have equilibria that can be predicted a priori. This hypothesis is tested here using a simulation of a simple market economy in which each agents behavior is based on the result of an iterative evolutionary process of variation and selection applied to competing internal models of its environment. The results indicate no tendency for convergence to stability or a long-term equilibrium and highlight fundamental differences between deterministic and stochastic models of complex adaptive systems.

Verifying Anaconda's expert rating by competing against Chinook: experiments in co-evolving a neural checkers player

Abstract Since the early days of artificial intelligence, there has been interest in having a computer teach itself how to play a game of skill, like checkers, at a level that is competitive with human experts. To be truly noteworthy, such efforts should minimize the amount of human intervention in the learning process. Recently, co-evolution has been used to evolve a neural network (called Anaconda) that, when coupled with a minimax search, can evaluate checker-boards and play to the level of a human expert, as indicated by its rating of 2045 on an international web site for playing checkers. The neural network uses only the location, type, and number of pieces on the board as input. No other features that would require human expertise are included. Experiments were conducted to verify the neural networks expert rating by competing it in 10 games against a “novice-level” version of Chinook, a world-champion checkers program. The neural network had 2 wins, 4 losses, and 4 draws in the 10-game match. Based on an estimated rating of Chinook at the novice level, the results corroborate Anacondas expert rating.

Revisiting evolutionary programming

Evolutionary programming is a method for simulating evolution that has been investigated for almost 40 years. When originally introduced, the available computing equipment was quite slow and difficult to use as measured by current standards. This paper provides a series of experiments that follow the framework of the original approach from the early 1960s, brought up to date with current computing machinery. A brief review of evolutionary programming and its relationship to other methods of evolutionary computation, specifically genetic algorithms and evolution strategies, is also offered.

A preliminary investigation into evolving modular finite state machines

Evolutionary programming was proposed more than thirty five years ago for generating artificial intelligence. The original experiments consisted of evolving populations of finite state machines (FSMs) for prediction, identification, and control. Since then, all of the studies with FSMs and evolutionary programming have been limited to the evolution of strictly non-modular FSMs. In this study, a modular FSM architecture is proposed and an evolutionary programming procedure for evolving such structures is presented. Preliminary results indicate that the proposed procedure is indeed capable of successfully evolving modular FSMs and that such modularity can result in a statistically significantly increased rate of optimization.