Amit Benbassat

Evolving Lose-Checkers players using genetic programming

We present the application of genetic programming (GP) to the zero-sum, deterministic, full-knowledge board game of Lose Checkers. Our system implements strongly typed GP trees, explicitly defined introns, local mutations, and multi-tree individuals. Explicitly defined introns in the genome allow for information selected out of the population to be kept as a reservoir for possible future use. Multi-tree individuals are implemented by a method inspired by structural genes in living organisms, whereby we take a single tree describing a state evaluator and split it.

EvoMCTS: Enhancing MCTS-based players through genetic programming

We present EvoMCTS, a genetic programming method for enhancing level of play in games. Our work focuses on the zero-sum, deterministic, perfect-information board game of Reversi. Expanding on our previous work on evolving board-state evaluation functions for alpha-beta search algorithm variants, we now evolve evaluation functions that augment the MTCS algorithm. We use strongly typed genetic programming, explicitly defined introns, and a selective directional crossover method. Our system regularly evolves players that outperform MCTS players that use the same amount of search. Our results prove scalable and EvoMCTS players whose search is increased offline still outperform MCTS counterparts. To demonstrate the generality of our method we apply EvoMCTS successfully to the game of Dodgem.

Evolving board-game players with genetic programming

We present the application of genetic programming (GP) to zero-sum, deterministic, full-knowledge board games. Our work expands previous results in evolving board-state evaluation functions for Lose Checkers to a 10x10 variant of Checkers, as well as Reversi. Our system implements strongly typed GP trees, explicitly defined introns, and a selective directional crossover method.

Complexity of Canadian traveler problem variants

The Canadian traveler problem (CTP) is the problem of traversing a given graph, where some of the edges may be blocked a state which is revealed only upon reaching an incident vertex. Originally st ...

Evolving both search and strategy for Reversi players using genetic programming

We present the application of genetic programming to the zero-sum, deterministic, full-knowledge board game of Reversi. Expanding on our previous work on evolving boardstate evaluation functions, we now evolve the search algorithm as well, by allowing evolved programs control of game-tree pruning. We use strongly typed genetic programming, explicitly defined introns, and a selective directional crossover method. We show that our system regularly churns out highly competent players and our results prove easy to scale.

From size perception to counting: an evolutionary computation point of view

The ability to perceive size is shared by humans and animals. Babies present this basic ability from birth, and it improves with age. Counting, on the other hand, is a more complex task than size perception. We examined the theory that the counting system evolved from a more primitive system of size perception (the leading alternative being that the two systems evolved separately). By using evolutionary computation techniques, we generated artificial neural networks (ANNs) that excelled in size perception and presented a significant advantage in evolving the ability to count over those that evolved this ability from scratch. This advantage was observed also when evolving from ANNs that master other simple classification tasks. We also show that ANNs who train to perceive size of continuous stimuli present better counting skills than those that train with discrete stimuli.

More or Less? Two Approaches to Evolving Game-Playing Strategies

We present two opposing approaches to the evolution of game strategies, one wherein a minimal amount of domain expertise is injected into the process, the other infusing the evolutionary setup with expertise in the form of domain heuristics. We show that the first approach works well for several popular board games, while the second produces top-notch solvers for the hard game of FreeCell.

EvoMCTS: A Scalable Approach for General Game Learning

In this paper, we present the application of genetic programming as a generic game learning approach to zero-sum, deterministic, full-knowledge board games by evolving board-state evaluation functions to be used in conjunction with Monte Carlo tree search (MCTS). Our method involves evolving board-evaluation functions that are then used to guide the MCTS playout strategy. We examine several variants of Reversi, Dodgem, and Hex using strongly typed genetic programming, explicitly defined introns, and a selective directional crossover method. Our results show a proficiency that surpasses that of baseline handcrafted players using equal and in some cases greater amounts of search, with little domain knowledge and no expert domain knowledge. Moreover, our results exhibit scalability.

Evolving players that use selective game-tree search with genetic programming

We present the application of genetic programming (GP) to evolving game-tree search in board games. Our work expands previous results in evolving board-state evaluation functions for multiple board games, now evolving a search-guiding evaluation function alongside it. Our system implements strongly typed GP trees, explicitly defined introns, and a selective directional crossover method.

Replicating the Stroop Effect Using a Developmental Spatial Neuroevolution System

We present an approach to the study of cognitive phenomena by using evolutionary computation. To this end we use a spatial, developmental, neuroevolution system. We use our system to evolve ANNs to perform simple abstractions of the cognitive tasks of color perception and color reading. We define these tasks to explore the nature of the Stroop effect. We show that we can evolve it to perform a variety of cognitive tasks, and also that evolved networks exhibit complex interference behavior when dealing with multiple tasks and incongruent data. We also show that this interference behavior can be manipulated by changing the learning parameters, a method that we successfully use to create a Stroop like interference pattern.