This paper presents a theoretical framework for generalization, the first time that generalization is defined and analyzed rigorously in coevolutionary learning, and shows that a small sample of test strategies can be used to estimate the generalization performance.
Abstract:
Coevolutionary learning involves a training process where training samples are instances of solutions that interact strategically to guide the evolutionary (learning) process. One main research issue is with the generalization performance, i.e., the search for solutions (e.g., input-output mappings) that best predict the required output for any new input that has not been seen during the evolutionary process. However, there is currently no such framework for determining the generalization performance in coevolutionary learning even though the notion of generalization is well-understood in machine learning. In this paper, we introduce a theoretical framework to address this research issue. We present the framework in terms of game-playing although our results are more general. Here, a strategy's generalization performance is its average performance against all test strategies. Given that the true value may not be determined by solving analytically a closed-form formula and is computationally prohibitive, we propose an estimation procedure that computes the average performance against a small sample of random test strategies instead. We perform a mathematical analysis to provide a statistical claim on the accuracy of our estimation procedure, which can be further improved by performing a second estimation on the variance of the random variable. For game-playing, it is well-known that one is more interested in the generalization performance against a biased and diverse sample of "good" test strategies. We introduce a simple approach to obtain such a test sample through the multiple partial enumerative search of the strategy space that does not require human expertise and is generally applicable to a wide range of domains. We illustrate the generalization framework on the coevolutionary learning of the iterated prisoner's dilemma (IPD) games. We investigate two definitions of generalization performance for the IPD game based on different performance criteria, e.g., in terms of the number of wins based on individual outcomes and in terms of average payoff. We show that a small sample of test strategies can be used to estimate the generalization performance. We also show that the generalization performance using a biased and diverse set of "good" test strategies is lower compared to the unbiased case for the IPD game. This is the first time that generalization is defined and analyzed rigorously in coevolutionary learning. The framework allows the evaluation of the generalization performance of any coevolutionary learning system quantitatively.
TL;DR: The Elements of Statistics Appendix as mentioned in this paper Theoretically, the classical limit theorem and the theory of infinite-divisible distributions have been studied in the context of random variables and distribution functions.
TL;DR: The paper introduces the concept of a subsuming HOM; one that is harder to kill than the first order mutants from which it is constructed, by definition, subsumed HOMs denote subtle fault combinations.
TL;DR: The proposed algorithm is named DEwSAcc and is based on Differential Evolution algorithm, which is a floating-point encoding evolutionary algorithm for global optimization over continuous spaces based on log-normal self-adaptation of its control parameters and combined with cooperative co-evolution as a dimension decomposition mechanism.
TL;DR: This paper argues that Search Based Software Engineering can provide a solution to this apparent problem of real faults in Higher Order Mutation Testing, citing results from recent work on search based optimization techniques for constructing higher order mutants.
TL;DR: It is proved that longer memory strategies outperform shorter memory strategies statistically in the sense of evolutionary stability and given an example of a memory-two strategy to show how the theoretical study of evolutionary Stability assists in developing novel strategies.
TL;DR: This is the first comprehensive treatment of feed-forward neural networks from the perspective of statistical pattern recognition, and is designed as a text, with over 100 exercises, to benefit anyone involved in the fields of neural computation and pattern recognition.
TL;DR: In this paper, a model based on the concept of an evolutionarily stable strategy in the context of the Prisoner's Dilemma game was developed for cooperation in organisms, and the results of a computer tournament showed how cooperation based on reciprocity can get started in an asocial world, can thrive while interacting with a wide range of other strategies, and can resist invasion once fully established.
TL;DR: The chapter discusses two important directions of research to improve learning algorithms: the dynamic node generation, which is used by the cascade correlation algorithm; and designing learning algorithms where the choice of parameters is not an issue.
TL;DR: A model is developed based on the concept of an evolutionarily stable strategy in the context of the Prisoner's Dilemma game to show how cooperation based on reciprocity can get started in an asocial world, can thrive while interacting with a wide range of other strategies, and can resist invasion once fully established.
Q1. What are the contributions in "Measuring generalization performance in co-evolutionary learning" ?
In this paper, the authors introduce a theoretical framework to address this research issue. The authors present the framework in terms of game-playing although their results are more general. Given that the true value may not be determined by solving analytically a closed-form formula and is computationally prohibitive, the authors propose an estimation procedure that computes the average performance against a small sample of random test strategies instead. The authors perform a mathematical analysis to provide a statistical claim on the accuracy of their estimation procedure, which can be further improved by performing a second estimation on the variance of the random variable. The authors introduce a simple approach to obtain such a test sample through the multiple partial enumerative search of the strategy space that does not require human expertise and is generally applicable to a wide range of domains. The authors investigate two definitions of generalization performance for the IPD game based on different performance criteria, e. g., in terms of the number of wins based on individual outcomes and in terms of average payoff. The authors show that a small sample of test strategies can be used to estimate the generalization performance. The authors also show that the generalization performance using a biased and diverse set of “ good ” test strategies is lower compared to the unbiased case for the IPD game.
Q2. What is the maximum variance for a random variable over a compact interval?
The maximum variance for a random variable over [a, b] is when half of the mass is at a and the other half is at b, i.e., σ2MAX = R 2 X/4 where RX = b−a.
Q3. How many combinations of tables can be used to calculate the proportion of “all cooperate” in S?
Given that there are 22 combinations of such tables, one can easily calculate the proportion of “all cooperate” in S as 22/25 (since the total number of unique combinations of the table is 22 2+1).
Q4. What is the true generalization performance of co-evolutionary learning?
The true generalization performance of co-evolutionary learning is defined as the expected performance of strategy i that is produced after a learning process (co-evolution) against all strategies j in the strategy space S.
Q5. Why do the authors require a larger population size for each partial enumerative search?
The authors require the population size for each partial enumerative search, PS, to be larger than the maximum number of unique strategies that can be obtained from an evolutionary run, i.e., PS > (generation × POPSIZE), for two reasons.
Q6. How can the authors obtain a tighter Chebyshev’s bound?
The authors also show how a strategy’s performance profile with respect to the strategy space, the variance σ2, can be used through a second estimation to obtain a tighter Chebyshev’s bound.