Learnable Evolution Model

Abstract: This paper addresses the problem of improving the accuracy of an hypothesis output by a learning algorithm in the distribution-free (PAC) learning model. A concept class is learnable (or strongly learnable) if, given access to a source of examples of the unknown concept, the learner with high probability is able to output an hypothesis that is correct on all but an arbitrarily small fraction of the instances. The concept class is weakly learnable if the learner can produce an hypothesis that performs only slightly better than random guessing. In this paper, it is shown that these two notions of learnability are equivalent. A method is described for converting a weak learning algorithm into one that achieves arbitrarily high accuracy. This construction may have practical applications as a tool for efficiently converting a mediocre learning algorithm into one that performs extremely well. In addition, the construction has some interesting theoretical consequences, including a set of general upper bounds on the complexity of any strong learning algorithm as a function of the allowed error e.

Abstract: Surrogate-assisted, or meta-model based evolutionary computation uses efficient computational models, often known as surrogates or meta-models, for approximating the fitness function in evolutionary algorithms. Research on surrogate-assisted evolutionary computation began over a decade ago and has received considerably increasing interest in recent years. Very interestingly, surrogate-assisted evolutionary computation has found successful applications not only in solving computationally expensive single- or multi-objective optimization problems, but also in addressing dynamic optimization problems, constrained optimization problems and multi-modal optimization problems. This paper provides a concise overview of the history and recent developments in surrogate-assisted evolutionary computation and suggests a few future trends in this research area.

Abstract: While Lawrence Fogel, John Holland, Ingo Rechenberg and others were the undoubted pioneers of the field we now know as evolutionary algorithms (EA), or evolutionary computation (EC), Ken De Jong’s doctoral thesis of 1975 deserves much of the credit for firing the enthusiasm of several research communities in the practical exploration of these methods. Moreover, as he has taken a very active part in the development of the field through the last 30 years, there could scarcely be anyone better placed to write a book on evolutionary computation. As the subtitle of his book promises, De Jong takes a unified approach. His first 4 chapters carefully explain and differentiate, whilst putting in their historical context, the common aspects of different EC paradigms (evolutionary programming—EP, evolution strategies—ES and genetic algorithms—GA). Chapters 1–4 use clear examples, rather than too many mathematical symbols. They form a truly superb introduction. Any novice coming to EC should come away with an excellent grasp of the basics. In chapter 5 he discusses the different uses to which EAs have been put as problem-solvers. The greater part is devoted to optimization (OPT-EA), with shorter sections on search, machine learning, and automated programming. There is a final, very brief, section on adaptive EAs. In the optimization part, considerable care is taken in the organisation of his material—again, presumably, with the novice in mind. Chapter 6 is the longest, and focuses on EC theory. De Jong carefully builds up a picture of the influences of selection, mutation and recombination on the behaviour of EAs. If you are expecting theory in the sense of a comprehensive, general model with well-understood effects, you will be disappointed. There are equations, but the argument is in fact founded on a series of experiments, whose results are displayed in a series of graphs. That is not to say that the insights gained are incorrect, or