Genetic Programming Theory and Practice XI (Genetic and Evolutionary Computation)
Format: PDF / Kindle (mobi) / ePub
These contributions, written by the foremost international researchers and practitioners of Genetic Programming (GP), explore the synergy between theoretical and empirical results on real-world problems, producing a comprehensive view of the state of the art in GP. Topics in this volume include: evolutionary constraints, relaxation of selection mechanisms, diversity preservation strategies, flexing fitness evaluation, evolution in dynamic environments, multi-objective and multi-modal selection, foundations of evolvability, evolvable and adaptive evolutionary operators, foundation of injecting expert knowledge in evolutionary search, analysis of problem difficulty and required GP algorithm complexity, foundations in running GP on the cloud – communication, cooperation, flexible implementation, and ensemble methods. Additional focal points for GP symbolic regression are: (1) The need to guarantee convergence to solutions in the function discovery mode; (2) Issues on model validation; (3) The need for model analysis workflows for insight generation based on generated GP solutions – model exploration, visualization, variable selection, dimensionality analysis; (4) Issues in combining different types of data. Readers will discover large-scale, real-world applications of GP to a variety of problem domains via in-depth presentations of the latest and most significant results.
binary strings are generated to be ANNs using a Backus-Naur grammar. The ANNs are guaranteed to be functional per sensible initialization (O’Neill and Ryan 2001, 2003). During sensible initialization, an expression tree is created using the specified grammar by randomly selecting grammar rules to construct the tree. The software recursively checks the expression tree to make sure the selected rule would not make the expression tree exceed the maximum depth (Maxdepth) allowed. Half of the
different model objectives that are each treated equally. We have previously used classification accuracy and model size as our two objectives (Moore et al. 2013). Here, we add a third dimension defined by the interaction information measure of interestingness. For a given GP population, models for which there are no better as measured by accuracy, model size and interestingness are selected. This subset of Pareto-optimal models is referred to as the Pareto front. The goal of the present study
Experimental Design and Post-processing The goal of this study was to apply CES to the genetic analysis of Alzheimer’s disease. We first pre-processed the data by estimating the interaction information for all pairs of SNPs as described by Moore et al. (2006). We considered pairs of SNPs that have higher interaction information more interesting. This pre-processed interestingness measure was used as expert knowledge (Attribute EK) in the CES solution modifiers and as an additional axis in a
work. Divergent user requirements and heterogeneous workload characteristics complicate the problem of identifying the best cloud provider for a specific application. The approach presented here is confined to only some of the characteristics and requirements, which we believe constitute a solid base to demonstrate the impact and performance of the prediction and scheduling mechanics. Our current application model focuses on web application workloads, because they have a long lifetime which
(Kotanchek et al. 2007), age layered population structures (Hornby 2006), age fitness pareto optimization (Schmidt and Lipson 2010), and specialized embedded abstract constant optimization (Korns 2010). In this chapter we enhance the previous baseline with a complex multi-island algorithm for modern symbolic regression which is extremely accurate for a large class of Symbolic Regression problems. The class of problems, on which SR is extremely accurate, is described in detail. A definition of