This paper presents a highly efficient, fully parallelized implementation of the compact genetic algorithm (cGA) to solve very large scale problems with millions to billions of variables. The paper presents principled results demonstrating the scalable solution of a difficult test function on instances over a billion variables using a parallel implementation of cGA. The problem addressed is a noisy, blind problem over a vector of binary decision variables. Noise is added equaling up to a tenth of the deterministic objective function variance of the problem, thereby making it difficult for simple hillclimbers to find the optimal solution. The compact GA, on the other hand, is able to find the optimum in the presence of noise quickly, reliably, and accurately, and the solution scalability follows known convergence theories. These results on noisy problem together with other results on problems involving varying modularity, hierarchy, and overlap foreshadow routine solution of billion-variable problems across the landscape of search problems.

This paper presents a learning methodology based on a substructural classification model to solve decomposable classification problems. The proposed method consists of three important components: (1) a structural model that represents salient interactions between attributes for a given data, (2) a surrogate model which provides a functional approximation of the output as a function of attributes, and (3) a classification model which predicts the class for new inputs. The structural model is used to infer the functional form of the surrogate and its coefficients are estimated using linear regression methods. The classification model uses a maximally-accurate, least-complex surrogate to predict the output for given inputs. The structural model that yields an optimal classification model is searched using an iterative greedy search heuristic. Results show that the proposed method successfully detects key variable interactions in hierarchical problems, group them in linkages groups, and build maximally accurate classification models. The initial results on non-trivial hierarchical test problems indicate that the proposed method holds promise and have also shed light on several improvements to enhance the capabilities of the proposed method. 1

This paper presents a learning methodology based on a substructural classification model to solve decomposable classification problems. The proposed method consists of three important components: (1) a structural model that represents salient interactions between attributes for a given data, (2) a surrogate model which provides a functional approximation of the output as a function of attributes, and (3) a classification model which predicts the class for new inputs. The structural model is used to infer the functional form of the surrogate and its coefficients are estimated using linear regression methods. The classification model uses a maximally-accurate, least-complex surrogate to predict the output for given inputs. The structural model that yields an optimal classification model is searched using an iterative greedy search heuristic. Results show that the proposed method successfully detects the interacting variables in hierarchical problems, group them in linkages groups, and build maximally accurate classification models. The initial results on non-trivial hierarchical test problems indicate that the proposed method holds promise and have also shed light on several improvements to enhance the capabilities of the proposed method.

Abstract not found

Effective and efficient multiscale modeling is essential to advance both the science and synthesis in a wide array of fields such as physics, chemistry, materials science, biology, biotechnology and pharmacology. This study investigates the efficacy and potential of using genetic algorithms for multiscale materials modeling and addresses some of the challenges involved in designing competent algorithms that solve hard problems quickly, reliably and accurately. In particular, this thesis demonstrates the use of genetic algorithms (GAs) and genetic programming (GP) in multiscale modeling with the help of two non-trivial case studies in materials science and chemistry. The first case study explores the utility of genetic programming (GP) in multi-timescaling alloy kinetics simulations. In essence, GP is used to bridge molecular dynamics and kinetic Monte Carlo methods to span orders-of-magnitude in simulation time. Specifically, GP is used to regress symbolically an inline barrier function from a limited set of molecular dynamics simulations to enable kinetic Monte Carlo that simulate seconds of real time. Results on a non-trivial example of vacancy-assisted migration on a surface of a face-centered cubic (fcc) Copper-Cobalt (CuxCo1−x) alloy show that GP predicts all barriers with 0.1 % error from calculations for less than 3 % of active

Estimation of Distribution Algorithms (EDAs) have been proposed as an extension of genetic algorithms. We assume that the function to be optimized is additively decomposed (ADF). The interaction graph of the ADF is used to create exact or approximate factorizations of the Boltzmann distribution. Convergence of the algorithm MN-GIBBS is proven. MN-GIBBS uses a Markov network easily derived from the ADF and Gibbs sampling. The Factorized Distribution Algorithm (FDA) uses a less general representation, a Bayesian network and probabilistic logic sampling (PLS). We shortly describe the algorithm LFDA which learns a Bayesian network from data. The relation between the network computed by LFDA and the optimal network used by FDA is investigated. Convergence of FDA to the optima is shown for finite samples if the factorization fulfills the running intersection property. The sample size is bounded by O(nm ln nm) where n is the size of the problem and m the number of sub-functions. For the proof results from statistical learning theory and Probably Approximately Correct (PAC) learning are used. Numerical experiments show that even for difficult test functions a sample size which scales linearly with n is often sufficient. We also show that a good approximation of the true distribution is not necessary, it suffices to use a factorization where the global optima have a large enough probability. This explains the success of EDAs in practical applications.