A genetic neural network approach to identifying reaction coordinates

Aaron Dinner
Department of Chemistry
University of Chicago
RI 340E, 5640 S Ellis Ave
Chicago, IL 60637
USA

To interpret simulations of a complex system to determine the physical mechanism of a dynamical process, it is necessary to identify the small number of coordinates that distinguish the stable states from the transition states. We develop an automatic method for identifying these degrees of freedom from a database of candidate physical variables. In the method, neural networks are used to determine the functional dependence of the probability of committing to a stable state (committor) on a set of coordinates, and a genetic algorithm selects the combination of inputs that yields the best fit. The method enables us to obtain the first set of coordinates that is demonstrably sufficient to specify the transition state of the C7eq to R isomerization of the alanine dipeptide in the presence of explicit water molecules. It is revealed that the solute-solvent coupling can be described by a solvent-derived electrostatic torque around one of the main chain bonds, and the collective, long-ranged nature of this interaction accounts for previous failures to characterize this reaction.

<more | download>

Evolving Recurrent Neural Networks with Non binary Encoding

Martin Mandischer
Systems Analysis Research Group
Department of Computer Science
University of Dortmund
Dortmund Germany

This paper presents an evolutionary approach for the design of feed forward and recurrent neural networks We show that Evolutionary Algorithms can be used for the construction of networks for real world tasks Therefore a data structure based genotypic network representation as well as genetic operators are introduced Results from the classification function approximation and time series domain are presented.

<more | download>

Integration of Genetic Algorithm and Neural Network for Financial Early Warning System: An Example of Taiwanese Banking Industry

Jih-Chang Hsieh(1), Pei-Chann Chang(2), Shih-Hsin Chen(2)
1. Department of Finance, Vanung University, Chung-Li 32061, Tao-Yuan, Taiwan.
2. Department of Industrial Engineering and Management, Yuan-Ze University, Chung-Li,
Tao-Yuan, Taiwan.

Genetic algorithm and neural network (GNN) are integrated to build a financial early warning system. An example of Taiwanese banking industry is discussed to test the hit ratio of each system. The performance is compared with other four early warning systems, namely, case-based reasoning, backpropagation neural network, logistic regression analysis, and quadratic discriminant analysis. The result indicates that the GNN proposed in this research is a little superior to the other two soft computing early warning systems. And the GNN outperforms the statistical early warning systems at least 13%.

<more | download>

Feedback Approximation of the Stochastic Growth Model by Genetic Neural Networks

Sibel Sirakaya(a) Stephen Turnovsky(b);y M. Nedim Alemdar(c)

a. Departments of Economics and Statistics, and CSSS,
University of Washington, Seattle, WA 98195
b. Department of Economics, University of Washington, Seattle, WA 98195
c. Department of Economics, Bilkent University, 06800 Bilkent, Ankara, Turkey

A direct numerical optimization method is developed to approximate the one-sector stochastic growth model. The feedback investment policy is parameterized as a neural network and trained by a genetic algorithm to maximize the utility functional over the space of time-invariant investment policies. To eliminate the dependence of training on the initial conditions, at any generation, the same stationary investment policy (the same network) is used to repeatedly solve the problem from di®ering initial conditions. The ¯tness of a given policy rule is then computed as the sum of payo®s over all initial conditions. The algorithm performs quite well under a wide set of parameters. Given the general purpose nature of the method, the °exibility of neural network parametrization and the global nature of the genetic algorithm search, it can be easily extended to tackle problems with higher dimensional nonlinearities, state spaces and/or discontinuities.

<more | download>

MASSP: A hybrid genetic-neural system for predicting protein secondary structure

Armano G.(1), Mancosu G.(2), Orro A.(1), Saba M.(1), and Vargiu E.(1)
(1) DIEE - Dept. of Electrical and Electronic Engineering, University of Cagliari Piazza d’Armi, I-09123
Cagliari, Italy (2) Shardna Life Sciences, Piazza Deffenu 4, I-09121 Cagliari, Italy

Being the prediction of protein structure a very complex task, most methodologies concentrate on the simplified task of predicting secondary structures. In this paper, we illustrate a technique based on multiple experts, aimed at predicting protein secondary structures. The prediction activity results from the interaction of a population of experts, each integrating genetic and neural technologies. Roughly speaking, an expert of this kind embodies a genetic classifier designed to control the activation of a feedforward artificial neural network for performing a locally-scoped prediction activity. Genetic and neural components (i.e., guard and embedded predictor, respectively) are devoted to perform different tasks and are supplied with different information: Each guard is aimed at (soft-)partitioning the input space, insomuch assuring both the diversity and the specialization of the corresponding embedded predictor, which in turn is devoted to perform the actual prediction.

<more | download>

GENETIC NEURAL NETWORK BASED DATA MINING AND APPLICATION IN CASE ANALYSIS OF POLICE OFFICE

LIU Han-li, LI Lin, ZHU Hai-hong
School of Resource and Environment Science, Wuhan University, 129 Luoyu Road, Wuhan, P.R.China, 430079
Tel: 86-27-87882209 E-mail: liuhl000@sohu.com; lilin@telecarto.com

This paper puts forward a method that combines the learning algorithm of BP neural network with genetic algorithm to train BP network and optimize the weight values of the network in a global scale. This method is featured as global optimization, high accuracy and fast convergence. The data-mining model based on genetic neural network has been widely applied to the procedure of data mining on case information in the command centre of police office. It achieves an excellent effect for assisting people to solve cases and make good decisions. In this paper, the principles and methods of this data-mining model are described in details. A real case of its application is also presented. From this case we can draw a conclusion that the data-mining model we have chosen is scientific, efficient and practicable.

<More | Download>

BROADBAND MODELING OF WIRE ANTENNAS BY GENETIC NEURAL NETWORKS

Abstract. The paper describes an approach to learning artificial neural networks (ANN) with the genetic algorithm (GA). Two types of ANN are used: the recurrent Elman ANN and the feed-forward one. Neural networks are implemented in MATLAB. Results of training abilities are discussed.

<more | download>