Learning polynomial feedforward neural networks by genetic programming and backpropagation

This paper presents an approach to learning polynomial feedforward neural networks (PFNNs). The approach suggests, first, finding the polynomial network structure by means of a population-based search technique relying on the genetic programming paradigm, and second, further adjustment of the best discovered network weights by an especially derived backpropagation algorithm for higher order networks with polynomial activation functions. These two stages of the PFNN learning process enable us to identify networks with good training as well as generalization performance. Empirical results show that this approach finds PFNN which outperform considerably some previous constructive polynomial network algorithms on processing benchmark time series.     

Adaptive evolutionary programming based on reinforcement learning

This paper studies evolutionary programming and adopts reinforcement learning theory to learn individual mutation operators. A novel algorithm named RLEP (Evolutionary Programming based on Reinforcement Learning) is proposed. In this algorithm, each individual learns its optimal mutation operator based on the immediate and delayed performance of mutation operators. Mutation operator selection is mapped into a reinforcement learning problem. Reinforcement learning methods are used to learn optimal policies by maximizing the accumulated rewards. According to the calculated Q function value of each candidate mutation operator, an optimal mutation operator can be selected to maximize the learned Q function value. Four different mutation operators have been employed as the basic candidate operators in RLEP and one is selected for each individual in different generations. Our simulation shows the performance of RLEP is the same as or better than the best of the four basic mutation operators.     

Adaptive memetic method of multi-objective genetic evolutionary algorithm for backpropagation neural network

Neural Computing and Applications - In recent years, multi-objective evolutionary optimization algorithms have shown success in different areas of research. Due to their efficiency and power, many...     

Bayesian methods for elucidating genetic regulatory networks

Bayesian network methods are useful for elucidating genetic regulatory networks because they can represent more than pair-wise relationships between variables, are resistant to overfitting, and remain robust in the face of noisy data.     

Adaptive learning of dynamic Bayesian networks with changing structures by detecting geometric structures of time series

A dynamic Bayesian network (DBN) is one of popular approaches for relational knowledge discovery such as modeling relations or dependencies, which change over time, between variables of a dynamic system. In this paper, we propose an adaptive learning method (autoDBN) to learn DBNs with changing structures from multivariate time series. In autoDBN, segmentation of time series is achieved first through detecting geometric structures transformed from time series, and then model regions are found from the segmentation by designed finding strategies; in each found model region, a DBN model is established by existing structure learning methods; finally, model revisiting is developed to refine model regions and improve DBN models. These techniques provide a special mechanism to find accurate model regions and discover a sequence of DBNs with changing structures, which are adaptive to changing relations between multivariate time series. Experimental results on simulated and real time series show that autoDBN is very effective in finding accurate/reasonable model regions and gives lower error rates, outperforming the switching linear dynamic system method and moving window method.

Recognizing multi-modal sensor signals using evolutionary learning of dynamic Bayesian networks

Multi-modal context-aware systems can provide user-adaptive services, but it requires complicated recognition models with larger resources. The limitations to build optimal models and infer the context efficiently make it difficult to develop practical context-aware systems. We developed a multi-modal context-aware system with various wearable sensors including accelerometers, gyroscopes, physiological sensors, and data gloves. The system used probabilistic models to handle the uncertain and noisy time-series sensor data. In order to construct the efficient probabilistic models, this paper uses an evolutionary algorithm to model structure and EM algorithm to determine parameters. The trained models are selectively inferred based on a semantic network which describes the semantic relations of the contexts and sensors. Experiments with the real data collected show the usefulness of the proposed method.     

Application of time series and polynomial learning networks to robot trajectory error control

A compensatory control scheme based on measured errors at the end-effector is proposed using polynomial learning networks and time series modeling. Based on experimental data from an industrial manipulator programmed forms straight-line motion, trajectory deviations are modeled using both techniques. The performance of the models are compared at different locations in the workspace. It is shown that the robot arm signature can be obtained and that models from both techniques can be used to forecast trajectory errors. A method to implement the proposed scheme is also given.     

Learning nonlinear multiregression networks based on evolutionary computation

This paper describes a novel knowledge discovery and data mining framework dealing with nonlinear interactions among domain attributes. Our network-based model provides an effective and efficient reasoning procedure to perform prediction and decision making. Unlike many existing paradigms based on linear models, the attribute relationship in our framework is represented by nonlinear nonnegative multiregressions based on the Choquet integral. This kind of multiregression is able to model a rich set of nonlinear interactions directly. Our framework involves two layers. The outer layer is a network structure consisting of network elements as its components, while the inner layer is concerned with a particular network element modeled by Choquet integrals. We develop a fast double optimization algorithm (FDOA) for learning the multiregression coefficients of a single network element. Using this local learning component and multiregression-residual-cost evolutionary programming (MRCEP), we propose a global learning algorithm, called MRCEP-FDOA, for discovering the network structures and their elements from databases. We have conducted a series of experiments to assess the effectiveness of our algorithm and investigate the performance under different parameter combinations, as well as sizes of the training data sets. The empirical results demonstrate that our framework can successfully discover the target network structure and the regression coefficients.     

A new genetic approach for structure learning of Bayesian networks: Matrix genetic algorithm

In this paper, a novel method for structure learning of a Bayesian network (BN) is developed. A new genetic approach called the matrix genetic algorithm (MGA) is proposed. In this method, an individual structure is represented as a matrix chromosome and each matrix chromosome is encoded as concatenation of upper and lower triangular parts. The two triangular parts denote the connection in the BN structure. Further, new genetic operators are developed to implement the MGA. The genetic operators are closed in the set of the directed acyclic graph (DAG). Finally, the proposed scheme is applied to real world and benchmark applications, and its effectiveness is demonstrated through computer simulation.     

基于进化规划的神经网络群的设计方法

提出一种基于进化规划的神经网络群的自动设计方法．该方法不仅使得神经网络群中的个体网络倾向于完成不同的子任务，同时各神经网络个体在进化过程中不断寻找最好的协作关系，而且神经网络群的规模和结构不需预先设定．仿真试验表明，该算法是有效的。     

D-Separation and computation of probability distributions in Bayesian networks

Consider a family ${(X_i)_{i \in I}}$ of random variables endowed with the structure of a Bayesian network, and a subset S of I. This paper examines the problem of computing the probability distribution of the subfamily ${(X_{a})_{a \in S}}$ (respectively the probability distribution of ${ (X_{b})_{b \in {\bar{S}}}}$ , where ${{\bar{S}} = I - S}$ , conditional on ${(X_{a})_{a \in S}}$ ). This paper presents some theoretical results that makes it possible to compute joint and conditional probabilities over a subset of variables by computing over separate components. In other words, it is demonstrated that it is possible to decompose this task into several parallel computations, each related to a subset of S (respectively of ${{\bar{S}}}$ ); these partial results are then put together as a final product. In computing the probability distribution over ${(X_a)_{a \in S}}$ , this procedure results in the production of a structure of level two Bayesian network structure for S.     

Bayesian learning of Bayesian networks with informative priors

This paper presents and evaluates an approach to Bayesian model averaging where the models are Bayesian nets (BNs). A comprehensive study of the literature on structural priors for BNs is conducted. A number of prior distributions are defined using stochastic logic programs and the MCMC Metropolis-Hastings algorithm is used to (approximately) sample from the posterior. We use proposals which are tightly coupled to the priors which give rise to cheaply computable acceptance probabilities. Experiments using data generated from known BNs have been conducted to evaluate the method. The experiments used 6 different BNs and varied: the structural prior, the parameter prior, the Metropolis-Hasting proposal and the data size. Each experiment was repeated three times with different random seeds to test the robustness of the MCMC-produced results. Our results show that with effective priors (i) robust results are produced and (ii) informative priors improve results significantly.     

Improving experimental methods on success rates in evolutionary computation

Abstract

Due to the complexity of theoretical approaches in evolutionary computation (EC), research has being largely performed on experimental basis. One popular measure used by the EC community is the success rate (SR), which is used alone or as part of more complex measures such as Koza’s computational effort in genetic programming. A common practice in EC is to report just a punctual estimation of the SR, without additional information about its associated uncertainty. We aim to motivate EC researchers to adopt more rigorous practices when working with SRs. In particular, we introduce the importance of correctly reporting this measure and highlight its binomial nature. Unfortunately, this fact is usually overlooked in the literature. Considering the binomiality of the SR opens the whole corpus of binomial statistics to EA research and practice. In particular, we focus on studying several methods to compute SR confidence intervals, the factors that determine their quality in terms of coverage probability and interval length. Due to its practical interest, we also briefly discuss the number of required runs to build confidence intervals with a certain quality, providing a sound method to set the number of runs, one of the most important experimental settings in EC. Evidence suggests that Wilson is, on average, a reliable and simple method to bound an estimation of SR with confidence intervals, while the standard method, which is quite popular because of its conceptual simplicity, should be avoided in any case. However, other methods can also be of interest under certain circumstances. We encourage to report the number of trials and successes, as well as the interval, to ease further comparability of the results.     

Interactive structural learning of Bayesian networks

We propose an hybrid approach for structure learning of Bayesian networks, in which a computer system and a human expert cooperate to search for the best structure. The system builds an initial tree structure which is graphically presented to the expert, and then the expert can modify this structure according to his knowledge of the domain. The system has several tools for aiding the human in this task: it allows for graphical editing (adding, deleting, inverting arcs) of the network, it shows graphically the correlation between variables, and it gives a measure of the quality and complexity for each structure. A measure which combines both quality and complexity, that we call quality, is defined. We have tested the tool in two domains: atmospheric pollution and car insurance, with good results.     

贝叶斯网络结构学习综述

贝叶斯网络是一种有效的不确定性知识表达和推理工具,在数据挖掘等领域得到了较好的应用,而结构学习是其重要研究内容之一.经过二十多年的发展,已经出现了一些比较成熟的贝叶斯网络结构学习算法,对迄今为止的贝叶斯网络结构学习方法进行了综述.现阶段获得的用于结构学习的观测数据都比较复杂,这些数据分为完备数据和不完备数据两种类型.针对完备数据,分别从基于依赖统计分析的方法、基于评分搜索的方法和混合搜索方法三个方面对已有的算法进行分析.对于不完备数据,给出了数据不完备情况下网络结构的学习框架.在此基础上归纳总结了贝叶斯网络结构学习各个方向的研究进展,给出了贝叶斯网络结构学习未来可能的研究方向.