On Convergence Properties of the EM Algorithm for Gaussian Mixtures

"Expectation-Maximization'' (EM) algorithm and gradient-based approaches for maximum likelihood learning of finite Gaussian mixtures. We show that the EM step in parameter space is obtained from the gradient via a projection matrix $P$, and we provide an explicit expression for the matrix. We then analyze the convergence of EM in terms of special properties of $P$ and provide new results analyzing the effect that $P$ has on the likelihood surface. Based on these mathematical results, we present a comparative discussion of the advantages and disadvantages of EM and other algorithms for the learning of Gaussian mixture models.

Efficient learning of reactive robot behaviors with a Neural-Q_learning approach

The purpose of this paper is to propose a Neural-Q_learning approach designed for online learning of simple and reactive robot behaviors. In this approach, the Q_function is generalized by a multi-layer neural network allowing the use of continuous states and actions. The algorithm uses a database of the most recent learning samples to accelerate and guarantee the convergence. Each Neural-Q_learning function represents an independent, reactive and adaptive behavior which maps sensorial states to robot control actions. A group of these behaviors constitutes a reactive control scheme designed to fulfill simple missions. The paper centers on the description of the Neural-Q_learning based behaviors showing their performance with an underwater robot in a target following task. Real experiments demonstrate the convergence and stability of the learning system, pointing out its suitability for online robot learning. Advantages and limitations are discussed

Semi-online neural-Q_leaming for real-time robot learning

Reinforcement learning (RL) is a very suitable technique for robot learning, as it can learn in unknown environments and in real-time computation. The main difficulties in adapting classic RL algorithms to robotic systems are the generalization problem and the correct observation of the Markovian state. This paper attempts to solve the generalization problem by proposing the semi-online neural-Q_learning algorithm (SONQL). The algorithm uses the classic Q_learning technique with two modifications. First, a neural network (NN) approximates the Q_function allowing the use of continuous states and actions. Second, a database of the most representative learning samples accelerates and stabilizes the convergence. The term semi-online is referred to the fact that the algorithm uses the current but also past learning samples. However, the algorithm is able to learn in real-time while the robot is interacting with the environment. The paper shows simulated results with the "mountain-car" benchmark and, also, real results with an underwater robot in a target following behavior

Validation of improved TAMANN neural network for operational satellite-derived rainfall estimation in Africa

Real-time rainfall monitoring in Africa is of great practical importance for operational applications in hydrology and agriculture. Satellite data have been used in this context for many years because of the lack of surface observations. This paper describes an improved artificial neural network algorithm for operational applications. The algorithm combines numerical weather model information with the satellite data. Using this algorithm, daily rainfall estimates were derived for 4 yr of the Ethiopian and Zambian main rainy seasons and were compared with two other algorithms-a multiple linear regression making use of the same information as that of the neural network and a satellite-only method. All algorithms were validated against rain gauge data. Overall, the neural network performs best, but the extent to which it does so depends on the calibration/validation protocol. The advantages of the neural network are most evident when calibration data are numerous and close in space and time to the validation data. This result emphasizes the importance of a real-time calibration system.

Neural and wavelet network models for financial distress classification

This work analyzes the use of linear discriminant models, multi-layer perceptron neural networks and wavelet networks for corporate financial distress prediction. Although simple and easy to interpret, linear models require statistical assumptions that may be unrealistic. Neural networks are able to discriminate patterns that are not linearly separable, but the large number of parameters involved in a neural model often causes generalization problems. Wavelet networks are classification models that implement nonlinear discriminant surfaces as the superposition of dilated and translated versions of a single "mother wavelet" function. In this paper, an algorithm is proposed to select dilation and translation parameters that yield a wavelet network classifier with good parsimony characteristics. The models are compared in a case study involving failed and continuing British firms in the period 1997-2000. Problems associated with over-parameterized neural networks are illustrated and the Optimal Brain Damage pruning technique is employed to obtain a parsimonious neural model. The results, supported by a re-sampling study, show that both neural and wavelet networks may be a valid alternative to classical linear discriminant models.

Application of neural network to hybrid systems with binary inputs

Boolean input systems are in common used in the electric industry. Power supplies include such systems and the power converter represents these. For instance, in power electronics, the control variable are the switching ON and OFF of components as thyristors or transistors. The purpose of this paper is to use neural network (NN) to control continuous systems with Boolean inputs. This method is based on classification of system variations associated with input configurations. The classical supervised backpropagation algorithm is used to train the networks. The training of the artificial neural network and the control of Boolean input systems are presented. The design procedure of control systems is implemented on a nonlinear system. We apply those results to control an electrical system composed of an induction machine and its power converter.

Modified radial basis function neural network using output transformation

A modified radial basis function (RBF) neural network and its identification algorithm based on observational data with heterogeneous noise are introduced. The transformed system output of Box-Cox is represented by the RBF neural network. To identify the model from observational data, the singular value decomposition of the full regression matrix consisting of basis functions formed by system input data is initially carried out and a new fast identification method is then developed using Gauss-Newton algorithm to derive the required Box-Cox transformation, based on a maximum likelihood estimator (MLE) for a model base spanned by the largest eigenvectors. Finally, the Box-Cox transformation-based RBF neural network, with good generalisation and sparsity, is identified based on the derived optimal Box-Cox transformation and an orthogonal forward regression algorithm using a pseudo-PRESS statistic to select a sparse RBF model with good generalisation. The proposed algorithm and its efficacy are demonstrated with numerical examples.

A new RBF neural network with boundary value constraints

We present a novel topology of the radial basis function (RBF) neural network, referred to as the boundary value constraints (BVC)-RBF, which is able to automatically satisfy a set of BVC. Unlike most existing neural networks whereby the model is identified via learning from observational data only, the proposed BVC-RBF offers a generic framework by taking into account both the deterministic prior knowledge and the stochastic data in an intelligent manner. Like a conventional RBF, the proposed BVC-RBF has a linear-in-the-parameter structure, such that it is advantageous that many of the existing algorithms for linear-in-the-parameters models are directly applicable. The BVC satisfaction properties of the proposed BVC-RBF are discussed. Finally, numerical examples based on the combined D-optimality-based orthogonal least squares algorithm are utilized to illustrate the performance of the proposed BVC-RBF for completeness.

A forward-constrained regression algorithm for sparse kernel density estimation

Using the classical Parzen window (PW) estimate as the target function, the sparse kernel density estimator is constructed in a forward-constrained regression (FCR) manner. The proposed algorithm selects significant kernels one at a time, while the leave-one-out (LOO) test score is minimized subject to a simple positivity constraint in each forward stage. The model parameter estimation in each forward stage is simply the solution of jackknife parameter estimator for a single parameter, subject to the same positivity constraint check. For each selected kernels, the associated kernel width is updated via the Gauss-Newton method with the model parameter estimate fixed. The proposed approach is simple to implement and the associated computational cost is very low. Numerical examples are employed to demonstrate the efficacy of the proposed approach.

A-optimality orthogonal forward regression algorithm using branch and bound

In this brief, we propose an orthogonal forward regression (OFR) algorithm based on the principles of the branch and bound (BB) and A-optimality experimental design. At each forward regression step, each candidate from a pool of candidate regressors, referred to as S, is evaluated in turn with three possible decisions: 1) one of these is selected and included into the model; 2) some of these remain in S for evaluation in the next forward regression step; and 3) the rest are permanently eliminated from S. Based on the BB principle in combination with an A-optimality composite cost function for model structure determination, a simple adaptive diagnostics test is proposed to determine the decision boundary between 2) and 3). As such the proposed algorithm can significantly reduce the computational cost in the A-optimality OFR algorithm. Numerical examples are used to demonstrate the effectiveness of the proposed algorithm.

A robust nonlinear identification algorithm using PRESS statistic and forward regression

This letter introduces a new robust nonlinear identification algorithm using the Predicted REsidual Sums of Squares (PRESS) statistic and for-ward regression. The major contribution is to compute the PRESS statistic within a framework of a forward orthogonalization process and hence construct a model with a good generalization property. Based on the properties of the PRESS statistic the proposed algorithm can achieve a fully automated procedure without resort to any other validation data set for iterative model evaluation.

Synapsing variable length crossover: An algorithm for crossing and comparing variable length genomes

The Synapsing Variable Length Crossover (SVLC) algorithm provides a biologically inspired method for performing meaningful crossover between variable length genomes. In addition to providing a rationale for variable length crossover it also provides a genotypic similarity metric for variable length genomes enabling standard niche formation techniques to be used with variable length genomes. Unlike other variable length crossover techniques which consider genomes to be rigid inflexible arrays and where some or all of the crossover points are randomly selected, the SVLC algorithm considers genomes to be flexible and chooses non-random crossover points based on the common parental sequence similarity. The SVLC Algorithm recurrently "glues" or synapses homogenous genetic sub-sequences together. This is done in such a way that common parental sequences are automatically preserved in the offspring with only the genetic differences being exchanged or removed, independent of the length of such differences. In a variable length test problem the SVLC algorithm is shown to outperform current variable length crossover techniques. The SVLC algorithm is also shown to work in a more realistic robot neural network controller evolution application.

A multi-level probabilistic neural network

Based on the idea of an important cluster, a new multi-level probabilistic neural network (MLPNN) is introduced. The MLPNN uses an incremental constructive approach, i.e. it grows level by level. The construction algorithm of the MLPNN is proposed such that the classification accuracy monotonically increases to ensure that the classification accuracy of the MLPNN is higher than or equal to that of the traditional PNN. Numerical examples are included to demonstrate the effectiveness of proposed new approach.

A hyperstable neural network for the modelling and control of nonlinear systems

A multivariable hyperstable robust adaptive decoupling control algorithm based on a neural network is presented for the control of nonlinear multivariable coupled systems with unknown parameters and structure. The Popov theorem is used in the design of the controller. The modelling errors, coupling action and other uncertainties of the system are identified on-line by a neural network. The identified results are taken as compensation signals such that the robust adaptive control of nonlinear systems is realised. Simulation results are given.

Neural network enhanced generalised minimum variance self-tuning controller for nonlinear discrete-time systems

A neural network enhanced self-tuning controller is presented, which combines the attributes of neural network mapping with a generalised minimum variance self-tuning control (STC) strategy. In this way the controller can deal with nonlinear plants, which exhibit features such as uncertainties, nonminimum phase behaviour, coupling effects and may have unmodelled dynamics, and whose nonlinearities are assumed to be globally bounded. The unknown nonlinear plants to be controlled are approximated by an equivalent model composed of a simple linear submodel plus a nonlinear submodel. A generalised recursive least squares algorithm is used to identify the linear submodel and a layered neural network is used to detect the unknown nonlinear submodel in which the weights are updated based on the error between the plant output and the output from the linear submodel. The procedure for controller design is based on the equivalent model therefore the nonlinear submodel is naturally accommodated within the control law. Two simulation studies are provided to demonstrate the effectiveness of the control algorithm.