Search

Adaptive importance sampling (AIS) methods are increasingly used for the approximation of distributions and related intractable integrals in the context of Bayesian inference. Population Monte Carlo (PMC) algorithms are a subclass of AIS methods, widely used due to their ease in the adaptation. In this paper, we propose a novel algorithm that exploits the benefits of the PMC framework and includes more efficient adaptive mechanisms, exploiting geometric information of the target distribution. In particular, the novel algorithm adapts the location and scale parameters of a set of importance densities (proposals). At each iteration, the location parameters are adapted by combining a versatile resampling strategy (i.e., using the information of previous weighted samples) with an advanced optimization-based scheme. Local second-order information of the target distribution is incorporated through a preconditioning matrix acting as a scaling metric onto a gradient direction. A damped Newton approach is adopted to ensure robustness of the scheme. The resulting metric is also used to update the scale parameters of the proposals. We discuss several key theoretical foundations for the proposed approach. Finally, we show the successful performance of the proposed method in three numerical examples, involving challenging distributions. [ABSTRACT FROM AUTHOR]

We present new variants of Estimation of Distribution Algorithms (EDA) for large-scale continuous optimisation that extend and enhance a recently proposed random projection (RP) ensemble based approach. The main novelty here is to depart from the theory of RPs that require (sub-)Gaussian random matrices for norm-preservation, and instead for the purposes of high-dimensional search we propose to employ random matrices with independent and identically distributed entries drawn from a t-distribution. We analytically show that the implicitly resulting high-dimensional covariance of the search distribution is enlarged as a result. Moreover, the extent of this enlargement is controlled by a single parameter, the degree of freedom. For this reason, in the context of optimisation, such heavy tailed random matrices turn out to be preferable over the previously employed (sub-)Gaussians. Based on this observation, we then propose novel covariance adaptation schemes that are able to adapt the degree of freedom parameter during the search, and give rise to a flexible approach to balance exploration versus exploitation. We perform a thorough experimental study on high-dimensional benchmark functions, and provide statistical analyses that demonstrate the state-of-the-art performance of our approach when compared with existing alternatives in problems with 1 000 search variables. [ABSTRACT FROM AUTHOR]

In robots with elastic joints, discrepancies exist between the motor side and the load side ( i.e., the link of the robotic joint) due to joint elasticity. Thus, sensing only at the motor side does not precisely provide the information of interest at the load side. In view that the load side states determine the position and orientation of the end-effector, it is critical to have the load side states sensed and/or estimated. In this paper, we present a comprehensive comparative study on load side state estimation utilizing two different models of robotic elastic joints. Low-cost MEMS inertia sensors, such as gyroscopes and accelerometers, are adopted on the load side to supplement the motor side sensing. We also incorporate sensor noises and biases in the measurement dynamics. Kalman filtering methods are designed based on the extended dynamic/kinematic models for the fusion of multiple sensor signals. We also study the specific issue related to the noise covariance adaptation. The effectiveness of the proposed schemes is experimentally demonstrated and compared on a robotic elastic joint testbed. [ABSTRACT FROM AUTHOR]

Abstract: An improved Generalised Regression Neural Network is proposed for function approximation that incorporates kernels which adapt to the local structural information of the training data. Unlike the standard network, it allows bandwidth information to vary efficiently with each pattern in order to allow better adaptation to the local spatial arrangements of the nearest neighbours. The proposed network allows the use of structural information by employing full covariances with adaptive kernel volumes that are trained to form the optimum regression surfaces. Experiments show improved accuracy over the standard regression models with computationally efficient training. [Copyright &y& Elsevier]