Bayesian optimization with partially specified queries

Bayesian optimization (BO) is an approach to optimizing an expensive-to-evaluate black-box function and sequentially determines the values of input variables to evaluate the function. However, it is expensive and in some cases becomes difficult to specify values for all input variables, for example, in outsourcing scenarios where production of input queries with many input variables involves significant cost. In this paper, we propose a novel Gaussian process bandit problem, BO with partially specified queries (BOPSQ). In BOPSQ, unlike the standard BO setting, a learner specifies only the values of some input variables, and the values of the unspecified input variables are randomly determined according to a known or unknown distribution. We propose two algorithms based on posterior sampling for cases of known and unknown input distributions. We further derive their regret bounds that are sublinear for popular kernels. We demonstrate the effectiveness of the proposed algorithms using test functions and real-world datasets.

Analysis of Precipitation Variability using Memory Based Artificial Neural Networks

This article analyzes the variability in precipitation of the Barak river basin using memory-based ANN models called Gamma Memory Neural Network(GMNN) and genetically optimized GMNN called GMNN-GA for precipitation downscaling precipitation. GMNN having adaptive memory depth is capable techniques in modeling time varying inputs with unknown input characteristics, while an integration of the model with GA can further improve its performances. NCEP reanalysis and HadCM3A2 (a) scenario data are used for downscaling and forecasting precipitation series for Barak river basin. Model performances are analyzed by using statistical criteria, RMSE and mean error and are compared with the standard SDSM model. Results obtained by using 24 years of daily data sets show that GMNN-GA is efficient in downscaling daily precipitation series with maximum daily annual mean error of 6.78%. The outcomes of the study demonstrate that execution of the GMNN-GA model is superior to the GMNN and similar with that of the standard SDSM.

Fault detection in a three-tank hydraulic system using unknown input observer and extended Kalman filter

This paper presents the problem of fault diagnosis in a three-tank hydraulic system. A mathematical model of the system is developed in order to apply two different observing algorithms. Unknown Input Observer (UIO) and Extended Kalman Filter (EKF) have been used to detect and isolate actuator and sensor faults. For Unknown Input Observer (UIO), residuals are calculated from the measured and estimated output according to the eigenvalues of the system after processed by Linear Matrix Inequality (LMI). Extended Kalman filter uses process and measurement noise variances for state estimation. Unknown Input Observer and Extended Kalman Filter's performance in fault estimation and isolation is evaluated under different scenarios. Using Extended Kalman Filter (EKF), faults can be diagnosed effectively in the presence of noise, while Unknown Input Observer (UIO) is working better in the absence of noise, and simulation results illustrate that clearly.