Search

Missing values are ubiquitous in the nonlinear regression research, and may lead to bias and a loss of efficiency. Even in a large dataset, values drop‐out can substantially reduce the available information for analysis. In this paper, we propose an improved hybrid model to solve the nonlinear regression problem under missing data scenarios, consisting of two parts: an overcomplete winner‐take‐all (WTA) autoencoder and a multilayer gated linear network. The WTA autoencoder is trained in an adversarial training process by taking advantage of gradually renewed teacher signals and the discrimination of missing values and observed values, and is designed to play two roles: (1) to impute missing components conditioned on observed samples; (2) to generate gate control sequences. On the other hand, the multilayer gated linear network with the generated gate control sequences implements a powerful piecewise linear regression model, whose parameters are optimized by formulating a support vector regression (SVR) with a deep quasi‐linear kernel. Experimental results based on different real‐world datasets demonstrate the effectiveness of our proposed hybrid model. © 2022 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

In both the research and engineering fields, missing data is a serious problem that cannot be overlooked. Therefore, available datasets with missing data are a challenge to be modeled by conventional global prediction models. In this paper, we propose a hybrid model consisting of an autoencoder and a gated linear network for solving the regression problem under missing value scenario. A sophisticated modeling and identifying algorithm is developed. First, an extended affinity propagation (AP) clustering algorithm is applied to obtain a self‐organized competitive net dividing the datasets into several clusters. Second, a multiple imputation tool with top p% winner‐take‐all denoising autoencoders (DAE) is introduced to realize better predictions of missing values, in which rough estimates of missing values by using the mean imputation and similarity method within the clusters are used as teacher signals of DAE. Finally, a gated linear network is designed to construct a piecewise linear regression model with interpolations in the exact same way as a support vector regression with a quasilinear kernel composed using the cluster information obtained in the AP clustering step. Based on the experiments of five datasets, our proposed method demonstrates its effectiveness and robustness compared with other traditional kernels and state‐of‐the‐art methods, even on datasets with a large percentage of missing values. © 2020 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

Missing data is a prevailing problem in predictive analytics. In this paper, a winner‐take‐all (WTA) autoencoder‐based piecewise linear model is developed to solve the nonlinear regression problem under the missing value scenario, which consists of two parts: an overcomplete WTA autoencoder and a gated linear network. The overcomplete WTA autoencoder is a stacked denoising autoencoder (SDAE) designed to play two roles: (1) to estimate the missing values; (2) to realize a sophisticated partitioning by generating a broad set of binary gate control sequences. Besides, an iterative algorithm with renewed teacher signals is developed to train the SDAE. On the other hand, the gated linear network with the generated binary gate control sequences implements a flexible piecewise linear model for nonlinear regression. By composing a quasi‐linear kernel based on the gate control sequences, the piecewise linear model is then identified in the same way as a support vector regression. Experimental results have shown that our proposed hybrid model has a better performance than traditional models. © 2021 Institute of Electrical Engineers of Japan. Published by Wiley Periodicals LLC. [ABSTRACT FROM AUTHOR]

It is commonly accepted that the horizontal-to-vertical spectral ratio (HVSR) technique enables the detection of the fundamental resonance frequency ( f HVSR ) of a given site. The utility of this f HVSR is analyzed using the nonlinear regression relationships between f HVSR and bedrock depth (h ). The derived relationships are mostly site-specific, so that the present paper consists of two main parts. The first is a literature review for the available empirical relationships between f HVSR and h . The aim of this part is to highlight the practical limitations of these established relationships and to make fair comparisons. The second is to generate new relationships, taking advantage of the very wide range of available lithological, geophysical, and geotechnical borehole drilling data of the 697 KiK-NET seismic stations in Japan. For this purpose, HVSR are calculated using 10,000 weak earthquakes or linear events recorded at KiK-NET stations to determine the f HVSR and correlate it with the corresponding h . The overlying layers/bedrock interface falling within sedimentary, igneous, or metamorphic layers significantly affect the derived frequency–depth relationships. In addition, these relationships are strongly reproduced by the V p / V s ratio of the bedrock in the range of 1.6–2.2. Interestingly, it is found that f HVSR less than 1 Hz corresponding to h more than 100 m leads the trend of the overall frequency–depth relationship. [ABSTRACT FROM AUTHOR]

Dose response functions (DRFs) are examined in terms of applicability to the data of outdoor exposure tests in Japan (JWTC data) in order to improve estimation methods for assessing the domestic atmospheric corrosivity of carbon steel and zinc. ISO9223DRF is selected from a literature survey as a suitable DRF. The corrosion rate estimated by the ISO9223DRF shows insufficient agreement with the measured rate from the JWTC data. The ISO9223DRF coefficients are redetermined by a nonlinear least-squares regression model based on the JWTC data (fitted DRF). The corrosion rate estimated for the fitted DRF shows better agreement with the measured rate from the JWTC data in terms of correlation with the measured one and the fraction of the estimated one within errors of --50...+100%. [ABSTRACT FROM AUTHOR]