OTL-PEM: an optimization-based two-layer pointwise ensemble of surrogate models

The ensemble of surrogate models is increasingly implemented in practice for its more flexibility and robustness compared to the individual surrogate models. In this work, a novel pointwise ensemble of surrogate models named the optimization-based two-layer pointwise ensemble of surrogate model (OTL-PEM) is proposed. In the OTL-PEM, the framework of two-layer surrogate models is defined, where the data-surrogate models containing different types of individual surrogate models are to fit the given dataset, while the weight-surrogate models are modeled based on the cross-validation errors aiming to fit the pointwise weights of different individual surrogate models. To avoid the negative influence of the poor individual surrogate models, the model selection problem is transformed into several optimization problems which can be solved easily by the mature optimization algorithm to eliminate the globally poor surrogate models. In addition, the optimization space is extracted to alleviating the predictive instability caused by the extrapolation of the weight-surrogate models. Forty test functions are used to select the appropriate hyperparameters of the OTL-PEM, and to evaluate the performance of the OTL-PEM. The results indicate that the OTL-PEM can provide more accurate and robust approximation performance compared with individual surrogate models as well as other ensembles of surrogate models.

Bayesian uncertainty quantification in aerosol optical depth retrieval applied to TROPOMI measurements

We present here an aerosol model selection based statistical method in Bayesian framework for retrieving atmospheric aerosol optical depth (AOD) and pixel-level uncertainty. Especially, we focus on to provide more realistic uncertainty estimate by taking into account a model selection problem when searching for the solution by fitting look-up table (LUT) models to a satellite measured top-of-atmosphere reflectance. By means of Bayesian model averaging over the best-fitting aerosol models we take into account an aerosol model selection uncertainty and get also a shared inference about AOD. Moreover, we acknowledge model discrepancy, i.e. forward model error, arising from approximations and assumptions done in forward model simulations. We have estimated the model discrepancy empirically by a statistical approach utilizing residuals of model fits. We use the measurements from the TROPOspheric Monitoring Instrument (TROPOMI) onboard the Sentinel-5 Precursor in ultraviolet and visible bands, and in one wavelength band 675 nm in near-infrared, in order to study the functioning of the retrieval in a broad wavelength range. We exploit a fundamental classification of the aerosol models as weakly absorbing, biomass burning and desert dust aerosols. For experimental purpose we have included some dust type of aerosols having non-spherical particle shapes. For this study we have created the aerosol model LUTs with radiative transfer simulations using the libRadtran software package. It is reasonably straightforward to experiment with different aerosol types and evaluate the most probable aerosol type by the model selection method. We demonstrate the method in wildfire and dust events in a pixel level. In addition, we have evaluated in detail the results against ground-based remote sensing data from the AErosol RObotic NETwork (AERONET). Based on the case studies the method has ability to identify the appropriate aerosol types, but in some wildfire cases the AOD is overestimated compared to the AERONET result. The resulting uncertainty when accounting for the model selection problem and the imperfect forward modelling is higher compared to uncertainty when only measurement error is included in an observation model, as can be expected.

Symmetry structures in dynamic models of biochemical systems

Symmetries provide a powerful concept for the development of mechanistic models by describing structures corresponding to the underlying dynamics of biological systems. In this paper, we consider symmetries of the non-linear Hill model describing enzymatic reaction kinetics, and derive a class of symmetry transformations for each order n of the model. We consider a minimal example consisting in the application of symmetry based methods to a model selection problem, where we are able to demonstrate superior performance compared to ordinary residual-based model selection. Finally, we discuss the role of symmetries in systematic model building in systems biology.

Pure donation or hybrid donation crowdfunding

Purpose Despite the growing research exploring the possibility and feasibility of financing socially oriented projects through crowdfunding, relatively little research examines which crowdfunding model is better to serve such purpose. The purpose of this paper is to offer novel insights to mitigate this research gap. Design/methodology/approach A unique data set collected from the largest Chinese crowdfunding platform is used to test the hypotheses. To solve the perceived self-selection problem, the propensity score matching method is adopted in this paper. Based on this approach, the results of similar prosocial campaigns in two different models (pure donation and hybrid donation) are compared. Findings The empirical results show that the hybrid donation model is negatively associated with the status of success and the extent of success of prosocial campaigns. Specifically, compared to the pure donation model, hybrid donation model leads to a lower probability of success, fewer contributors, a lower funding amount and a lower completion ratio. Originality/value The paper contributes to a relatively understudied theme in crowdfunding, namely, donations. It does so by introducing the concepts of pure vs hybrid donation models and investigates the model selection problem in financing social projects through crowdfunding based on cognitive evaluation theory.

Hidden Node Detection between Observable Nodes Based on Bayesian Clustering

Structure learning is one of the main concerns in studies of Bayesian networks. In the present paper, we consider networks consisting of both observable and hidden nodes, and propose a method to investigate the existence of a hidden node between observable nodes, where all nodes are discrete. This corresponds to the model selection problem between the networks with and without the middle hidden node. When the network includes a hidden node, it has been known that there are singularities in the parameter space, and the Fisher information matrix is not positive definite. Then, the many conventional criteria for structure learning based on the Laplace approximation do not work. The proposed method is based on Bayesian clustering, and its asymptotic property justifies the result; the redundant labels are eliminated and the simplest structure is detected even if there are singularities.

The Quality of the Covariance Selection through Detection Problem and AUC Bounds

This paper considers the problem of quantifying the quality of a model selection problem for a graphical model. The model selection problem often uses a distance measure such as the Kulback-Leibler (KL) distance to quantify the quality of the approximation between the original distribution and the model distribution. We extend this work by formulating the problem as a detection problem between the original distribution and the model distribution. In particular, we focus on the covariance selection problem by Dempster, [1], and consider the cases where the distributions are Gaussian distributions. Previous work showed that if the approximation model is a tree, that the optimal tree that minimizes the KL divergence can be found by using the Chow-Liu algorithm [2]. While the algorithm minimizes the KL divergence it does not minimize other measures such as other divergences and the area under the curve (AUC). These measures all depend on the eigenvalues of the correlation approximation measure (CAM). We find expressions for KL divergence, log-likelihood ratio, and AUC as a function of the CAM. Easily computable upper and lower bounds are also found for the AUC. The paper concludes by computing these measures for real and synthetic simulation data.