Nonparametric Machine Learning Research Articles

Fusion and Visualization of Bridge Deck Nondestructive Evaluation Data via Machine Learning

To maintain infrastructure safety and integrity, nondestructive evaluation (NDE) technologies are often used for detection of subsurface defects and for holistic condition assessment of structures. While the rapid advances in data collection and the diversity of available sensing technologies provide new opportunities, the ability to efficiently process data and combine heterogeneous data sources to make robust decisions remains a challenge. Heterogeneous NDE measurements often conflict with one another and methods to visualize integrated results are often developed ad hoc. In this work, we present a framework to support fusion of multiple NDE techniques in order to improve both detection and quantification accuracy while also improving the visualization of NDE results. For data sources with waveform representations, the discrete wavelet transform (DWT) is used to extract salient features and facilitate fusion with scalar-valued NDE measurements. The description of a signal in terms of its salient features using a wavelet transform allows for capturing the significance of the original data, while suppressing measurement noise. The complete set of measurements is then fused using nonparametric machine learning so as to relax the need for Bayesian assumptions regarding statistical distributions. A novel visualization schema based on classifier confidence intervals is then employed to support holistic visualization and decision making. To validate the capabilities of the proposed methodology, an experimental prototype system was created and tested from NDE measurements of laboratory-scale bridge decks at Turner-Fairbank highway research center (TFHRC). The laboratory decks exhibit various types of artificial defects and several non-destructive tests have already been carried out by research center technicians to characterize the mechanical properties of the materials and the existing damages. The behaviors of several different data fusion approaches are compared here. Based on the presented analyses, it is demonstrated that fusion via support vector machines provided the most robust and consistent data fusion and defect detection capabilities.

Surface Electromyography-Based Muscle Fatigue Analysis Using Binary and Weighted Visibility Graph Features

Surface electromyography (sEMG) is a non-invasive technique to assess the electrical activity of contracting skeletal muscles. sEMG-based muscle fatigue detection plays a key role in sports medicine, ergonomics and rehabilitation. These signals are random, multicomponent, nonlinear and the degree of fluctuations is higher in dynamic contractions. Hence, the extraction of reliable biomarkers remains a challenging task. In this work, an attempt has been made to differentiate non-fatigue, and fatigue conditions using nonlinear techniques, namely, binary and weighted Visibility Graph (VG) features. For this, signals are recorded from the biceps brachii muscle of 52 healthy adult volunteers. These signals are preprocessed, and the contractions associated with the non-fatigue and fatigue conditions are segmented. The graph transformation is performed, and first-order and second-order statistics, along with entropy measures, are extracted from the degree distribution. Parametric and non-parametric machine learning methods are applied for the classification. The results show that the proposed VG approach is able to capture the fluctuations of the signals in non-fatigue and fatigue conditions. Further, all extracted features exhibit a significant difference with [Formula: see text] [Formula: see text]. Maximum accuracy of 89.1% is achieved with information gain selected features and extreme learning machines classifier. Additionally, weighted VG features perform better than the binary version with a difference in the accuracy of 5%. It appears that the proposed approach could be used in real-time implementation for the monitoring of muscle fatigue conditions.

Early Warning Signs: Targeting Neonatal and Infant Mortality Using Machine Learning

This paper builds predictive models for the incidences of neonatal and infant mortality using multiple parametric and non-parametric Machine Learning (ML) techniques. The consensus of the top predictors from the interpret-able ML algorithms (that we use) serve as leading indicators of neonatal and infant mortality and enables us to identify a ‘high-mortality risk’ group of mothers and infants using a household survey data from India. Given the imbalance nature of the data (‘rare-event’ problem) as robustness check we use additional ML methods that are tailor-made for the purpose. The identification of this at-risk population is an important aspect of the government of India’s ‘India Newborn Action Plan’ which has been launched as part of World Health Organization’s Global ‘Every Newborn Action Plan’ and is particularly useful for the creation of targeted public health policies. In addition to identifying some socio-economic, health and behavioral characteristics/predictors of the target group, our analysis also sheds lights on some policy relevance.

Machine Learning for Plant Breeding and Biotechnology

Classical univariate and multivariate statistics are the most common methods used for data analysis in plant breeding and biotechnology studies. Evaluation of genetic diversity, classification of plant genotypes, analysis of yield components, yield stability analysis, assessment of biotic and abiotic stresses, prediction of parental combinations in hybrid breeding programs, and analysis of in vitro-based biotechnological experiments are mainly performed by classical statistical methods. Despite successful applications, these classical statistical methods have low efficiency in analyzing data obtained from plant studies, as the genotype, environment, and their interaction (G × E) result in nondeterministic and nonlinear nature of plant characteristics. Large-scale data flow, including phenomics, metabolomics, genomics, and big data, must be analyzed for efficient interpretation of results affected by G × E. Nonlinear nonparametric machine learning techniques are more efficient than classical statistical models in handling large amounts of complex and nondeterministic information with “multiple-independent variables versus multiple-dependent variables” nature. Neural networks, partial least square regression, random forest, and support vector machines are some of the most fascinating machine learning models that have been widely applied to analyze nonlinear and complex data in both classical plant breeding and in vitro-based biotechnological studies. High interpretive power of machine learning algorithms has made them popular in the analysis of plant complex multifactorial characteristics. The classification of different plant genotypes with morphological and molecular markers, modeling and predicting important quantitative characteristics of plants, the interpretation of complex and nonlinear relationships of plant characteristics, and predicting and optimizing of in vitro breeding methods are the examples of applications of machine learning in conventional plant breeding and in vitro-based biotechnological studies. Precision agriculture is possible through accurate measurement of plant characteristics using imaging techniques and then efficient analysis of reliable extracted data using machine learning algorithms. Perfect interpretation of high-throughput phenotyping data is applicable through coupled machine learning-image processing. Some applied and potentially applicable capabilities of machine learning techniques in conventional and in vitro-based plant breeding studies have been discussed in this overview. Discussions are of great value for future studies and could inspire researchers to apply machine learning in new layers of plant breeding.

ENSO Dynamics, Trends, and Prediction Using Machine Learning

AbstractEl Niño–Southern Oscillation (ENSO) has global effects on the hydrological cycle, agriculture, ecosystems, health, and society. We present a novel nonhomogeneous hidden Markov model (NHMM) for studying the underlying dynamics of sea surface temperature anomalies (SSTA) over the region 15°N–15°S, 150°E–80°W from January 1856 to December 2019, using the monthly SSTA data from the Kaplan extended SST v2 product. This nonparametric machine learning scheme dynamically simulates and predicts the spatiotemporal evolution of ENSO patterns, including their asymmetry, long-term trends, persistence, and seasonal evolution. The model identifies five hidden states whose spatial SSTA patterns are similar to the so-called ENSO flavors in the literature. From the fitted NHMM, the model shows that there are systematic trends in the frequency and persistence of the regimes over the last 160 years that may be related to changes in the mean state of basin temperature and/or global warming. We evaluated the ability of NHMM to make out-of-sample probabilistic predictions of the spatial structure of temperature anomalies for the period 1995–2016 using a training period from January 1856 to December 1994. The results show that NHMMs can simulate the behavior of the Niño-3.4 and Niño-1.2 regions quite well. The NHMM results over this period are comparable or superior to the commonly available ENSO prediction models, with the additional advantage of directly providing insights as to the space patterns, seasonal, and longer-term trends of the SSTA in the equatorial Pacific region.

Lazily Adapted Constant Kinky Inference for nonparametric regression and model-reference adaptive control

Techniques known as Nonlinear Set Membership prediction or Lipschitz Interpolation are approaches to supervised machine learning that utilise presupposed Lipschitz properties to perform inference over unobserved function values. Provided a bound on the true best Lipschitz constant of the target function is known a priori, they offer convergence guarantees, as well as bounds around the predictions. Considering a more general setting that builds on Lipschitz continuity, we propose an online method for estimating the Lipschitz constant online from function value observations that are possibly corrupted by bounded noise. Utilising this as a data-dependent hyper-parameter gives rise to a nonparametric machine learning method, for which we establish strong universal approximation guarantees. That is, we show that our prediction rule can learn any continuous function on compact support in the limit of increasingly dense data, up to a worst-case error that can be bounded by the level of observational error. We also consider applications of our nonparametric regression method to learning-based control. For a class of discrete-time settings, we establish convergence guarantees on the closed-loop tracking error of our online learning-based controllers. To provide evidence that our method can be beneficial not only in theory but also in practice, we apply it in the context of nonparametric model-reference adaptive control (MRAC). Across a range of simulated aircraft roll-dynamics and performance metrics our approach outperforms recently proposed alternatives that were based on Gaussian processes and RBF-neural networks.

Not There Yet: Using Data‐Driven Methods to Predict Who Becomes Costly Among Low‐Cost Patients with Type 2 Diabetes

Research ObjectiveDiabetes is a leading cause of Medicare spending, and predicting which individuals are likely to have high health care costs is essential for targeting interventions. Current approaches to predicting spending generally focus on composite measures, short time horizons, or patients who are already high utilizers and, therefore, whose costs may be harder to modify. In contrast, dynamic approaches that model monthly spending over time may help discriminate between low‐cost patients with diabetes who do and do not subsequently become costly.Study DesignWe identified Medicare beneficiaries with type 2 diabetes whose spending was in the bottom 90% of spending on diabetes care during a one‐year baseline period, a threshold for low‐cost patients used in prior studies. We used group‐based trajectory modeling, a data‐driven method, to classify unique clusters of these low‐cost patients by their diabetes spending patterns over the subsequent two‐year follow‐up period. Prediction models for these spending groups were then estimated with generalized boosted regression, a nonparametric machine learning method, using sets of all baseline predictors, diabetes predictors, and predictors that are potentially‐modifiable through interventions. For clinical context, we explored the relative influence of each predictor from the regression models to provide insight into baseline factors that may help distinguish patients who may become costly later. We evaluated each model through discrimination (C‐statistic) measures and split‐sample validation.Population StudiedWe identified Medicare fee‐for‐service beneficiaries with type 2 diabetes from a random, nationwide sample from 2011 to 2014 and used their enrollment, medical, and prescription claims data.Principal FindingsBaseline median diabetes spending for the 33 789 Medicare beneficiaries classified as low cost was $4153. We identified five distinct diabetes spending patterns among these individuals during the two‐year follow‐up period, including those in three groups with different, yet consistent, spending levels (68.1% of beneficiaries), a group with spending that rose quickly (25.3%), and a group with spending that rose progressively (6.6%). The ability to predict these spending groups was moderate to strong, with validated C‐statistics ranging from 0.63 to 0.87. The most influential factors for those in the group with progressively rising spending, in particular, were their age, generosity of Medicare insurance coverage, prior spending on diabetes care, and average adherence to medications.ConclusionsUsing data‐driven approaches, we identified distinct diabetes spending patterns among a nationally‐representative cohort of Medicare patients who were initially low utilizers. These patterns could be predicted using baseline characteristics, including factors that may be potentially modifiable.Implications for Policy or PracticeMany health care organizations use claims data to identify and predict patients for cost‐containment interventions. The approach we describe could inform the design and timing of cost‐containment interventions and target them to those at greatest need among patients with type 2 diabetes or other costly, chronic diseases.Primary Funding SourceNational Institute for Health Care Management (NIHCM) Foundation.

Dissecting Time-Evolved Conductance Behavior of Single Molecule Junctions by Nonparametric Machine Learning.

Improved understanding of charge transport in single molecules is essential for utilizing their potential as circuit components at the nanosize limit. However, reliable analyses of varying tunneling current acquired by break junction experiments remain an ongoing challenge to find molecular feature structure-property relationships. In this work, we report on an unsupervised learning approach for investigating molecular signatures in conductance traces. Our hybrid machine learning algorithm compares grids of data in conductance-time domains and judges the similarity without any researcher-crafted parameters to identify fine molecular components that may otherwise be obscured by background fluctuations. We demonstrate its ability for classifying Au-alkanedithiol-Au conductance traces acquired with microfabricated mechanically controllable break junctions. The unbiased procedure was able to not only judge the presence or absence of the carbon chains in the electrode gap but also to identify multiple conductance states of the molecular tunneling junctions with different conformations. This finding may offer a useful tool for studying single-molecule properties using break junction methods.

Advancement of weather-related crash prediction model using nonparametric machine learning algorithms

This paper evaluates the machine learning-based weather-related crash prediction model in Connecticut. Crash severity prediction has always been the principal focus of safety professionals and emergency responders for appropriate policy making and resource management. Over the years, different statistical methodologies (e.g., random forest, support vector machine) have been explored in various research efforts to develop efficient crash severity prediction models. As technology is advancing and computing has started becoming more efficient, machine learning-based models for crash severity prediction are being brought into light for more accurate data-driven prediction. However, some machine learning methodologies provide increased efficiency and better performance compared to others from the same genre. To explore different machine learning methodologies for crash severity prediction, this study considered two machine learning applications—random forest (RF) and bayesian additive regression trees (BART) for performance comparison. RF model produced higher prediction probabilities for determining crash severity compare to BART. The results were evaluated using various prediction probability analyses obtained from these two machine learning models and showed the model capabilities to generate prediction consistent with the observed data. We found the performance of the RF model to be highly promising with a higher skill score (0.73) than BART (0.61). Overall, our findings demonstrate the robust performances of the RF algorithm in predicting weather-related crashes. The analysis of this study confirms that stakeholders can use weather-related RF model with confidence to obtain a better prediction of crash severity that enables them to facilitate appropriate emergency responses and support essential preemptive measures.

A Hybrid Predictive Approach for Chromium Layer Thickness in the Hard Chromium Plating Process Based on the Differential Evolution/Gradient Boosted Regression Tree Methodology

The purpose of the industrial process of chromium plating is the creation of a hard and wear-resistant layer of chromium over a metallic surface. One of the main properties of chromium plating is its resistance to both wear and corrosion. This research presents an innovative nonparametric machine learning approach that makes use of a hybrid gradient boosted regression tree (GBRT) methodology for hard chromium layer thickness prediction. GBRT is a non-parametric statistical learning technique that produces a prediction model in the form of an ensemble of weak prediction models. The motivation for boosting is a procedure that combines the output of many weak classifiers to produce a powerful committee. In this study, the GBRT hyperparameters were optimized with the help of differential evolution (DE). DE is an optimization technique within evolutionary computing. The results found that this model was able to predict the thickness of the chromium layer formed in this industrial process with a determination coefficient equal to 0.9842 and a root-mean-square error value of 0.01590. The two most important variables of the model were the time of the hard-chromium process and the thickness of the layer removed by electropolishing. Thus, these results provide a foundation for an accurate predictive model of hard chromium layer thickness. The derived model also allowed the ranking of the importance of the independent input variables that were examined. Finally, the high performance and simplicity of the model make the DE/GBRT method attractive compared to conventional forecasting techniques.

Deep Gaussian processes for biogeophysical parameter retrieval and model inversion

Parameter retrieval and model inversion are key problems in remote sensing and Earth observation. Currently, different approximations exist: a direct, yet costly, inversion of radiative transfer models (RTMs); the statistical inversion with in situ data that often results in problems with extrapolation outside the study area; and the most widely adopted hybrid modeling by which statistical models, mostly nonlinear and non-parametric machine learning algorithms, are applied to invert RTM simulations. We will focus on the latter. Among the different existing algorithms, in the last decade kernel based methods, and Gaussian Processes (GPs) in particular, have provided useful and informative solutions to such RTM inversion problems. This is in large part due to the confidence intervals they provide, and their predictive accuracy. However, RTMs are very complex, highly nonlinear, and typically hierarchical models, so that very often a single (shallow) GP model cannot capture complex feature relations for inversion. This motivates the use of deeper hierarchical architectures, while still preserving the desirable properties of GPs. This paper introduces the use of deep Gaussian Processes (DGPs) for bio-geo-physical model inversion. Unlike shallow GP models, DGPs account for complicated (modular, hierarchical) processes, provide an efficient solution that scales well to big datasets, and improve prediction accuracy over their single layer counterpart. In the experimental section, we provide empirical evidence of performance for the estimation of surface temperature and dew point temperature from infrared sounding data, as well as for the prediction of chlorophyll content, inorganic suspended matter, and coloured dissolved matter from multispectral data acquired by the Sentinel-3 OLCI sensor. The presented methodology allows for more expressive forms of GPs in big remote sensing model inversion problems.

Quantification of macropore flow in Danish soils using near-saturated hydraulic properties

Larger pores allowing macropore flow can rapidly transport contaminants as well as nutrients from the soil surface to the subsoil and groundwater. Modeling of the involved macropore flow is complex due to large number of required parameters. One way to deal with this difficulty is the precise evaluation of the soil hydraulic conductivity function using robust pedotransfer functions for the near-saturated part. Aiming to develop pedotransfer functions to estimate saturated and unsaturated hydraulic conductivities [Ks and K(h)] as well as water contents at specific pressure heads, we used Gaussian Process Regression (GPR), a non-parametric machine learning model, to obtain van Genuchten (1980) retention and conductivity parameters for Danish soils based on soil texture and organic matter. We defined K10 as K(h) at − 10 cm pressure head. The difference between the logarithm of Ks and K10 [log(Ks) − log(K10)] was denominated log(Kjump) and was assumed to convey the potential degree of macropore flow considering only soil hydraulic properties. Macropore flow in Denmark was evaluated through the developed PTFs for log(Ks) and log(K10) with an average RMSE of 0.635 and 0.594 (for K in cm d−1) under a confidence interval obtained by bootstrapping. Dynamics of pressure head in an initially saturated soil column with a zero-flux top boundary condition and a 2 m deep groundwater level were simulated using the Hydrus-1D model. The result of this modeling showed that soils in the eastern part of Denmark had a higher likelihood to experience macropore flow. Finally, using the spatial distribution of meteorological data, a macropore flow map of Denmark was produced. Precipitation was a dominating factor on macropore flow in sandy clay loam and sandy loam soils when compared to sandy and loamy sand soils. Finally, maps of relative risk classes of macropore flow qualitatively differentiated land areas in Denmark for vulnerability of nutrient loss and groundwater contamination.

Nonparametric machine learning for mapping forest cover and exploring influential factors

The contribution of forest ecosystem services to human well-being varies over space following the dynamics in forest cover. Use of machine learning models is increasing in projecting forest cover changes and investigating the drivers, yet references are still lacking for selecting machine learning models for spatial projection of forest cover patterns. We assessed the ability of nonparametric machine learning techniques to project the spatial distribution of forest cover and identify its drivers using a case study of Tasmania, Australia. We developed, evaluated, and compared the performance of four nonparametric machine learning models: support vector regression (SVR), artificial neural networks (ANN), random forest (RF), and gradient boosted regression trees (GBRT). The results demonstrated that RF far outperformed the other three models in both fitting and projection accuracy, and required less computional costs. GBRT outperformed SVR and ANN in projection accuracy. However, RF exhibited serious overfitting due to the full growth of its decision trees. The influence rankings of explanatory variables on spatial patterns of forest cover were different under the four models. Land tenure type and rainfall were identified among the top four most influential variables by all four models. The ranking produced by the RF model was significantly different with topographic factors associated with land clearing and production costs (elevation and distance to timber facilities) being the two most influential variables. We encourage practitioners to consider nonparametric machine learning methods, especially RF, when facing problems of complex environmental data modelling.

Land Use/Land Cover Mapping Using Multitemporal Sentinel-2 Imagery and Four Classification Methods—A Case Study from Dak Nong, Vietnam

Information on land use and land cover (LULC) including forest cover is important for the development of strategies for land planning and management. Satellite remotely sensed data of varying resolutions have been an unmatched source of such information that can be used to produce estimates with a greater degree of confidence than traditional inventory estimates. However, use of these data has always been a challenge in tropical regions owing to the complexity of the biophysical environment, clouds, and haze, and atmospheric moisture content, all of which impede accurate LULC classification. We tested a parametric classifier (logistic regression) and three non-parametric machine learning classifiers (improved k-nearest neighbors, random forests, and support vector machine) for classification of multi-temporal Sentinel 2 satellite imagery into LULC categories in Dak Nong province, Vietnam. A total of 446 images, 235 from the year 2017 and 211 from the year 2018, were pre-processed to gain high quality images for mapping LULC in the 6516 km2 study area. The Sentinel 2 images were tested and classified separately for four temporal periods: (i) dry season, (ii) rainy season, (iii) the entirety of the year 2017, and (iv) the combination of dry and rainy seasons. Eleven different LULC classes were discriminated of which five were forest classes. For each combination of temporal image set and classifier, a confusion matrix was constructed using independent reference data and pixel classifications, and the area on the ground of each class was estimated. For overall temporal periods and classifiers, overall accuracy ranged from 63.9% to 80.3%, and the Kappa coefficient ranged from 0.611 to 0.813. Area estimates for individual classes ranged from 70 km2 (1% of the study area) to 2200 km2 (34% of the study area) with greater uncertainties for smaller classes.

25 Years of European Merger Control

We study the evolution of EC merger decisions over the first 25 years of common European merger policy. Using a novel dataset at the level of the relevant antitrust markets and containing all merger cases scrutinized by the Commission over the 1990-2014 period, we evaluate how consistently arguments related to structural market parameters – dominance, concentration, barriers to entry, and foreclosure – were applied over time and across different dimensions such as the geographic market definition and the complexity of the merger. Simple, linear probability models as usually applied in the literature overestimate on average the effects of the structural indicators. Using non-parametric machine learning techniques, we find that dominance is positively correlated with competitive concerns, especially in concentrated markets and in complex mergers. Yet, its importance has decreased over time and significantly following the 2004 merger policy reform. The Commission's competitive concerns are also correlated with concentration and the more so, the higher the entry barriers and the risks of foreclosure. These patterns are not changing over time. The role of the structural indicators in explaining competitive concerns does not change depending on the geographic market definition.

Broad Learning System for Nonparametric Modeling of Clay Parameters

Due to the complex and uncertain nature of geomaterial properties, establishing representative parametric models between clay parameters is a challenging task. Nonparametric machine learning offers an accessible approach to develop empirical transformation models of clay parameters based on the available measurement. In this study, nonparametric modeling of clay parameter relationships via broad learning system (BLS) is introduced. Broad learning architecture provides an effective tool for nonparametric modeling based on noise-corrupted data. The architecture of deep learning is configured with stacks of hierarchical layers, which consume expensive computational cost for network training. In contrast, the network of BLS is established in a flat architecture and it can be modified incrementally. As a result, the broad learning flat network can be reconfigured efficiently to accommodate additional training data. To demonstrate the performance of the learning algorithm for clay parameters, the comprehensive global database CLAY/10/7490 with 7,490 data points from over 250 studies worldwide is utilized and analyzed.

Predicting the excretion of feces, urine and nitrogen using support vector regression: A case study with Holstein dry cows

Predicting the excretion of feces, urine and nitrogen (N) from dairy cows is an effective way to prevent and control the environmental pollution caused by scaled farming. The traditional prediction methods such as pollutant generation coefficient (PGC) and mathematical model based on linear regression (LR) may be limited by prediction range and regression function assumption, and sometimes may deviate from the actual condition. In order to solve these problems, the support vector regression (SVR) was applied for predicting the cows' feces, urine and N excretions, taking Holstein dry cows as a case study. SVR is a typical non-parametric machine learning model that does not require any specific assumptions about the regression function in advance and only by learning the training sample data, and also it can fit the function closest to the actual in most cases. To evaluate prediction accuracy effectively, the SVR technique was compared with the LR and radial basis function artificial neural network (RBF-ANN) methods, using the required sample data obtained from actual feeding experiments. The prediction results indicate that the proposed technique is superior to the other two conventional (especially LR) methods in predicting the main indicators of feces, urine, and N excretions of Holstein dry cows. Keywords: cow farming pollution, feces/urine excretion prediction, nitrogen excretion prediction, non-parametric model, SVR technique DOI: 10.25165/j.ijabe.20201302.4781 Citation: Fu Q, Shen W Z, Wei X L, Yin Y L, Zheng P, Zhang Y G, et al. Predicting the excretion of feces, urine and nitrogen using support vector regression: A case study with Holstein dry cows. Int J Agric & Biol Eng, 2020; 13(2): 48–56.

Noise level penalizing robust Gaussian process regression for NIR spectroscopy quantitative analysis

In Near-infrared (NIR) spectroscopy qualitative analysis, noise caused data quality problem has been a bottleneck to further enhance the prediction accuracy. Appropriate preprocessing methods can reduce the influence of noise; and robust models have higher tolerance for noise disturbance. However, these methods treat all the wavelengths equally. In fact, the spectra at different wavelengths may have highly different level of noise. This paper presents a new noise-level-penalizing robust Gaussian process (NLP-RGP) regression for NIR spectroscopy quantitative analysis. The novel noise level penalizing mechanism penalize the spectra features according to their noise level, i.e., encourage the model to prefer the less noisy features over high noisy features. Gaussian process (GP) is a nonparametric machine learning method based on kernel and Bayesian inference framework; with a noise model of heavy-tailed distribution, robust Gaussian process can handle the abnormal sample data better. Experiments were taken on the determination of the total soluble solids content of navel oranges based on their surface NIR spectra. The NLP-GP outperforms the robust Gaussian process model and least squares support vector machines (LS-SVM), the state of art method. Moreover, the NLP-RGP performs even better than the NLP-GP, achieving the best prediction accuracy among all the models. This demonstrates the effectiveness of noise level penalizing mechanism, and the noise level penalizing mechanism and robust mechanism of Gaussian process can be integrated together well.

Robust learning-based MPC for nonlinear constrained systems

This paper presents a robust learning-based predictive control strategy for nonlinear systems subject to both input and output constraints, under the assumption that the model function is not known a priori and only input–output data are available. The proposed controller is obtained using a nonparametric machine learning technique to estimate a prediction model. Based on this prediction model, a novel stabilizing robust predictive controller without terminal constraint is proposed. The design procedure is purely based on data and avoids the estimation of any robust invariant set, which is in general a hard task. The resulting controller has been validated in a simulated case study.

Gaussian Process Panel Modeling-Machine Learning Inspired Analysis of Longitudinal Panel Data.

In this article, we extend the Bayesian nonparametric regression method Gaussian Process Regression to the analysis of longitudinal panel data. We call this new approach Gaussian Process Panel Modeling (GPPM). GPPM provides great flexibility because of the large number of models it can represent. It allows classical statistical inference as well as machine learning inspired predictive modeling. GPPM offers frequentist and Bayesian inference without the need to resort to Markov chain Monte Carlo-based approximations, which makes the approach exact and fast. GPPMs are defined using the kernel-language, which can express many traditional modeling approaches for longitudinal data, such as linear structural equation models, multilevel models, or state-space models but also various commonly used machine learning approaches. As a result, GPPM is uniquely able to represent hybrid models combining traditional parametric longitudinal models and nonparametric machine learning models. In the present paper, we introduce GPPM and illustrate its utility through theoretical arguments as well as simulated and empirical data.