Joachim van der Herten

A Fuzzy Hybrid Sequential Design Strategy for Global Surrogate Modeling of High-Dimensional Computer Experiments

Complex real-world systems can accurately be modeled by simulations. Evaluating high-fidelity simulators can take several days, making them impractical for use in optimization, design space exploration, and analysis. Often, these simulators are approximated by relatively simple math known as a surrogate model. The data points to construct this model are simulator evaluations meaning the choice of these points is crucial: each additional data point can be very expensive in terms of computing time. Sequential design strategies offer a huge advantage over one-shot experimental design because information gathered from previous data points can be used in the process of determining new data points. Previously, LOLA-Voronoi was presented as a hybrid sequential design method which balances exploration and exploitation: the former involves selecting data points in unexplored regions of the design space, while the latter suggests adding data points in interesting regions which were previously discovered. Although t...

Predictive modelling of survival and length of stay in critically ill patients using sequential organ failure scores

INTRODUCTION The length of stay of critically ill patients in the intensive care unit (ICU) is an indication of patient ICU resource usage and varies considerably. Planning of postoperative ICU admissions is important as ICUs often have no nonoccupied beds available. PROBLEM STATEMENT Estimation of the ICU bed availability for the next coming days is entirely based on clinical judgement by intensivists and therefore too inaccurate. For this reason, predictive models have much potential for improving planning for ICU patient admission. OBJECTIVE Our goal is to develop and optimize models for patient survival and ICU length of stay (LOS) based on monitored ICU patient data. Furthermore, these models are compared on their use of sequential organ failure (SOFA) scores as well as underlying raw data as input features. METHODOLOGY Different machine learning techniques are trained, using a 14,480 patient dataset, both on SOFA scores as well as their underlying raw data values from the first five days after admission, in order to predict (i) the patient LOS, and (ii) the patient mortality. Furthermore, to help physicians in assessing the prediction credibility, a probabilistic model is tailored to the output of our best-performing model, assigning a belief to each patient status prediction. A two-by-two grid is built, using the classification outputs of the mortality and prolonged stay predictors to improve the patient LOS regression models. RESULTS For predicting patient mortality and a prolonged stay, the best performing model is a support vector machine (SVM) with GA,D=65.9% (area under the curve (AUC) of 0.77) and GS,L=73.2% (AUC of 0.82). In terms of LOS regression, the best performing model is support vector regression, achieving a mean absolute error of 1.79 days and a median absolute error of 1.22 days for those patients surviving a nonprolonged stay. CONCLUSION Using a classification grid based on the predicted patient mortality and prolonged stay, allows more accurate modeling of the patient LOS. The detailed models allow to support the decisions made by physicians in an ICU setting.

Adaptive classification under computational budget constraints using sequential data gathering

The extension of the SUMO-Toolbox for classification is presented.State-of-the-art tool for modeling and optimization of budget-constrained problems.Includes the Neighborhood-Voronoi method for sequential sampling.Illustrative examples, including complex expensive constrained optimization. Classification algorithms often handle large amounts of labeled data. When a label is the result of a very expensive computer experiment (in terms of computational time), sequential selection of samples can be used to limit the overall cost of acquiring the labeled data. This paper outlines the concept of sequential design for classification, and the extension of an existing state-of-the-art research platform for surrogate modeling to handle classification problems with sequential design. The capabilities of the platform are illustrated on a number of use cases including real-world applications such as an ElectroMagnetic Compatibility (EMC) and a Computational Fluid Dynamics (CFD) problem. The CFD problem also illustrates how classification can be used together with regression techniques to solve multi-objective constrained optimization problems of complex systems.

Random Survival Forests for Predicting the Bed Occupancy in the Intensive Care Unit

Predicting the bed occupancy of an intensive care unit (ICU) is a daunting task. The uncertainty associated with the prognosis of critically ill patients and the random arrival of new patients can lead to capacity problems and the need for reactive measures. In this paper, we work towards a predictive model based on Random Survival Forests which can assist physicians in estimating the bed occupancy. As input data, we make use of the Sequential Organ Failure Assessment (SOFA) score collected and calculated from 4098 patients at two ICU units of Ghent University Hospital over a time period of four years. We compare the performance of our system with a baseline performance and a standard Random Forest regression approach. Our results indicate that Random Survival Forests can effectively be used to assist in the occupancy prediction problem. Furthermore, we show that a group based approach, such as Random Survival Forests, performs better compared to a setting in which the length of stay of a patient is individually assessed.

ALBATROS: adaptive line-based sampling trajectories for sequential measurements

Measurements in 2D or 3D spaces are ubiquitous among many fields of science and engineering. Often, data samples are gathered via autonomous robots or drones. The path through the measurement space and the location of the samples is traditionally determined upfront using a one-shot design of experiments. However, in certain cases, a sequential approach is preferred. For example, when dealing with a limited sampling budget or when a quick low-resolution overview is desired followed by a steady uniform increase in sampling density, instead of a slow high-resolution one-shot sampling. State-of-the-art sequential design of experiment methods are point-based and are often used to set up experiments both in virtual (simulation) as well as real-world (measurement) environments. In contrast to virtual experimentation, physical measurements require movement of a sensor probe through the measurement space. In these cases, the algorithm not only needs to optimize the sample locations and order but also the path to be traversed by sampling points along measurement lines. In this work, a sequential line-based sampling method is proposed which aims to gradually increase the sampling density across the entire measurement space while minimizing the overall path length. The algorithm is illustrated on a 2D and 3D unit space as well as a complex 3D space and the effectiveness is validated on an engineering measurement use-case. A computer code implementation of the algorithm is provided as an open-source toolbox.

Surrogate Modelling with Sequential Design for Expensive Simulation Applications

The computational demands of virtual experiments for modern product development processes can get out of control due to fine resolution and detail incorporation in simu‐ lation packages. These demands for appropriate approximation strategies and reliable selection of evaluations to keep the amount of required evaluations were limited, without compromising on quality and requirements specified upfront. Surrogate models provide an appealing data‐driven strategy to accomplish these goals for applications including design space exploration, optimization, visualization or sensitivity analysis. Extended with sequential design, satisfactory solutions can be identified quickly, greatly motivat‐ ing the adoption of this technology into the design process.

Adaptive modeling and sampling methodologies for Internet of Things applications

The past few years several cloud services offer automated machine learning software. This enables non-experts to build sophisticated predictive models so they can focus on their area of expertise instead, and use these state-of-the-art machine learning techniques. These were the same principles that guided the development of the surrogate modeling (SUMO) toolbox to assist engineers during (virtual) product design and rapid prototyping with state-of-the-art machine learning methods. A proof of concept was developed, which exposes the technologies of the SUMO toolbox as a network service, offering them to the devices attached to the same network. Both the implementation of the service as well as the possibilities for Internet of Things are discussed.

Constructing classifiers of expensive simulation-based data by sequential experimental design

Sequential experimental design for computer experiments is frequently used to construct surrogate regression models of complex blackbox simulators when evaluations are expensive. The same methodology can be used to train classifiers of labeled data which is expensive to obtain. For certain problems classification can be a more appropriate method to obtain a solution with fewer samples.