Miha Mlakar - Academia.edu (original) (raw)

Papers by Miha Mlakar

Studies in Health Technology and Informatics, 2018

Physical fitness is important in view of reducing risks for a number of non-communicable diseases... more Physical fitness is important in view of reducing risks for a number of non-communicable diseases, both for individuals and policy-makers. In this paper, we present a prototype tool that combines forecasting of individual fitness parameters of schoolchildren to early adulthood with estimation of relative risk for all-cause early mortality in adulthood based on the forecasted fitness. This tool is a first step in the development of a platform that will show age, gender, and geographical distributions of risk and suggest potential interventions.

Copyright © 2014 Miha Mlakar et al. This is an open access article distributed under the Creative... more Copyright © 2014 Miha Mlakar et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Due to various reasons the solutions in real-world optimization problems cannot always be exactly evaluated but are sometimes represented with approximated values and confidence intervals. In order to address this issue, the comparison of solutions has to be done differently than for exactly evaluated solutions. In this paper, we define new relations under uncertainty between solutions in multiobjective optimization that are represented with approximated values and confidence intervals. The new relations extend the Pareto dominance relations, can handle constraints, and can be used to compare solutions, both with and without the confidence interval. We also show that by including confidence intervals into the comparisons, the possibilit...

In surrogate-model-based optimization, the selection of an appropriate surrogate model is very im... more In surrogate-model-based optimization, the selection of an appropriate surrogate model is very important. If so-lution approximations returned by a surrogate model are accurate and with narrow confidence intervals, an algo-rithm using this surrogate model needs less exact solu-tion evaluations to obtain results comparable to an algo-rithm without surrogate models. In this paper we com-pare two well known modeling techniques, random forest (RF) and Gaussian process (GP) modeling. The compar-ison includes the approximation accuracy and confidence in the approximations (expressed as the confidence inter-val width). The results show that GP outperforms RF and that it is more suitable for use in a surrogate-model-based multiobjective evolutionary algorithm. 1

Informatica (Slovenia), 2015

This paper presents a summary of the doctoral dissertation of the author, which addresses the tas... more This paper presents a summary of the doctoral dissertation of the author, which addresses the task of evolutionary multiobjective optimization using surrogate models. The main contributions are done for the optimization problems, where solutions are presented with uncertainty. To compare solutions under uncertainty and improve the optimization results the new relations for comparing solutions under uncertainty are defined. These relations reduce the possibility of incorrect comparisons due to the inaccurate approximations. The relations under uncertainty are then used in the new surrogate-model-based multiobjective evolutionary algorithm called GP-DEMO. The algorithm is thoroughly tested on benchmark and real-world problems and the results show that GP-DEMO, in comparison to other multiobjective evolutionary algorithms, produces comparable results while requiring fewer exact evaluations of the original objective functions.

The Sussex-Huawei Locomotion-Transportation Recognition Challenge presented a unique opportunity ... more The Sussex-Huawei Locomotion-Transportation Recognition Challenge presented a unique opportunity to the activity-recognition community to test their approaches on a large, real-life benchmark dataset with activities different from those typically being recognized. The goal of the challenge was to recognize eight locomotion activities (Still, Walk, Run, Bike, Car, Bus, Train, Subway). This chapter describes the submissions winning the first and second place. They both start with data preprocessing, including a normalization of the phone orientation. Then, a wide set of hand-crafted domain features in both frequency and time domain are computed and their quality evaluated. The second-place submission feeds the best features into an XGBoost machine-learning model with optimized hyper-parameters, achieving the accuracy of 90.2%. The first-place submission builds an ensemble of models, including deep learning models, and finally refines the ensemble’s predictions by smoothing with a Hidd...

JMIR Medical Informatics

Background Congestive heart failure (CHF) is a disease that requires complex management involving... more Background Congestive heart failure (CHF) is a disease that requires complex management involving multiple medications, exercise, and lifestyle changes. It mainly affects older patients with depression and anxiety, who commonly find management difficult. Existing mobile apps supporting the self-management of CHF have limited features and are inadequately validated. Objective The HeartMan project aims to develop a personal health system that would comprehensively address CHF self-management by using sensing devices and artificial intelligence methods. This paper presents the design of the system and reports on the accuracy of its patient-monitoring methods, overall effectiveness, and patient perceptions. Methods A mobile app was developed as the core of the HeartMan system, and the app was connected to a custom wristband and cloud services. The system features machine learning methods for patient monitoring: continuous blood pressure (BP) estimation, physical activity monitoring, and...

The paper presents a multiobjective optimization approach to process parameter optimization in co... more The paper presents a multiobjective optimization approach to process parameter optimization in continuous casting of steel, which is the most widely used steel production process. The optimization task is to find parameter values such that the target values of the empirical metallurgical optimization criteria are approached as closely as possible, since this in turn results in high quality of the cast steel. The problem is being solved with a multiobjective evolutionary algorithm coupled with a numerical simulator of the casting process. The resulting trade-o↵ solutions are visualized to support decision-making about the preferred solutions.

The present-day evolutionary multi-objective optimization (EMO) algorithms had a demonstrated his... more The present-day evolutionary multi-objective optimization (EMO) algorithms had a demonstrated history of evolution over the years. The initial EMO methodologies involved additional niching parameters which made them somewhat subjective to the user. Fortunately, soon enough parameter-less EMO methodologies have been suggested thereby making the earlier EMO algorithms unpopular and obsolete. In this paper, we present a functional decomposition of a viable EMO methodology and discuss the critical components which require special attention for making the complete algorithm free from any additional parameter. A critical evaluation of existing EMO methodologies suggest that the elitist non-dominated sorting GA (NSGA-II) is one of EMO algorithms which does not require any additional implicit or explicit parameters other than the standard EA parameters, such as population size, operator probabilities, etc. This parameter-less property of NSGA-II is probably the reason for its popularity to most EMO studies thus far.

Proceedings of the 2017 ACM International Joint Conference on Pervasive and Ubiquitous Computing and Proceedings of the 2017 ACM International Symposium on Wearable Computers, 2017

Wearable devices are heavily used in many sports. However, the existing sports wearables are eith... more Wearable devices are heavily used in many sports. However, the existing sports wearables are either not tennis-specific, or are limited to information on shots. We therefore added tennis-specific information to a leading commercial device. Firstly, we developed a method for classifying shot types into forehand, backhand and serve. Secondly, we used multi-objective optimization to distinguish active play from the time in-between points. By combining both parts with the general movement information already provided by the device, we get comprehensive metrics for professional players and coaches to objectively measure a player's performance and enable in-depth tactical analysis.

Proceedings of the 2018 ACM International Joint Conference and 2018 International Symposium on Pervasive and Ubiquitous Computing and Wearable Computers

Proceedings of the 2018 ACM International Joint Conference and 2018 International Symposium on Pervasive and Ubiquitous Computing and Wearable Computers

Journal of Sports Analytics

Taking advantage of space and time is a major focus of tennis coaching yet few statistical measur... more Taking advantage of space and time is a major focus of tennis coaching yet few statistical measures exist to evaluate a player's spatio-temporal performance in matches. The present study proposed the time to net as a single metric capturing both space and time characteristics of the quality of a shot. Tracking data from 2017 Australian Open allowed a detailed investigation of the characteristics and predictive value of the time-to-net in 33,913 men's and 19,195 women's shots. For groundstroke shots, the majority of men's and women's shots have a time-to-net between 200 and 800 ms. The expected time to net was found to vary significantly by gender, shot type, and where in a rally it occurred. We found considerable between-player differences in average time-to-net of groundstrokes when serving or receiving, indicating the potential for time-to-net to capture differences in playing style. Time-to-net increased prediction accuracy of point outcomes by 8 percentage points. These findings show that time to net is a simple spatio-temporal statistic that has descriptive and predictive value for performance analysis in tennis.

Studies in Health Technology and Informatics, 2018

Physical fitness is important in view of reducing risks for a number of non-communicable diseases... more Physical fitness is important in view of reducing risks for a number of non-communicable diseases, both for individuals and policy-makers. In this paper, we present a prototype tool that combines forecasting of individual fitness parameters of schoolchildren to early adulthood with estimation of relative risk for all-cause early mortality in adulthood based on the forecasted fitness. This tool is a first step in the development of a platform that will show age, gender, and geographical distributions of risk and suggest potential interventions.

Copyright © 2014 Miha Mlakar et al. This is an open access article distributed under the Creative... more Copyright © 2014 Miha Mlakar et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Due to various reasons the solutions in real-world optimization problems cannot always be exactly evaluated but are sometimes represented with approximated values and confidence intervals. In order to address this issue, the comparison of solutions has to be done differently than for exactly evaluated solutions. In this paper, we define new relations under uncertainty between solutions in multiobjective optimization that are represented with approximated values and confidence intervals. The new relations extend the Pareto dominance relations, can handle constraints, and can be used to compare solutions, both with and without the confidence interval. We also show that by including confidence intervals into the comparisons, the possibilit...

In surrogate-model-based optimization, the selection of an appropriate surrogate model is very im... more In surrogate-model-based optimization, the selection of an appropriate surrogate model is very important. If so-lution approximations returned by a surrogate model are accurate and with narrow confidence intervals, an algo-rithm using this surrogate model needs less exact solu-tion evaluations to obtain results comparable to an algo-rithm without surrogate models. In this paper we com-pare two well known modeling techniques, random forest (RF) and Gaussian process (GP) modeling. The compar-ison includes the approximation accuracy and confidence in the approximations (expressed as the confidence inter-val width). The results show that GP outperforms RF and that it is more suitable for use in a surrogate-model-based multiobjective evolutionary algorithm. 1

Informatica (Slovenia), 2015

This paper presents a summary of the doctoral dissertation of the author, which addresses the tas... more This paper presents a summary of the doctoral dissertation of the author, which addresses the task of evolutionary multiobjective optimization using surrogate models. The main contributions are done for the optimization problems, where solutions are presented with uncertainty. To compare solutions under uncertainty and improve the optimization results the new relations for comparing solutions under uncertainty are defined. These relations reduce the possibility of incorrect comparisons due to the inaccurate approximations. The relations under uncertainty are then used in the new surrogate-model-based multiobjective evolutionary algorithm called GP-DEMO. The algorithm is thoroughly tested on benchmark and real-world problems and the results show that GP-DEMO, in comparison to other multiobjective evolutionary algorithms, produces comparable results while requiring fewer exact evaluations of the original objective functions.

The Sussex-Huawei Locomotion-Transportation Recognition Challenge presented a unique opportunity ... more The Sussex-Huawei Locomotion-Transportation Recognition Challenge presented a unique opportunity to the activity-recognition community to test their approaches on a large, real-life benchmark dataset with activities different from those typically being recognized. The goal of the challenge was to recognize eight locomotion activities (Still, Walk, Run, Bike, Car, Bus, Train, Subway). This chapter describes the submissions winning the first and second place. They both start with data preprocessing, including a normalization of the phone orientation. Then, a wide set of hand-crafted domain features in both frequency and time domain are computed and their quality evaluated. The second-place submission feeds the best features into an XGBoost machine-learning model with optimized hyper-parameters, achieving the accuracy of 90.2%. The first-place submission builds an ensemble of models, including deep learning models, and finally refines the ensemble’s predictions by smoothing with a Hidd...

JMIR Medical Informatics

Background Congestive heart failure (CHF) is a disease that requires complex management involving... more Background Congestive heart failure (CHF) is a disease that requires complex management involving multiple medications, exercise, and lifestyle changes. It mainly affects older patients with depression and anxiety, who commonly find management difficult. Existing mobile apps supporting the self-management of CHF have limited features and are inadequately validated. Objective The HeartMan project aims to develop a personal health system that would comprehensively address CHF self-management by using sensing devices and artificial intelligence methods. This paper presents the design of the system and reports on the accuracy of its patient-monitoring methods, overall effectiveness, and patient perceptions. Methods A mobile app was developed as the core of the HeartMan system, and the app was connected to a custom wristband and cloud services. The system features machine learning methods for patient monitoring: continuous blood pressure (BP) estimation, physical activity monitoring, and...

The paper presents a multiobjective optimization approach to process parameter optimization in co... more The paper presents a multiobjective optimization approach to process parameter optimization in continuous casting of steel, which is the most widely used steel production process. The optimization task is to find parameter values such that the target values of the empirical metallurgical optimization criteria are approached as closely as possible, since this in turn results in high quality of the cast steel. The problem is being solved with a multiobjective evolutionary algorithm coupled with a numerical simulator of the casting process. The resulting trade-o↵ solutions are visualized to support decision-making about the preferred solutions.

The present-day evolutionary multi-objective optimization (EMO) algorithms had a demonstrated his... more The present-day evolutionary multi-objective optimization (EMO) algorithms had a demonstrated history of evolution over the years. The initial EMO methodologies involved additional niching parameters which made them somewhat subjective to the user. Fortunately, soon enough parameter-less EMO methodologies have been suggested thereby making the earlier EMO algorithms unpopular and obsolete. In this paper, we present a functional decomposition of a viable EMO methodology and discuss the critical components which require special attention for making the complete algorithm free from any additional parameter. A critical evaluation of existing EMO methodologies suggest that the elitist non-dominated sorting GA (NSGA-II) is one of EMO algorithms which does not require any additional implicit or explicit parameters other than the standard EA parameters, such as population size, operator probabilities, etc. This parameter-less property of NSGA-II is probably the reason for its popularity to most EMO studies thus far.

Proceedings of the 2017 ACM International Joint Conference on Pervasive and Ubiquitous Computing and Proceedings of the 2017 ACM International Symposium on Wearable Computers, 2017

Wearable devices are heavily used in many sports. However, the existing sports wearables are eith... more Wearable devices are heavily used in many sports. However, the existing sports wearables are either not tennis-specific, or are limited to information on shots. We therefore added tennis-specific information to a leading commercial device. Firstly, we developed a method for classifying shot types into forehand, backhand and serve. Secondly, we used multi-objective optimization to distinguish active play from the time in-between points. By combining both parts with the general movement information already provided by the device, we get comprehensive metrics for professional players and coaches to objectively measure a player's performance and enable in-depth tactical analysis.

Proceedings of the 2018 ACM International Joint Conference and 2018 International Symposium on Pervasive and Ubiquitous Computing and Wearable Computers

Proceedings of the 2018 ACM International Joint Conference and 2018 International Symposium on Pervasive and Ubiquitous Computing and Wearable Computers

Journal of Sports Analytics

Taking advantage of space and time is a major focus of tennis coaching yet few statistical measur... more Taking advantage of space and time is a major focus of tennis coaching yet few statistical measures exist to evaluate a player's spatio-temporal performance in matches. The present study proposed the time to net as a single metric capturing both space and time characteristics of the quality of a shot. Tracking data from 2017 Australian Open allowed a detailed investigation of the characteristics and predictive value of the time-to-net in 33,913 men's and 19,195 women's shots. For groundstroke shots, the majority of men's and women's shots have a time-to-net between 200 and 800 ms. The expected time to net was found to vary significantly by gender, shot type, and where in a rally it occurred. We found considerable between-player differences in average time-to-net of groundstrokes when serving or receiving, indicating the potential for time-to-net to capture differences in playing style. Time-to-net increased prediction accuracy of point outcomes by 8 percentage points. These findings show that time to net is a simple spatio-temporal statistic that has descriptive and predictive value for performance analysis in tennis.