Holger Trittenbach
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Holger Trittenbach has a total of 12 co-authors. Here are the top co-authors by number of shared publications:
- Klemens Böhm
Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany
7 shared publications - Adrian Englhardt
LinkedIn, Mountain View, CA, USA
2 shared publications - Jakob Bach
Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany
2 shared publications - Bernhard Sick
University of Kassel
1 shared publication - Daniel Kottke
University of Kassel
1 shared publication - Dominik Werle
Karlsruhe Institute of Technology (KIT)
1 shared publication
Latest Publications
- Open Access
- 17 August 2022
- Published by Springer Nature in SN Computer Science
- Vol. 3 (6) , 1-25
An Empirical Evaluation of Constrained Feature Selection
While feature selection helps to get smaller and more understandable prediction models, most existing feature-selection techniques do not consider domain knowledge. One way to use domain knowledge is via constraints on sets of selected features. However, the impact of constraints, e.g., on the predictive quality of selected features, is currently unclear. This article is an empirical study that evaluates the impact of propositional and arithmetic constraints on filter feature selection. First, we systematically generate constraints from various types, using datasets from different domains. As expected, constraints tend to decrease the predictive quality of feature sets, but this effect is non-linear. So we observe feature sets both adhering to constraints and with high predictive quality. Second, we study a concrete setting in materials science. This part of our study sheds light on how one can analyze scientific hypotheses with the help of constraints.
- 17 August 2022
- Published by Springer Nature in SN Computer Science
- Vol. 3 (6) , 1-25
Open AccessAn Empirical Evaluation of Constrained Feature Selection
While feature selection helps to get smaller and more understandable prediction models, most existing feature-selection techniques do not consider domain knowledge. One way to use domain knowledge is via constraints on sets of selected features. However, the impact of constraints, e.g., on the predictive quality of selected features, is currently unclear. This article is an empirical study that evaluates the impact of propositional and arithmetic constraints on filter feature selection. First, we systematically generate constraints from various types, using datasets from different domains. As expected, constraints tend to decrease the predictive quality of feature sets, but this effect is non-linear. So we observe feature sets both adhering to constraints and with high predictive quality. Second, we study a concrete setting in materials science. This part of our study sheds light on how one can analyze scientific hypotheses with the help of constraints.
- Open Access
- 7 April 2022
- Published by Springer Nature in Machine Learning
- Vol. 111 (4) , 1349-1375
Efficient SVDD sampling with approximation guarantees for the decision boundary
Support Vector Data Description (SVDD) is a popular one-class classifier for anomaly and novelty detection. But despite its effectiveness, SVDD does not scale well with data size. To avoid prohibitive training times, sampling methods select small subsets of the training data on which SVDD trains a decision boundary hopefully equivalent to the one obtained on the full data set. According to the literature, a good sample should therefore contain so-called boundary observations that SVDD would select as support vectors on the full data set. However, non-boundary observations also are essential to not fragment contiguous inlier regions and avoid poor classification accuracy. Other aspects, such as selecting a sufficiently representative sample, are important as well. But existing sampling methods largely overlook them, resulting in poor classification accuracy. In this article, we study how to select a sample considering these points. Our approach is to frame SVDD sampling as an optimization problem, where constraints guarantee that sampling indeed approximates the original decision boundary. We then propose RAPID, an efficient algorithm to solve this optimization problem. RAPID does not require any tuning of parameters, is easy to implement and scales well to large data sets. We evaluate our approach on real-world and synthetic data. Our evaluation is the most comprehensive one for SVDD sampling so far. Our results show that RAPID outperforms its competitors in classification accuracy, in sample size, and in runtime.
- 7 April 2022
- Published by Springer Nature in Machine Learning
- Vol. 111 (4) , 1349-1375
Open AccessEfficient SVDD sampling with approximation guarantees for the decision boundary
Support Vector Data Description (SVDD) is a popular one-class classifier for anomaly and novelty detection. But despite its effectiveness, SVDD does not scale well with data size. To avoid prohibitive training times, sampling methods select small subsets of the training data on which SVDD trains a decision boundary hopefully equivalent to the one obtained on the full data set. According to the literature, a good sample should therefore contain so-called boundary observations that SVDD would select as support vectors on the full data set. However, non-boundary observations also are essential to not fragment contiguous inlier regions and avoid poor classification accuracy. Other aspects, such as selecting a sufficiently representative sample, are important as well. But existing sampling methods largely overlook them, resulting in poor classification accuracy. In this article, we study how to select a sample considering these points. Our approach is to frame SVDD sampling as an optimization problem, where constraints guarantee that sampling indeed approximates the original decision boundary. We then propose RAPID, an efficient algorithm to solve this optimization problem. RAPID does not require any tuning of parameters, is easy to implement and scales well to large data sets. We evaluate our approach on real-world and synthetic data. Our evaluation is the most comprehensive one for SVDD sampling so far. Our results show that RAPID outperforms its competitors in classification accuracy, in sample size, and in runtime.
- Open Access
Active Learning of SVDD Hyperparameter Values
Support Vector Data Description is a popular method for outlier detection. However, its usefulness largely depends on selecting good hyperparameter values -- a difficult problem that has received significant attention in literature. Existing methods to estimate hyperparameter values are purely heuristic, and the conditions under which they work well are unclear. In this article, we propose LAMA (Local Active Min-Max Alignment), the first principled approach to estimate SVDD hyperparameter values by active learning. The core idea bases on kernel alignment, which we adapt to active learning with small sample sizes. In contrast to many existing approaches, LAMA provides estimates for both SVDD hyperparameters. These estimates are evidence-based, i.e., rely on actual class labels, and come with a quality score. This eliminates the need for manual validation, an issue with current heuristics. LAMA outperforms state-of-the-art competitors in extensive experiments on real-world data. In several cases, LAMA even yields results close to the empirical upper bound.
Open AccessActive Learning of SVDD Hyperparameter Values
Support Vector Data Description is a popular method for outlier detection. However, its usefulness largely depends on selecting good hyperparameter values -- a difficult problem that has received significant attention in literature. Existing methods to estimate hyperparameter values are purely heuristic, and the conditions under which they work well are unclear. In this article, we propose LAMA (Local Active Min-Max Alignment), the first principled approach to estimate SVDD hyperparameter values by active learning. The core idea bases on kernel alignment, which we adapt to active learning with small sample sizes. In contrast to many existing approaches, LAMA provides estimates for both SVDD hyperparameters. These estimates are evidence-based, i.e., rely on actual class labels, and come with a quality score. This eliminates the need for manual validation, an issue with current heuristics. LAMA outperforms state-of-the-art competitors in extensive experiments on real-world data. In several cases, LAMA even yields results close to the empirical upper bound.
- Open Access
An Overview and a Benchmark of Active Learning for Outlier Detection with One-Class Classifiers
Active learning methods increase classification quality by means of user feedback. An important subcategory is active learning for outlier detection with one-class classifiers. While various methods in this category exist, selecting one for a given application scenario is difficult. This is because existing methods rely on different assumptions, have different objectives, and often are tailored to a specific use case. All this calls for a comprehensive comparison, the topic of this article. This article starts with a categorization of the various methods. We then propose ways to evaluate active learning results. Next, we run extensive experiments to compare existing methods, for a broad variety of scenarios. Based on our results, we formulate guidelines on how to select active learning methods for outlier detection with one-class classifiers.
Open AccessAn Overview and a Benchmark of Active Learning for Outlier Detection with One-Class Classifiers
Active learning methods increase classification quality by means of user feedback. An important subcategory is active learning for outlier detection with one-class classifiers. While various methods in this category exist, selecting one for a given application scenario is difficult. This is because existing methods rely on different assumptions, have different objectives, and often are tailored to a specific use case. All this calls for a comprehensive comparison, the topic of this article. This article starts with a categorization of the various methods. We then propose ways to evaluate active learning results. Next, we run extensive experiments to compare existing methods, for a broad variety of scenarios. Based on our results, we formulate guidelines on how to select active learning methods for outlier detection with one-class classifiers.
- 1 April 2019
- Published by Walter de Gruyter GmbH in it - Information Technology
- Vol. 61 (2-3) , 111-123
Understanding the effects of temporal energy-data aggregation on clustering quality
Energy data often is available at high temporal resolution, which challenges the scalability of data-analysis methods. A common way to cope with this is to aggregate data to, say, 15-minute-interval summaries. But it often is not known how much information is lost with this, i. e., how good analysis results on aggregated data actually are. In this article, we study the effects of aggregating energy data on clustering. We propose an experimental design to compare a wide range of clustering methods found in literature. We then introduce different ways to compare clustering results obtained with different aggregation schemes. Our evaluation shows that aggregation affects the clustering quality significantly. Finally, we propose guidelines to select an aggregation scheme.
- 1 April 2019
- Published by Walter de Gruyter GmbH in it - Information Technology
- Vol. 61 (2-3) , 111-123
Understanding the effects of temporal energy-data aggregation on clustering quality
Energy data often is available at high temporal resolution, which challenges the scalability of data-analysis methods. A common way to cope with this is to aggregate data to, say, 15-minute-interval summaries. But it often is not known how much information is lost with this, i. e., how good analysis results on aggregated data actually are. In this article, we study the effects of aggregating energy data on clustering. We propose an experimental design to compare a wide range of clustering methods found in literature. We then introduce different ways to compare clustering results obtained with different aggregation schemes. Our evaluation shows that aggregation affects the clustering quality significantly. Finally, we propose guidelines to select an aggregation scheme.
- 14 June 2018
- Published by Springer Nature in International Journal of Data Science and Analytics
- Vol. 7 (2) , 87-101
Dimension-based subspace search for outlier detection
- 14 June 2018
- Published by Springer Nature in International Journal of Data Science and Analytics
- Vol. 7 (2) , 87-101
Dimension-based subspace search for outlier detection