Konstantin Kobs, M.Sc.

Chair of Data Science (Informatik X)
University of Würzburg
Am Hubland
97074 Würzburg
Germany

Email: kobs@informatik.uni-wuerzburg.de

Phone: (+49 931) 31 - 84707

Office: Room B100 (Computer Science Building M2)

Projects and Research Interests

I am working on Machine Learning and Neural Network models to process text, images, audio, and sensor data. The idea is to utilize additional information such as data attributes to learn representations that can be used in tasks, such as infromation retrieval or classification.

In the BigData@Geo project, we work with climate data, so time series and geospatial data. Our goal is to help the geography department to improve climate modeling using Machine Learning.

Education

2015–2017: M.Sc. Computer Science at the University of Hamburg
2012–2015: B.Sc. Computer Science at the University of Hamburg

Publications

2021[ to top ]

Kobs, K., Steininger, M., Dulny, A., and Hotho, A. (2021) “Do Different Deep Metric Learning Losses Lead to Similar Learned Features?”, Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 10644–10654.
- [ BibTeX ]
- [ Download ]
@article{noauthororeditor, author = {Kobs, Konstantin and Steininger, Michael and Dulny, Andrzej and Hotho, Andreas}, journal = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)}, keywords = {deepmetriclearning}, month = 10, pages = {10644-10654}, title = {Do Different Deep Metric Learning Losses Lead to Similar Learned Features?}, year = 2021 }
Pfister, J., Kobs, K., and Hotho, A. (2021) “Self-Supervised Multi-Task Pretraining Improves Image Aesthetic Assessment”, in Proceedings of the IEEE CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 816–825.
- [ BibTeX ]
- [ Download ]
@inproceedings{Pfister_2021_CVPR, author = {Pfister, Jan and Kobs, Konstantin and Hotho, Andreas}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops}, keywords = {myown}, month = {06}, pages = {816-825}, title = {Self-Supervised Multi-Task Pretraining Improves Image Aesthetic Assessment}, year = 2021 }
Steininger, M., Kobs, K., Davidson, P., Krause, A., and Hotho, A. (2021) “Density-based weighting for imbalanced regression”, Machine Learning, available: https://doi.org/10.1007/s10994-021-06023-5.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
- [ Download ]
In many real world settings,imbalanced data impedes model performance of learning algorithms, like neural networks, mostly for rare cases. This is especially problematic for tasks focusing on these rare occurrences. For example, when estimating precipitation, extreme rainfall events are scarce but important considering their potential consequences. While there are numerous well studied solutions for classification settings, most of them cannot be applied to regression easily. Of the few solutions for regression tasks, barely any have explored cost-sensitive learning which is known to have advantages compared to sampling-based methods in classification tasks. In this work, we propose a sample weighting approach for imbalanced regression datasets called DenseWeight and a cost-sensitive learning approach for neural network regression with imbalanced data called DenseLoss based on our weighting scheme. DenseWeight weights data points according to their target value rarities through kernel density estimation (KDE). DenseLoss adjusts each data point’s influence on the loss according to DenseWeight, giving rare data points more influence on modeltraining compared to common data points. We show on multiple differently distributed datasets that DenseLoss significantly improves model performance for rare data points through its density-based weighting scheme. Additionally, we compare DenseLoss to the state-of-the-art method SMOGN, finding that our method mostly yields better performance. Our approach provides more control over model training as it enables us to actively decide on the trade-off between focusing on common or rare cases through a single hyperparameter, allowing the training of better models for rare data points.

@article{steininger2021densitybased, abstract = {In many real world settings,imbalanced data impedes model performance of learning algorithms, like neural networks, mostly for rare cases. This is especially problematic for tasks focusing on these rare occurrences. For example, when estimating precipitation, extreme rainfall events are scarce but important considering their potential consequences. While there are numerous well studied solutions for classification settings, most of them cannot be applied to regression easily. Of the few solutions for regression tasks, barely any have explored cost-sensitive learning which is known to have advantages compared to sampling-based methods in classification tasks. In this work, we propose a sample weighting approach for imbalanced regression datasets called DenseWeight and a cost-sensitive learning approach for neural network regression with imbalanced data called DenseLoss based on our weighting scheme. DenseWeight weights data points according to their target value rarities through kernel density estimation (KDE). DenseLoss adjusts each data point’s influence on the loss according to DenseWeight, giving rare data points more influence on modeltraining compared to common data points. We show on multiple differently distributed datasets that DenseLoss significantly improves model performance for rare data points through its density-based weighting scheme. Additionally, we compare DenseLoss to the state-of-the-art method SMOGN, finding that our method mostly yields better performance. Our approach provides more control over model training as it enables us to actively decide on the trade-off between focusing on common or rare cases through a single hyperparameter, allowing the training of better models for rare data points.}, author = {Steininger, Michael and Kobs, Konstantin and Davidson, Padraig and Krause, Anna and Hotho, Andreas}, editor = {Appice, Annalisa and Escalera, Sergio and Gamez, Jose A. and Trautmann, Heike}, journal = {Machine Learning}, keywords = {myown}, title = {Density-based weighting for imbalanced regression}, year = 2021 }
Kobs, K., Schäfer, C., Steininger, M., Krause, A., Baumhauer, R., Paeth, H., and Hotho, A. (2021) “Semi-Supervised Learning for Grain Size Distribution Interpolation”, in Pattern Recognition. ICPR International Workshops and Challenges: Virtual Event, January 10--15, 2021, Proceedings, Part VI, Springer International Publishing, 34–44.
- [ BibTeX ]
- [ Download ]
@inproceedings{kobs2021semi, author = {Kobs, Konstantin and Schäfer, Christian and Steininger, Michael and Krause, Anna and Baumhauer, Roland and Paeth, Heiko and Hotho, Andreas}, booktitle = {Pattern Recognition. ICPR International Workshops and Challenges: Virtual Event, January 10–15, 2021, Proceedings, Part VI}, keywords = {myown}, organization = {Springer International Publishing}, pages = {34–44}, title = {Semi-Supervised Learning for Grain Size Distribution Interpolation}, year = 2021 }

2020[ to top ]

Kobs, K., Koopmann, T., Zehe, A., Fernes, D., Krop, P., and Hotho, A. (2020) “Where to Submit? Helping Researchers to Choose the Right Venue”, in Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: Findings, Online: Association for Computational Linguistics, 878–883, available: https://www.aclweb.org/anthology/2020.findings-emnlp.78.
- [ BibTeX ]
- [ URL ]
- [ Download ]
@inproceedings{kobs-etal-2020-submit, address = {Online}, author = {Kobs, Konstantin and Koopmann, Tobias and Zehe, Albin and Fernes, David and Krop, Philipp and Hotho, Andreas}, booktitle = {Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: Findings}, keywords = {myown}, month = 11, pages = {878–883}, publisher = {Association for Computational Linguistics}, title = {Where to Submit? Helping Researchers to Choose the Right Venue}, year = 2020 }
Davidson, P., Steininger, M., Lautenschlager, F., Kobs, K., Krause, A., and Hotho, A. (2020) “Anomaly Detection in Beehives using Deep Recurrent Autoencoders”, in Proceedings of the 9th International Conference on Sensor Networks (SENSORNETS 2020), SCITEPRESS – Science and Technology Publications, Lda., 142–149.
- [ BibTeX ]
@inproceedings{davidson2020anomaly, author = {Davidson, Padraig and Steininger, Michael and Lautenschlager, Florian and Kobs, Konstantin and Krause, Anna and Hotho, Andreas}, booktitle = {Proceedings of the 9th International Conference on Sensor Networks (SENSORNETS 2020)}, keywords = {beehive}, number = 9, pages = {142-149}, publisher = {SCITEPRESS – Science and Technology Publications, Lda.}, title = {Anomaly Detection in Beehives using Deep Recurrent Autoencoders}, year = 2020 }
Kobs, K., Potthast, M., Wiegmann, M., Zehe, A., Stein, B., and Hotho, A. (2020) “Towards Predicting the Subscription Status of Twitch.tv Users”, Proceedings of ECML-PKDD 2020 ChAT Discovery Challenge on Chat Analytics for Twitch, available: http://ceur-ws.org/Vol-2661/paper1.pdf.
- [ BibTeX ]
- [ URL ]
- [ Download ]
@article{kobstowards, author = {Kobs, Konstantin and Potthast, Martin and Wiegmann, Matti and Zehe, Albin and Stein, Benno and Hotho, Andreas}, journal = {Proceedings of ECML-PKDD 2020 ChAT Discovery Challenge on Chat Analytics for Twitch}, keywords = {myown}, title = {Towards Predicting the Subscription Status of Twitch.tv Users}, year = 2020 }
Lautenschlager, F., Becker, M., Kobs, K., Steininger, M., Davidson, P., Krause, A., and Hotho, A. (2020) “OpenLUR: Off-the-shelf air pollution modeling with open features and machine learning”, Atmospheric Environment, 233, 117535, available: https://doi.org/https://doi.org/10.1016/j.atmosenv.2020.117535.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
To assess the exposure of citizens to pollutants like NOx or particulate matter in urban areas, land use regression (LUR) models are a well established method. LUR models leverage information about environmental and anthropogenic factors such as cars, heating, or industry to predict air pollution in areas where no measurements have been made. However, existing approaches are often not globally applicable and require tedious hyper-parameter tuning to enable high quality predictions. In this work, we tackle these issues by introducing OpenLUR, an off-the-shelf approach for modeling air pollution that (i) works on a set of novel features solely extracted from the globally and openly available data source OpenStreetMap and (ii) is based on state-of-the-art machine learning featuring automated hyper-parameter tuning in order to minimize manual effort. We show that our proposed features are able to outperform their counterparts from local and closed sources, and illustrate how automated hyper parameter tuning can yield competitve results while alleviating the need for expert knowledge in machine learning and manual effort. Importantly, we further demonstrate the potential of the global availability of our features by applying cross-learning across different cities in order to reduce the need for a large amount of training samples. Overall, OpenLUR represents an off-the-shelf approach that facilitates easily reproducible experiments and the development of globally applicable models.

@article{lautenschlager2020openlur, abstract = {To assess the exposure of citizens to pollutants like NOx or particulate matter in urban areas, land use regression (LUR) models are a well established method. LUR models leverage information about environmental and anthropogenic factors such as cars, heating, or industry to predict air pollution in areas where no measurements have been made. However, existing approaches are often not globally applicable and require tedious hyper-parameter tuning to enable high quality predictions. In this work, we tackle these issues by introducing OpenLUR, an off-the-shelf approach for modeling air pollution that (i) works on a set of novel features solely extracted from the globally and openly available data source OpenStreetMap and (ii) is based on state-of-the-art machine learning featuring automated hyper-parameter tuning in order to minimize manual effort. We show that our proposed features are able to outperform their counterparts from local and closed sources, and illustrate how automated hyper parameter tuning can yield competitve results while alleviating the need for expert knowledge in machine learning and manual effort. Importantly, we further demonstrate the potential of the global availability of our features by applying cross-learning across different cities in order to reduce the need for a large amount of training samples. Overall, OpenLUR represents an off-the-shelf approach that facilitates easily reproducible experiments and the development of globally applicable models.}, author = {Lautenschlager, Florian and Becker, Martin and Kobs, Konstantin and Steininger, Michael and Davidson, Padraig and Krause, Anna and Hotho, Andreas}, journal = {Atmospheric Environment}, keywords = {myown}, pages = 117535, title = {OpenLUR: Off-the-shelf air pollution modeling with open features and machine learning}, volume = 233, year = 2020 }
Kobs, K., Steininger, M., Zehe, A., Lautenschlager, F., and Hotho, A. (2020) “SimLoss: Class Similarities in Cross Entropy”, available: http://arxiv.org/abs/2003.03182.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
- [ Download ]
One common loss function in neural network classification tasks is Categorical Cross Entropy (CCE), which punishes all misclassifications equally. However, classes often have an inherent structure. For instance, classifying an image of a rose as "violet" is better than as "truck". We introduce SimLoss, a drop-in replacement for CCE that incorporates class similarities along with two techniques to construct such matrices from task-specific knowledge. We test SimLoss on Age Estimation and Image Classification and find that it brings significant improvements over CCE on several metrics. SimLoss therefore allows for explicit modeling of background knowledge by simply exchanging the loss function, while keeping the neural network architecture the same. Code and additional resources can be found at https://github.com/konstantinkobs/SimLoss.

@misc{kobs2020simloss, abstract = {One common loss function in neural network classification tasks is Categorical Cross Entropy (CCE), which punishes all misclassifications equally. However, classes often have an inherent structure. For instance, classifying an image of a rose as "violet" is better than as "truck". We introduce SimLoss, a drop-in replacement for CCE that incorporates class similarities along with two techniques to construct such matrices from task-specific knowledge. We test SimLoss on Age Estimation and Image Classification and find that it brings significant improvements over CCE on several metrics. SimLoss therefore allows for explicit modeling of background knowledge by simply exchanging the loss function, while keeping the neural network architecture the same. Code and additional resources can be found at https://github.com/konstantinkobs/SimLoss.}, author = {Kobs, Konstantin and Steininger, Michael and Zehe, Albin and Lautenschlager, Florian and Hotho, Andreas}, keywords = {lossfunction}, note = {cite arxiv:2003.03182Comment: This paper is going to be published in the proceedings of the 25th International Symposium on Methodologies for Intelligent Systems (ISMIS)}, title = {SimLoss: Class Similarities in Cross Entropy}, year = 2020 }
Schlör, D., Zehe, A., Kobs, K., Veseli, B., Westermeier, F., Brübach, L., Roth, D., Latoschik, M.E., and Hotho, A. (2020) “Improving Sentiment Analysis with Biofeedback Data”, in Proceedings of the Workshop on PeOple in LaNguage, VIsiOn and the MiNd (ONION), available: https://downloads.hci.informatik.uni-wuerzburg.de/2018-ieeevr-lugrin-vr-teacher-training/2020-onion-sentiment-eeg-preprint.pdf.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
- [ Download ]
Humans frequently are able to read and interpret emotions of others by directly taking verbal and non-verbal signals in human-to-human communication into account or to infer or even experience emotions from mediated stories. For computers, however, emotion recognition is a complex problem: Thoughts and feelings are the roots of many behavioural responses and they are deeply entangled with neurophysiological changes within humans. As such, emotions are very subjective, often are expressed in a subtle manner, and are highly depending on context. For example, machine learning approaches for text-based sentiment analysis often rely on incorporating sentiment lexicons or language models to capture the contextual meaning. This paper explores if and how we further can enhance sentiment analysis using biofeedback of humans which are experiencing emotions while reading texts. Specifically, we record the heart rate and brain waves of readers that are presented with short texts which have been annotated with the emotions they induce. We use these physiological signals to improve the performance of a lexicon-based sentiment classifier. We find that the combination of several biosignals can improve the ability of a text-based classifier to detect the presence of a sentiment in a text on a per-sentence level.

@inproceedings{schlor2020improving, abstract = {Humans frequently are able to read and interpret emotions of others by directly taking verbal and non-verbal signals in human-to-human communication into account or to infer or even experience emotions from mediated stories. For computers, however, emotion recognition is a complex problem: Thoughts and feelings are the roots of many behavioural responses and they are deeply entangled with neurophysiological changes within humans. As such, emotions are very subjective, often are expressed in a subtle manner, and are highly depending on context. For example, machine learning approaches for text-based sentiment analysis often rely on incorporating sentiment lexicons or language models to capture the contextual meaning. This paper explores if and how we further can enhance sentiment analysis using biofeedback of humans which are experiencing emotions while reading texts. Specifically, we record the heart rate and brain waves of readers that are presented with short texts which have been annotated with the emotions they induce. We use these physiological signals to improve the performance of a lexicon-based sentiment classifier. We find that the combination of several biosignals can improve the ability of a text-based classifier to detect the presence of a sentiment in a text on a per-sentence level.}, author = {Schlör, Daniel and Zehe, Albin and Kobs, Konstantin and Veseli, Blerta and Westermeier, Franziska and Brübach, Larissa and Roth, Daniel and Latoschik, Marc Erich and Hotho, Andreas}, booktitle = {Proceedings of the Workshop on peOple in laNguage, vIsiOn and the miNd (ONION)}, keywords = {myown}, title = {Improving Sentiment Analysis with Biofeedback Data}, year = 2020 }
Steininger, M., Kobs, K., Zehe, A., Lautenschlager, F., Becker, M., and Hotho, A. (2020) “MapLUR: Exploring a new Paradigm for Estimating Air Pollution using Deep Learning on Map Images”, available: https://arxiv.org/abs/2002.07493.
- [ Abstract ]
- [ BibTeX ]
- [ URL ]
Land-use regression (LUR) models are important for the assessment of air pollution concentrations in areas without measurement stations. While many such models exist, they often use manually constructed features based on restricted, locally available data. Thus, they are typically hard to reproduce and challenging to adapt to areas beyond those they have been developed for. In this paper, we advocate a paradigm shift for LUR models: We propose the Data-driven, Open, Global (DOG) paradigm that entails models based on purely data-driven approaches using only openly and globally available data. Progress within this paradigm will alleviate the need for experts to adapt models to the local characteristics of the available data sources and thus facilitate the generalizability of air pollution models to new areas on a global scale. In order to illustrate the feasibility of the DOG paradigm for LUR, we introduce a deep learning model called MapLUR. It is based on a convolutional neural network architecture and is trained exclusively on globally and openly available map data without requiring manual feature engineering. We compare our model to state-of-the-art baselines like linear regression, random forests and multi-layer perceptrons using a large data set of modeled NO2 concentrations in Central London. Our results show that MapLUR significantly outperforms these approaches even though they are provided with manually tailored features. Furthermore, we illustrate that the automatic feature extraction inherent to models based on the DOG paradigm can learn features that are readily interpretable and closely resemble those commonly used in traditional LUR approaches.

@misc{steininger2020maplur, abstract = {Land-use regression (LUR) models are important for the assessment of air pollution concentrations in areas without measurement stations. While many such models exist, they often use manually constructed features based on restricted, locally available data. Thus, they are typically hard to reproduce and challenging to adapt to areas beyond those they have been developed for. In this paper, we advocate a paradigm shift for LUR models: We propose the Data-driven, Open, Global (DOG) paradigm that entails models based on purely data-driven approaches using only openly and globally available data. Progress within this paradigm will alleviate the need for experts to adapt models to the local characteristics of the available data sources and thus facilitate the generalizability of air pollution models to new areas on a global scale. In order to illustrate the feasibility of the DOG paradigm for LUR, we introduce a deep learning model called MapLUR. It is based on a convolutional neural network architecture and is trained exclusively on globally and openly available map data without requiring manual feature engineering. We compare our model to state-of-the-art baselines like linear regression, random forests and multi-layer perceptrons using a large data set of modeled NO2 concentrations in Central London. Our results show that MapLUR significantly outperforms these approaches even though they are provided with manually tailored features. Furthermore, we illustrate that the automatic feature extraction inherent to models based on the DOG paradigm can learn features that are readily interpretable and closely resemble those commonly used in traditional LUR approaches.}, author = {Steininger, Michael and Kobs, Konstantin and Zehe, Albin and Lautenschlager, Florian and Becker, Martin and Hotho, Andreas}, keywords = {myown}, note = {cite arxiv:2002.07493}, title = {MapLUR: Exploring a new Paradigm for Estimating Air Pollution using Deep Learning on Map Images}, year = 2020 }
Fischer, M., Kobs, K., and Hotho, A. (2020) “NICER — Aesthetic Image Enhancement with Humans in the Loop”, in ACHI 2020: The Thirteenth International Conference on Advances in Computer-Human Interactions, 357–362.
- [ Abstract ]
- [ BibTeX ]
- [ Download ]
Fully- or semi-automatic image enhancement software helps users to increase the visual appeal of photos and does not require in-depth knowledge of manual image editing. However, fully-automatic approaches usually enhance the image in a black-box manner that does not give the user any control over the optimization process, possibly leading to edited images that do not subjectively appeal to the user. Semi-automatic methods mostly allow for controlling which pre-defined editing step is taken, which restricts the users in their creativity and ability to make detailed adjustments, such as brightness or contrast. We argue that incorporating user preferences by guiding an automated enhancement method simplifies image editing and increases the enhancement’s focus on the user. This work thus proposes the Neural Image Correction & Enhancement Routine (NICER), a neural network based approach to no-reference image enhancement in a fully-, semi-automatic or fully manual process that is interactive and user-centered. NICER iteratively adjusts image editing parameters in order to maximize an aesthetic score based on image style and content. Users can modify these parameters at any time and guide the optimization process towards a desired direction. This interactive workflow is a novelty in the field of human-computer interaction for image enhancement tasks. In a user study, we show that NICER can improve image aesthetics without user interaction and that allowing user interaction leads to diverse enhancement outcomes that are strongly preferred over the unedited image. We make our code publicly available to facilitate further research in this direction.

@inproceedings{fischer2020nicer, abstract = {Fully- or semi-automatic image enhancement software helps users to increase the visual appeal of photos and does not require in-depth knowledge of manual image editing. However, fully-automatic approaches usually enhance the image in a black-box manner that does not give the user any control over the optimization process, possibly leading to edited images that do not subjectively appeal to the user. Semi-automatic methods mostly allow for controlling which pre-defined editing step is taken, which restricts the users in their creativity and ability to make detailed adjustments, such as brightness or contrast. We argue that incorporating user preferences by guiding an automated enhancement method simplifies image editing and increases the enhancement’s focus on the user. This work thus proposes the Neural Image Correction & Enhancement Routine (NICER), a neural network based approach to no-reference image enhancement in a fully-, semi-automatic or fully manual process that is interactive and user-centered. NICER iteratively adjusts image editing parameters in order to maximize an aesthetic score based on image style and content. Users can modify these parameters at any time and guide the optimization process towards a desired direction. This interactive workflow is a novelty in the field of human-computer interaction for image enhancement tasks. In a user study, we show that NICER can improve image aesthetics without user interaction and that allowing user interaction leads to diverse enhancement outcomes that are strongly preferred over the unedited image. We make our code publicly available to facilitate further research in this direction.}, author = {Fischer, Michael and Kobs, Konstantin and Hotho, Andreas}, journal = {ACHI 2020: The Thirteenth International Conference on Advances in Computer-Human Interactions}, keywords = {image-assessment}, pages = {357-362}, title = {NICER — Aesthetic Image Enhancement with Humans in the Loop}, year = 2020 }
Kobs, K., Zehe, A., Bernstetter, A., Chibane, J., Pfister, J., Tritscher, J., and Hotho, A. (2020) “Emote-Controlled: Obtaining Implicit Viewer Feedback through Emote based Sentiment Analysis on Comments of Popular Twitch.tv Channels”, ACM Transactions on Social Computing, available: https://doi.org/10.1145/3365523.
- [ BibTeX ]
- [ Download ]
@article{noauthororeditor, author = {Kobs, Konstantin and Zehe, Albin and Bernstetter, Armin and Chibane, Julian and Pfister, Jan and Tritscher, Julian and Hotho, Andreas}, journal = {ACM Transactions on Social Computing}, keywords = {sentiment-analysis}, note = {accepted but not published}, title = {Emote-Controlled: Obtaining Implicit Viewer Feedback through Emote based Sentiment Analysis on Comments of Popular Twitch.tv Channels}, year = 2020 }

Konstantin Kobs, M.Sc.

Projects and Research Interests

Education

Publications

Hinweis zum Datenschutz

Hinweis zum Datenschutz

Bildnachweise