Department of Computer Science

Prof. Dr. Gjergji Kasneci

After my PhD in Graph-based Mining and Retrieval at the Max Planck Institute for Computer Science in Saarbrücken, Germany (2009), I joined Microsoft Research Cambridge, UK, as a postdoctoral researcher working on probabilistic inference methods in knowledge bases. In 2011, I joined the Hasso Plattner Institute in Potsdam, Germany, where I led the research group "Web Mining and Analytics". In mid-2014, I joined SCHUFA Holding AG, where I held the CTO position from April 2017 to April 2022 and headed the "Innovation and Strategic Analysis" division. Since 2018, I have been heading the research group "Data Science and Analytics" at the University of Tübingen.

Research Interests

  • Data engineering and analytics
  • Explainable analytics on very large tabular data sets
  • Uncertainty quantification and drift detection in online learning and streaming applications
  • Algorithmic recourse / counterfactual explanations
  • Analysis of fairness and bias in Machine Learning applications
  • Quality assurance of Machine Learning applications

Publications

2023

ChatGPT for Good? On Opportunities and Challenges of Large Language Models for Education by Enkelejda Kasneci et al. Preprint.

Language Models are Realistic Tabular Data Generators by Vadim Borisov, Kathrin Sessler, Tobias Leemann, Martin Pawelczyk, and Gjergji Kasneci. International Conference on Learning Representations (ICLR). To appear.

On the Trade-Off between Actionable Explanations and the Right to be Forgotten by Martin Pawelczyk, Tobias Leemann, Asia Biega, and Gjergji KasneciInternational Conference on Learning Representations (ICLR). To appear.

Probabilistically Robust Recourse: Navigating the Trade-offs between Costs and Robustness in Algorithmic Recourse by Martin Pawelczyk, Teresa Datta, Johannes van-den-Heuvel, Gjergji Kasneci, and Himabindu Lakkaraju. International Conference on Learning Representations (ICLR). To appear.

2022

Deep Neural Networks and Tabular Data: A Survey by Vadim Borisov, Tobias Leemann, Kathrin Sessler, Johannes Haug, Martin Pawelczyk and Gjergji Kasneci accepted to IEEE Transactions on Neural Networks and Learning Systems (TNNLS).

Interventional SHAP Values and Interaction Values for Piecewise Linear Regression Trees by Artjom Zern, Klaus Broelemann, Gjergji Kasneci accepted at AAAI 2023

I Prefer not to Say: Operationalizing Fair and User-guided Data Minimization by Tobias Leemann, Martin Pawelczyk, Christian Thomas Eberle and Gjergji Kasneci was accepted at the NeuIPS22 Workshop on Algorithmic Fairness through the Lens of Causality and Privacy.

On the Trade-Off between Actionable Explanations and the Right to be Forgotten by Martin Pawelczyk, Tobias Leemann, Asia Biega und Gjergji Kasneci was accepted at the NeuIPS22 Workshop on Trustworthy and Socially Responsible Machine Learning.

Change Detection for Local Explainability in Evolving Data Streams by Johannes Haug, Alexander Braun, Stefan Zürn, and Gjergji Kasneci was accepted at CIKM 2022 (github).

Disentangling Embedding Spaces with Minimal Distributional Assumptions by Tobias Leemann, Michael KirchhofYao RongEnkelejda Kasneci and Gjergji Kasneci was published as a preprint on arXiv.

Regressive Saccadic Eye Movements on Fake News von Efe Bozkir, Gjergji Kasneci, Sonja Utz und Enkelejda Kasneci was accepted at ETRA 2022.

A Constistent and Efficient Evaluation Strategy for Attribution Methods by Yao Rong*, Tobias Leemann*, Vadim Borisov, Gjergji Kasneci and Enkelejda Kasneci was accepted at the International Conference on Machine Learning (ICML) 2022.

Fairness in Agreement with European Values: An Interdisciplinary Perspective on AI Regulation by Alejandra Bringas Colmenarejo, Luca Nannini, Alisa Rieger, Kristen Marie Scott, Xuan Zhao, Gourab Patro, Gjergji Kasneci, and Katharina Kinder-Kurlanda was accepted at AIES 2022.

Standardized Evaluation of Machine Learning Methods for Evolving Data Streams by Johannes Haug, Effi Tramountani and Gjergji Kasneci was published on arXiv. The corresponding float-evaluation framework (Python) is available at Github or Pypi.

Coherence Evaluation of Visual Concepts with Objects and Language by Tobias Leemann, Yao Rong, Stefan Kraft, Enkelejda Kasneci and Gjergji Kasneci was presented at the ICLR2022 Workshop on Objects, Structure and Causality.

Algorithmic recourse in the face of noisy human responses by Martin Pawelczyk, Teresa Datta, Johannes van-den-Heuvel, Gjergji Kasneci and Himabindu Lakkaraju was presented at the ICLR2022 on Socially Responsible Machine Learning.

Disentangling Algorithmic Recourse by Martin Pawelczyk, Lea Tiyavorabun and Gjergji Kasneci was presented at the ICLR2022 on Socially Responsible Machine Learning.

Dynamic Model Tree for Interpretable Data Stream Learning by Johannes Haug, Klaus Broelemann and Gjergji Kasneci has been accepted at ICDE 2022 (github).

Do your eye movements reveal your performance on an IQ test? A study linking eye movements and socio-demographic information to fluid intelligence by Enkelejda KasneciGjergji Kasneci, Ulrich Trautwein, Tobias Appel, Maike Tibus, Susanne M. Jaeggi & Peter Gerjets was accepted at the PLOS ONE journal.

2021

Aggregating the Gaussian Experts' Predictions via Undirected Graphical Models by Hamed Jalali and Gjergji Kasneci was accepted at IEEE BigComp 2022.

Model Selection in Local Approximation Gaussian Processes: A Markov Random Fields Approach by Hamed Jalali, Martin Pawelczyk, and Gjergji Kasneci was accepted at IEEE International Conference on Big Data 2021.

Gaussian Graphical Models as an Ensemble Method for Distributed Gaussian Processes by Hamed Jalali and Gjergji Kasneci was accepted at OPT2021: 13th Annual Workshop on Optimization for Machine Learning @ NeurIPS 2021.

SPARROW: Semantically Coherent Prototypes for Image Classification by Stefan Kraft, Klaus Broelemann, Andreas Theissler, and Gjergji Kasneci was accepted at BMVC 2021.

A Robust Unsupervised Ensemble of Feature-Based Explanations using Restricted Boltzmann Machines by Vadim Borisov, Johannes Meier, Johan Van den Heuvel, Hamed Jalali, and Gjergji Kasneci was accepted at the workshop on eXplainable AI approaches for debugging and diagnosis @ NeurIPS 2021.

Deep Neural Networks and Tabular Data: A Survey by Vadim Borisov, Tobias Leemann, Kathrin Seßler, Johannes Haug, Martin Pawelczyk, and Gjergji Kasneci was published on arXiv.

CARLA: A Python Library to Benchmark Algorithmic Recourse and Counterfactual Explanation Algorithms by Martin Pawelczyk, Sascha Bielawski, Johan Van den Heuvel, Tobias Richter and Gjergji Kasneci was accepted at NeurIPS 2021 (Datasets and Benchmarks Track) (github).

TEyeD: Over 20 million real-world eye images with Pupil, Eyelid, and Iris 2D and 3D Segmentations, 2D and 3D Landmarks, 3D Eyeball, Gaze Vector, and Eye Movement Types by Wolfgang Fuhl, Gjergji Kasneci and Enkelejda Kasneci was accepted at ISMAR 2021 (dataset).

"TüEyeQ, a rich IQ test performance data set with eye movement, educational and socio-demographic information" by Enkelejda Kasneci, Gjergji Kasneci, Tobias Appel, Johannes Haug, Franz Wortha, Maike Tibus, Ulrich Trautwein and Peter Gerjets was accepted at the Nature Scientific Data Journal (8, 2021) (dataset, github)

On Baselines for Local Feature Attributions by Johannes Haug, Stefan Zürn, Peter El-Jiz and Gjergji Kasneci was accepted at the workshop on Explainably Agency in AI @ AAAI 2021 (Abstract, .bib-file, github)

2020

Aggregating Dependent Gaussian Experts in Local Approximation by Hamed Jalali and Gjergji Kasneci was accepted at ICPR 2020 (a link to the paper will follow soon).

Learning Parameter Distributions to Detect Concept Drift in Data Streams by Johannes Haug and Gjergji Kasneci was accepted at ICPR 2020 (Abstract, .bib-file, github).

Leveraging Model Inherent Variable Importance for Stable Online Feature Selection by Johannes Haug, Martin Pawelczyk, Klaus Broelemann and Gjergji Kasneci was accepted at KDD 2020 (Abstract, .bib-file, github).

On Learning Invariant Counterfactual Explanations under Predictive Multiplicity by Martin Pawelczyk, Klaus Broelemann and Gjergji Kasneci was accepted at UAI 2020 (Abstract, .bib-file). 

A MinHash Approach for Fast Scanpath Classification by David Geisler, Nora Castner, Gjergji Kasneci and Enkelejda Kasneci was accepted at ETRA 2020 and won the Best-Paper Award (Abstract.bib-file).

PLAY: A Profiled Linear Weighting Scheme for Understanding the Influence of Input Variables on the Output of a Deep Artificial Neural Network by Torsten Dietl, Gjergji Kasneci, Johannes Fürnkranz and Eneldo Loza Mencía was accepted at the Archives of Data Science, Series A (Abstract, .bib-file).

Learning Model Agnostic Counterfactual Explanations for Tabular Data by Martin Pawelczyk, Klaus Broelemann and Gjergji Kasneci was accepted at WWW 2020 (Abstract.bib-file).

Bias in Data-driven AI Systems - An Introductory Survey by Ntoutsi et al. was accepted at WIRE's Data Mining and Knowledge Discovery Journal. This is the first multi-disciplinary publication from our EU "NoBias" Project (Abstract, .bib-file​​​​​​​).

Training Decision Trees as Replacement for Convolution Layers by Wolfgang Fuhl, Gjergji Kasneci, Wolfgang Rosenstiel and Enkelejda Kasneci was accepted at AAAI 2020 and is currently published on arXiv (Abstract, .bib-file).

2019

Towards User Empowerment (2019) by Martin Pawelczyk, Johannes Haug, Klaus Broelemann and Gjergji Kasneci was accepted at the NeurIPS 2019 workshop on Human-Centric Machine Learning and was published at arXiv (Abstract, .bib-file).

CancelOut: A Layer for Feature Selection in Deep Neural Networks (2019) by Vadim Borisov, Johannes Haug and Gjergji Kasneci was published at ICANN 2019 (Abstract, .bib-file).

A Gradient-Based Split Criterion for Highly Accurate and Transparent Model Trees (2019) by Klaus Broelemann and Gjergji Kasneci was published at IJCAI 2019 (Abstract, .bib-file).

Validation Loss for Landmark Detection (2019) by Wolfgang FuhlThomas Kübler, Rene Alexander Lotz, Gjergji Kasneci, Wolfgang Rosenstiel and Enkelejda Kasneci was published as a preprint on arXiv (Abstract, .bib-file).