Research Group Gjergji Kasneci
Gjergji Kasneci
Prof. Dr.
Core PI
Gjergji Kasneci
is Professor of Responsible Data Science at TU Munich.
His research focuses on transparency, robustness, bias, and fairness in machine learning algorithms and involves ethical, legal, and societal considerations with the goal of using artificial intelligence responsibly for the benefit of individuals and society.
Team members @MCML
PostDocs
Bardh Prenkaj
Dr.
→ Group Gjergji Kasneci
Responsible Data Science
PhD Students
Jan Batzner
→ Group Gjergji Kasneci
Responsible Data Science
Mahdi Dhaini
→ Group Gjergji Kasneci
Responsible Data Science
Alina Fastowski
→ Group Gjergji Kasneci
Responsible Data Science
Sven Jacob
→ Group Gjergji Kasneci
Responsible Data Science
Rohit Jagtani
→ Group Gjergji Kasneci
Responsible Data Science
Yuxiao Li
→ Group Gjergji Kasneci
Responsible Data Science
Felix Steinbauer
→ Group Gjergji Kasneci
Responsible Data Science
Shuo Yang
→ Group Gjergji Kasneci
Responsible Data Science
Chenchen Yuan
→ Group Gjergji Kasneci
Responsible Data Science
Efstratios Zaradoukas
→ Group Gjergji Kasneci
Responsible Data Science
Zheyu Zhang
→ Group Gjergji Kasneci
Responsible Data Science
Recent News @MCML
Publications @MCML
2026
[20]
Y. Li • G. Kasneci
Safety Cost of Steering Vectors Is Separable and Reducible.
AI4GOOD @ICML 2026 - Workshop on Trustworthy AI for Good at the 43rd International Conference on Machine Learning. Seoul, South Korea, Jul 06-11, 2026. To be published. Preprint available. URL
Safety Cost of Steering Vectors Is Separable and Reducible.
AI4GOOD @ICML 2026 - Workshop on Trustworthy AI for Good at the 43rd International Conference on Machine Learning. Seoul, South Korea, Jul 06-11, 2026. To be published. Preprint available. URL
[19]
Y. Li • G. Kasneci
Safety Cost of Steering Vectors Is Separable and Reducible.
CompLearn @ICML 2026 - 2nd Workshop on Compositional Learning: Safety, Interpretability, and Agents at the 43rd International Conference on Machine Learning. Seoul, South Korea, Jul 06-11, 2026. To be published. Preprint available. URL
Safety Cost of Steering Vectors Is Separable and Reducible.
CompLearn @ICML 2026 - 2nd Workshop on Compositional Learning: Safety, Interpretability, and Agents at the 43rd International Conference on Machine Learning. Seoul, South Korea, Jul 06-11, 2026. To be published. Preprint available. URL
[18]
M. Akhtar • A. Reuel • P. Soni • S. Ahuja • P. S. Ammanamanchi • R. Rawal • V. Zouhar • S. Yadav • C. Whitehouse • D. Ki • J. Mickel • L. Choshen • M. Šuppa • J. Batzner • J. Chim • J. Sania • Y. Long • H. A. Rahmani • C. Knight • Y. Nan • J. Raj • Y. Fan • S. Singh • S. Sahoo • E. Habba • U. Gohar • S. Pawar • R. Scholz • A. Subramonian • J. Ni • M. Kochenderfer • S. Koyejo • M. Sachan • S. Biderman • Z. Talat • A. Ghosh • I. Solaiman
When AI Benchmarks Plateau: A Systematic Study of Benchmark Saturation.
ICML 2026 - 43rd International Conference on Machine Learning. Seoul, South Korea, Jul 06-11, 2026. To be published. Preprint available. URL
When AI Benchmarks Plateau: A Systematic Study of Benchmark Saturation.
ICML 2026 - 43rd International Conference on Machine Learning. Seoul, South Korea, Jul 06-11, 2026. To be published. Preprint available. URL
[17]
E. Kasneci • G. Kasneci
Sycophancy is an Educational Safety Risk: Why LLM Tutors Need Sycophancy Benchmarks.
ICML 2026 - 43rd International Conference on Machine Learning. Seoul, South Korea, Jul 06-11, 2026. To be published. Preprint available. URL GitHub
Sycophancy is an Educational Safety Risk: Why LLM Tutors Need Sycophancy Benchmarks.
ICML 2026 - 43rd International Conference on Machine Learning. Seoul, South Korea, Jul 06-11, 2026. To be published. Preprint available. URL GitHub
[16]
A. Reuel • A. Ghosh • J. Chim • A. Tran • Y. Long • J. Mickel • U. Gohar • S. Yadav • P. Ammanamanchi • M. Allaham • H. A. Rahmani • M. Akhtar • F. Friedrich • R. Scholz • M. A. Riegler • J. Batzner • E. Habba • A. Saxena • A. Kornilova • K. Wei • P. Soni • Y. Mathew • K. Klyman • J. Sania • S. Sahoo • O. Bruvik • P. Sadeghi • S. Goswami • A. Wang • Y. Jernite • Z. Talat • S. Biderman • M. Kochenderfer • S. Koyejo • I. Solaiman
Who Evaluates AI's Social Impacts? Mapping Coverage and Gaps in First and Third Party Evaluations.
ICML 2026 - 43rd International Conference on Machine Learning. Seoul, South Korea, Jul 06-11, 2026. To be published. Preprint available. URL
Who Evaluates AI's Social Impacts? Mapping Coverage and Gaps in First and Third Party Evaluations.
ICML 2026 - 43rd International Conference on Machine Learning. Seoul, South Korea, Jul 06-11, 2026. To be published. Preprint available. URL
[15]
Z. Zhang • S. Yang • B. Prenkaj • G. Kasneci
Active Tabular Augmentation via Policy-Guided Diffusion Inpainting.
ICML 2026 - 43rd International Conference on Machine Learning. Seoul, South Korea, Jul 06-11, 2026. To be published. Preprint available. URL GitHub
Active Tabular Augmentation via Policy-Guided Diffusion Inpainting.
ICML 2026 - 43rd International Conference on Machine Learning. Seoul, South Korea, Jul 06-11, 2026. To be published. Preprint available. URL GitHub
[14]
F. Weeber • V. Neplenbroek • J. Batzner • S. Padó
One Persona, Many Cues, Different Results: How Sociodemographic Cues Impact LLM Personalization.
ACL 2026 - 64th Annual Meeting of the Association for Computational Linguistics. San Diego, CA, USA, Jul 02-07, 2026. DOI
One Persona, Many Cues, Different Results: How Sociodemographic Cues Impact LLM Personalization.
ACL 2026 - 64th Annual Meeting of the Association for Computational Linguistics. San Diego, CA, USA, Jul 02-07, 2026. DOI
[13]
S. Yang • Z. Zhang • B. Prenkaj • G. Kasneci
SAGE: Sparse Adaptive Guidance for Dependency-Aware Tabular Data Generation.
ACL 2026 - 64th Annual Meeting of the Association for Computational Linguistics. San Diego, CA, USA, Jul 02-07, 2026. DOI GitHub
SAGE: Sparse Adaptive Guidance for Dependency-Aware Tabular Data Generation.
ACL 2026 - 64th Annual Meeting of the Association for Computational Linguistics. San Diego, CA, USA, Jul 02-07, 2026. DOI GitHub
[12]
C. Yuan • Z. Zhang • G. Kasneci
Where Paths Split: Localized, Calibrated Control of Moral Reasoning in Large Language Models.
ACL 2026 - 64th Annual Meeting of the Association for Computational Linguistics. San Diego, CA, USA, Jul 02-07, 2026. DOI GitHub
Where Paths Split: Localized, Calibrated Control of Moral Reasoning in Large Language Models.
ACL 2026 - 64th Annual Meeting of the Association for Computational Linguistics. San Diego, CA, USA, Jul 02-07, 2026. DOI GitHub
[11]
Y. Li • A. Fastowski • E. Zaradoukas • B. Prenkaj • G. Kasneci
Analysing the Safety Pitfalls of Steering Vectors.
Findings @ACL 2026 - Findings at the 64th Annual Meeting of the Association for Computational Linguistics. San Diego, CA, USA, Jul 02-07, 2026. DOI GitHub
Analysing the Safety Pitfalls of Steering Vectors.
Findings @ACL 2026 - Findings at the 64th Annual Meeting of the Association for Computational Linguistics. San Diego, CA, USA, Jul 02-07, 2026. DOI GitHub
[10]
J. Batzner • S. H. Nelaturu • D. Stachura • A. Kornilova • J. Crall • T. Cerruti • Y. Long • Y. Mai • S. Ahuja • A. Yehudai • M. Šuppa • J. P. Lalor • O. Olowe • J. Ganhotra • B. H. Hu • E. Habba • A. M. Bean • C. Liu • S. Land • S. Dillmann • A. Garikaparthi • E. Bandel • S. Imai • J. Edgell • W. M. Kennedy • J. Chim • P. Meusling • A. Kaeberlein • V. R. K. Chundi • M. Patwardhan • M. Ku • A. Meek • L. Knauer • B. Wingenroth • S. Yadav • U. Gohar • F. Friedrich • M. Lin • J. Mickel • A. Cohan • S. Biderman • I. Solaiman • Z. Talat • A. Reuel • M. Akhtar • G. Kasneci • A. Ghosh • L. Choshen
Every Eval Ever: A Unifying Schema and Community Repository for AI Evaluation Results.
Preprint (Jun. 2026). arXiv URL
Every Eval Ever: A Unifying Schema and Community Repository for AI Evaluation Results.
Preprint (Jun. 2026). arXiv URL
[9]
Z. Zhang • S. Yang • G. Kasneci
Consolidating Rewarded Perturbations for LLM Post-Training.
Preprint (May. 2026). arXiv GitHub
Consolidating Rewarded Perturbations for LLM Post-Training.
Preprint (May. 2026). arXiv GitHub
[8]
E. Zaradoukas • B. Prenkaj • G. Kasneci
Reinforcement Unlearning via Group Relative Policy Optimization.
ICLR 2026 - 14th International Conference on Learning Representations. Rio de Janeiro, Brazil, Apr 23-27, 2026. To be published. Preprint available. arXiv
Reinforcement Unlearning via Group Relative Policy Optimization.
ICLR 2026 - 14th International Conference on Learning Representations. Rio de Janeiro, Brazil, Apr 23-27, 2026. To be published. Preprint available. arXiv
[7]
S. Jacob • B. Prenkaj • W. Shao • G. Kasneci
TabSCM: A practical Framework for Generating Realistic Tabular Data.
Preprint (Apr. 2026). arXiv GitHub
TabSCM: A practical Framework for Generating Realistic Tabular Data.
Preprint (Apr. 2026). arXiv GitHub
[6]
F. Steinbauer • E. Oswald • K. Kirchheim • B. Prenkaj • F. Kofler • G. Kasneci
Efficient General Intelligence through Multi-Component Integration.
Preprint (Apr. 2026). URL
Efficient General Intelligence through Multi-Component Integration.
Preprint (Apr. 2026). URL
[5]
S. Firmani • F. Steinbauer • G. Kasneci • A. Marsico • M. Horlacher
RIBEX: Predicting and Explaining RNA Binding Across Structured and Intrinsically Disordered Regions (IDR)-rich Proteins.
Preprint (Mar. 2026). DOI
RIBEX: Predicting and Explaining RNA Binding Across Structured and Intrinsically Disordered Regions (IDR)-rich Proteins.
Preprint (Mar. 2026). DOI
[4]
C. Yuan • B. Ma • Z. Zhang • B. Prenkaj • F. Kreuter • G. Kasneci
Moral Lenses, Political Coordinates: Towards Ideological Positioning of Morally Conditioned LLMs.
Preprint (Jan. 2026). arXiv
Moral Lenses, Political Coordinates: Towards Ideological Positioning of Morally Conditioned LLMs.
Preprint (Jan. 2026). arXiv
2025
[3]
C. Wu • B. Ma • Z. Zhang • N. Deng • Y. He • Y. Xue
Evaluating Zero-Shot Multilingual Aspect-Based Sentiment Analysis with Large Language Models.
International Journal of Machine Learning and Cybernetics 16.10. Jun. 2025. DOI
Evaluating Zero-Shot Multilingual Aspect-Based Sentiment Analysis with Large Language Models.
International Journal of Machine Learning and Cybernetics 16.10. Jun. 2025. DOI
2023
[2]
Z. Zhang • H. Yang • B. Ma • D. Rügamer • E. Nie
Baby’s CoThought: Leveraging Large Language Models for Enhanced Reasoning in Compact Models.
BabyLM Challenge @CoNLL 2023) - 1st BabyLM Challenge at the 27th Conference on Computational Natural Language Learning. Singapore, Dec 06-10, 2023. DOI GitHub
Baby’s CoThought: Leveraging Large Language Models for Enhanced Reasoning in Compact Models.
BabyLM Challenge @CoNLL 2023) - 1st BabyLM Challenge at the 27th Conference on Computational Natural Language Learning. Singapore, Dec 06-10, 2023. DOI GitHub
[1]
M. Muschalik • F. Fumagalli • R. Jagtani • B. Hammer • E. Hüllermeier
iPDP: On Partial Dependence Plots in Dynamic Modeling Scenarios.
xAI 2023 - 1st World Conference on eXplainable Artificial Intelligence. Lisbon, Portugal, Jul 26-28, 2023. Best Paper Award. DOI
iPDP: On Partial Dependence Plots in Dynamic Modeling Scenarios.
xAI 2023 - 1st World Conference on eXplainable Artificial Intelligence. Lisbon, Portugal, Jul 26-28, 2023. Best Paper Award. DOI
©all images: LMU | TUM
Back to Top
2025-10-06 - Last modified: 2026-07-03