Web page of Philipp Geiger

Overview

Welcome! My name is Philipp Geiger, on this web page I give an overview over my work in research and software applications, and general background. I'm a research scientist at Bosch Center for Artificial Intelligence. Before, I did my doctorate in computer science and a postdoc at Max Planck Institute for Intelligent Systems and University of Stuttgart, including a stay at Microsoft Research Cambridge, and my diplom in mathematics with philosophy as side subject at Heidelberg University and Humboldt University of Berlin.

Generally, I conduct research and development in the areas of machine learning (core part of artificial intelligence) and multiagent systems (analysis and optimization of multiple intelligent decision makers' interaction behavior). Often, this combines project-driven software engineering of machine learning pipelines (currently mainly Python/PyTorch) and experimental aspects with statistical/mathematical aspects. With this, I aim to contribute to the analysis and solution of application problems that are important to business and society as a whole. Currently, I am focusing on multiagent imitation learning and reinforcement learning often combined with aspects of deep learning and generative models, with application (proof of concept and protoypes) to data-driven driving behavior models, autonomous driving and its safety validation (see also further background).

Selected publications (see also all publications and further material):

1. Fail-Safe Adversarial Generative Imitation Learning. (2022). TMLR.
2. Learning game-theoretic models of multiagent trajectories using implicit layers. (2021). AAAI.
3. Coordinating users of shared facilities via data-driven predictive assistants and game theory. (2019). UAI. [Slides.]

Further links: my profiles on Google Scholar, DBLP, OpenReview, GitHub.

News

• Analyzing Closed-loop Training Techniques for Realistic Traffic Agent Models in Autonomous Highway Driving Simulations:

  At Bosch Research, we conducted an extensive comparison of different approaches for (generative) imitation learning of realistic traffic agents from real-world human driving data, for driving behavior simulations, check out our preprint.

  [Posted on 2024-12-27.]

• On Learning the Tail Quantiles of Driving Behavior Distributions via Quantile Regression and Flows:

  In our recent ITSC workshop paper, we investigate two approaches for estimating the (tail) quantiles of driver action distributions (policies), including a version of conditional normalizing flows.

  [Posted on 2023-06-08. Updated on 2024-08-22.]

• Fail-Safe Adversarial Generative Imitation Learning paper accepted at TMLR:

  Using the flexible scalability of deep generative imitation learning, adding guaranteed safety via worst-case sample-based reachability analysis, while keeping end-to-end trainability – check out our paper accepted at TMLR.

  [Posted on 2022-08-05. Updated on 2022-10-10.]

Publications

All peer-reviewed publications

1. Tee, J. Y., De Candido, O., Utschick, W., & Geiger, P. (2023). On Learning the Tail Quantiles of Driving Behavior Distributions via Quantile Regression and Flows. 2023 IEEE 26th International Conference on Intelligent Transportation Systems (ITSC). [publication.]
2. Geiger, P., & Straehle, C.-N. (2022). Fail-Safe Adversarial Generative Imitation Learning. Transactions on Machine Learning Research. [publication.]
3. Geiger, P., & Straehle, C.-N. (2021). Learning game-theoretic models of multiagent trajectories using implicit layers. Proceedings of the Thirty-Fifth AAAI Conference on Artificial Intelligence (AAAI). [publication.]
4. Etesami, J., & Geiger, P. (2020). Causal Transfer for Imitation Learning and Decision Making under Sensor­‐shift. Proceedings of the Thirty-Third AAAI Conference on Artificial Intelligence (AAAI). [publication.]
5. Geiger, P., Besserve, M., Winkelmann, J., Proissl, C., & Schölkopf, B. (2019). Coordinating users of shared facilities based on data-driven assistants and game-theoretic analysis. Proceedings of the 35th Conference on Uncertainty in Artificial Intelligence (UAI). [publication, slides.]
6. Geiger, P., Hofmann, K., & Schölkopf, B. (2016). Experimental and causal view on information integration in autonomous agents. Proceedings of the 6th International Workshop on Combinations of Intelligent Methods and Applications (CIMA), 21–28. [publication, slides.]
7. Geiger, P., Zhang, K., Gong, M., Janzing, D., & Schölkopf, B. (2015). Causal inference by identification of vector autoregressive processes with hidden components. Proceedings of the 32nd International Conference on Machine Learning (ICML). [publication, slides.]
8. Gong, M., Zhang, K., Schoelkopf, B., Tao, D., & Geiger, P. (2015). Discovering temporal causal relations from subsampled data. Proceedings of the 32nd International Conference on Machine Learning (ICML). [publication.]
9. Geiger, P., Janzing, D., & Schölkopf, B. (2014). Estimating causal effects by bounding confounding. Proceedings of the 30th Conference on Uncertainty in Artificial Intelligence (UAI). [publication, supplement, slides.]

Preprints

1. Bitzer, M., Cimurs, R., Coors, B., Goth, J., Ziesche, S., Geiger, P., & Naumann, M. (2024). Analyzing Closed-loop Training Techniques for Realistic Traffic Agent Models in Autonomous Highway Driving Simulations. ArXiv Preprint ArXiv:2410.15987.
2. Geiger, P., Carata, L., & Schoelkopf, B. (2016). Causal inference for data-driven debugging and decision making in cloud computing. ArXiv Preprint ArXiv:1603.01581. [publication.]

Theses

1. Geiger, P. (2017). Causal models for decision making via integrative inference [PhD thesis]. [publication, slides.]
2. Geiger, P. (2012). Mutual information and Gödel incompleteness [Diploma thesis]. [publication.]

Notes

1. Geiger, P. (2016). Notes on socio-economic transparency mechanisms. ArXiv Preprint ArXiv:1606.04703. [publication.]

Further material, slides

1. On Mathematical Guarantees in Machine Learning for Safe Autonomous Driving. (2023). Mathematics in Sciences, Engineering, and Economics Symposium, Karlsruhe Institute of Technology. [, slides.]

Bio

Short bio:

• Currently, I am a research scientist at the Bosch Center for Artificial Intelligence (BCAI), Renningen, Germany.
• From 2017 to 2018, I was a postdoctoral researcher at the Empirical Inference Department of the Max Planck Institute for Intelligent Systems, Tübingen, Germany, working with Bernhard Schölkopf.
• From 2013 to 2017 I pursued my PhD (Dr. rer. nat.) under the supervision of Dominik Janzing, Bernhard Schölkopf and Marc Toussaint at the Max Planck Institute for Intelligent Systems and the University of Stuttgart, Germany.
• In 2015, I did an internship at Microsoft Research Cambridge, United Kingdom, supervised by Katja Hofmann.
• From 2006 till 2012, I studied mathematics and philosophy at Humboldt University of Berlin and at Heidelberg University, Germany.
• In 2005, I finished high school in Karlsruhe, Germany

Areas of research, implementation and further interests:

• Methodology: mathematics (proving), imitation learning, inverse reinforcement learning, interpretable/causal machine learning, generative/probablistic models (in particular generative adversarial networks (GANs), normalizing flows), deep/representation learning, statistics, multiagent systems, game theory, sequential decision making, dynamical systems and mathematical logic (in particular Goedel incompleteness, computability theory).
• Coding: software engineering in projects, in particular data/machine learning pipelines.
• Applications: autononomous and highly automated driving, resource-efficient sharing economy.
• Further areas of interest (without research contributions yet): economics, politics and psychology.

Contact

Name: Philipp Geiger,
email address (at Bosch): (insert first name).w.(insert last name)@de.bosch.com.

See also my MPI web site (it's still valid, even though I moved away from the MPI).

About this web page: I used these tools to build this web page as well as my CV:

• Font Awesome,
• Jekyll (static CMS).