Imprecise Probabilistic Machine Learning

This seminar explores the emerging field of imprecise probabilistic machine learning (IPML). While probability theory is the standard mathematical framework for modeling uncertainty and randomness in machine learning, its reliance on single, precise probability distributions often falls short when capturing the multifaceted uncertainties inherent in complex real-world systems. This limitation can lead to undesirable model behavior in practice. To address this, researchers are increasingly turning to generalizations of standard probability theory, encompassing approaches like Dempster-Shafer theory, interval-valued probabilities, the Choquet integral, upper/lower probabilities, and comparative probabilities. Though distinct, these methods all fall under the unifying framework of imprecise probability (IP).

This seminar offers participants a deep dive into the theoretical foundations and practical applications of imprecise probability (IP) in machine learning. Through the reading, presentation, and discussion of curated research papers, we will explore the field's breadth, from philosophical debates surrounding the nature and interpretation of probability to cutting-edge applications in areas such as classification, conformal prediction, out-of-distribution generalization, reinforcement learning, causal inference, foundation models, and large language models (LLMs).

Format

As the field of IPML is still immature, the goal of this seminar is to explore the foundation of IP and its applications in machine learning. We will specifically focus on the following topics: 

1. Topic 1: Foundations and Representations of Imprecise Probability
2. Topic 2: Imprecise Classification and Regression
3. Topic 3: Conformal Prediction
4. Topic 4: Uncertainty Quantification
5. Topic 5: Imprecise Probabilistic Forecast and Calibration
6. Topic 6: Decision Making with Imprecise Probability
7. Topic 7: Imprecise Probability in Modern ML (Deep Learning, Foundation Models, LLM, GenAI)
8. Topic 8: Use Cases of Imprecise Probability in Fairness, Privacy, Ethics, Safety, etc

The students will be assigned one topic out of the above areas. Based on their preference and academic background, they will then pick 1-2 papers to study in detail, including the related literature. After studying the paper(s), they must submit the initial report and presentation for feedback. After receiving the feedback, the students prepare the presentation and then deliver it to the rest of the class. Finally, the students submit the final report and presentation.

Schedule

• Topic Assignment (8 May 2025) -- Slides
• Submit initial report and presentation (June 2025)
• Receive feedback on the initial report and presentation (mid-June)
• Student presentation (July-August 2025)
• Submit the final report and presentation (September 2025)

Deliverable

Students who participate in this seminar are expected to deliver

1. Report
   • A summary of the topic of your choice (1-2 papers)
   • A template is available here.
2. Presentation
   • 30 minutes talk + 15 minutes for Q&A
   • Your classmates are your target audience
   • A template is available here.

IMPORTANT: The live presentation is the primary determinant of the final grade. It must clearly demonstrate a thorough understanding of the assigned paper(s), which will be further evaluated during the Q&A session. While the final report contributes to the grade, it carries less weight. The use of GenAI for writing assistance is permitted; however, students remain fully responsible for any resulting scientific misconduct, including plagiarism. 

Contact

Dr. Krikamol Muandet (muandet@cispa.de)
CISPA-D2 building, Im Oberen Werk 1, 66386 Sankt Ingbert

Anurag Singh (anurag.singh@cispa.de)
CISPA-D2 building, Im Oberen Werk 1, 66386 Sankt Ingbert