Causality and Reinforcement Learning

Organized by: Prof. Joschka Boedecker, Hanne Raum
Course number: 411LE13S-7344-M
Language: English
Introductory Lecture: 18.4.2024, 10:00-12:00, IMBIT, Georges-Köhler-Allee 201, Nexus-Lab
ILIAS link

Overview:

While machine learning excels at identifying correlations within data, it often falls short in uncovering and understanding the underlying causal mechanisms, leaving the fundamental question of "why?" unanswered. Integrating causal relationships can help transfer models to different environments, ensure invariance to distribution shifts, provide clearer explanations for black box models or help to find compact meaningful models.
In this seminar, we will explore how causality can help us in different machine learning approaches.

Format:

The course will be given in person, in the form of a block seminar, where papers are read and presented by students

Application process:

Submitting a short motivational statement at ILIAS.

Paper Selection:

1.	A Causality Inspired Framework for Model Interpretation
2.	Causal Discovery for Non-stationary Non-linear Time Series Data Using Just-In-Time Modeling
3.	Causal Influence Detection for Improving Efficiency in Reinforcement Learning
4.	Causality-driven Hierarchical Structure Discovery for Reinforcement Learning
5.	Counterfactual Data Augmentation using Locally Factored Dynamics
6.	Deep learning of causal structures in high dimensions under data limitations
7.	Explainable Reinforcement Learning via a Causal World Model
8.	Generalizing Goal-Conditioned Reinforcement Learning with Variational Causal Reasoning
9.	Learning Independent Causal Mechanisms
10.	Learning to induce causal structure
11.	Explanations for Occluded Images
12.	Variational Causal Dynamics: Discovering Modular World Models from Interventions
13.	What Went Wrong? Closing the Sim-to-Real Gap via Differentiable Causal Discovery
14.	Causal machine learning for predicting treatment outcomes
15.	Neural Causal Structure Discovery from Interventions
16.	Causal Reinforcement Learning using Observational and Interventional Data