causality-inspired ML

Learning Dynamic Attribute-factored World Models for Efficient Multi-object Reinforcement Learning

In many reinforcement learning tasks, the agent has to learn to interact with many objects of different types and generalize to unseen …

Fan Feng, Sara Magliacane

AdaRL: What, Where, and How to Adapt in Transfer Reinforcement Learning

One practical challenge in reinforcement learning (RL) is how to make quick adaptations when faced with new environments. In this …

Biwei Huang, Fan Feng, Chaochao Lu, Sara Magliacane, Kun Zhang

AdaRL: What, Where, and How to Adapt in Transfer Reinforcement Learning

Factored Adaptation for Non-Stationary Reinforcement Learning

Dealing with non-stationarity in environments (e.g., in the transition dynamics) and objectives (e.g., in the reward functions) is a …

Fan Feng, Biwei Huang, Kun Zhang, Sara Magliacane

Domain Adaptation by Using Causal Inference to Predict Invariant Conditional Distributions

We show that can use labelled data in source domains and unlabelled data in the target domain to identify features that are robust to the specific shift that happens in the target dataset. While we use ideas from causality, we do not need to recover the causal graph (which in this case is not identifiable) to find these sets of features.

Sara Magliacane, Thijs van Ommen, Tom Claassen, Stephan Bongers, Philip Versteeg, Joris M Mooij