On September 3^rd, 2021 at 03:00 pm Pierre Etienne Valentin Liotet, PhD Student in Information Technology, will hold the online seminar titled "Introduction to Causal Inference and Applications to Reinforcement Learning".


Most machine learning relies on the assumption that data is iid. However, this assumption seldom holds in realistic scenarios, thus producing undesired effects. Causal inference accounts for this limitation by adding more information to the framework, by considering underlying causal mechanisms.

Because causal learning is a broader framework that statistical learning, it requires a more careful analysis, but it possibly yields models with better generalization abilities.

Interestingly, from the current state of the machine learning field, reinforcement learning is the most suited to causal inference. Reinforcement learning natively deals with non iid data. Moreover, it considers the ability for the agent to intervene in its environment by way of actions, and interventions is a central concept of the causal inference framework. The association of causal inference and reinforcement learning into the so-called causal reinforcement learning is therefore promising.


To participate in the online event, please, use this link