Causal methods hold significant potential to improve the explainability and robustness of AI systems. These methods enable the discovery and estimation of cause and effect, which is critical for human-like cognition. While research on causal techniques for supervised learning has flourished in recent years, there is a lack of such work in the domain of reinforcement learning (RL). This oversight is surprising given RL’s inherently interventional nature, which aligns well with the principles of causality. My ongoing PhD project is focused on leveraging causal techniques within RL systems to bolster their explainability, safety, and robustness. The primary objective is to develop causal discovery methods which would help to explain behaviour of RL agents.
I chose to do my PhD with the STAI CDT because it is in perfect alignment with my research interests in explainable and robust AI, causality, and automated decision-making for critical applications. I value the cohort-based approach, which fosters learning from fellow PhD students and cultivates relationships with researchers who share similar interests.
–
Masters Qualification: MSc Econometrics, Erasmus University Rotterdam
Undergraduate Qualification: BSc (Hons) Mathematical Economics & Data Science, Higher School of Economics
Work Experience:
- Data Scientist, causaLens (AI Company)
- Quantitative Analyst, London and Netherlands
Website: https://maxanisimov.github.io/
