Curiosity driven reinforcement learning for molecular design

Organic molecules are used in many areas, for example in drugs or solar cells. However, the development of new organic molecules with desired properties is difficult, time-intensive, and requires a lot of trial and error. A new line of research uses artificial intelligence to partially automate this process and make it more reliable. This project aims to further advance the approach. Specifically, the use of curiosity-driven reinforcement learning will be explored. Reinforcement learning is a kind of algorithm, that teaches an artificially intelligent agent to achieve a defined goal, like predicting molecules with desired properties. Curiosity in reinforcement learning generally helps the agent to explore its environment. Here, it has the potential to help agents better explore the chemical space, which can lead to faster development of better molecules that were previously overlooked by human chemists.

Faculty Supervisor:

Alán Aspuru-Guzik

Student:

Partner:

Georg-August-Universität Göttingen

Discipline:

Computer science

Sector:

Education

University:

University of Toronto

Program: