Application of Curriculum Learning for de novo design of small molecules
Typ
Examensarbete för masterexamen
Program
Publicerad
2021
Författare
Arango, Juan Diego
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
In recent years, Deep Learning has given new energy to the field of de novo design. This field is the
generation of novel chemical compound ideas, which can be used for new applications. AstraZeneca
has developed software for this task called REINVENT. The software uses Reinforcement Learning and
a user-defined scoring function to create new compound ideas. The objective of this work is to
implement, within REINVENT, the technique of Curriculum Learning. Here, the scoring function during
the training phase is actively modified. Besides the implementation, this work explores how this
approach improves performance compared to the classical Reinforcement Learning approach.