Public defence in Computer Science, M.Sc. Tianyu Cui
Opponent: Professor Christopher Yau, University of Oxford, England
Custos: Professor Samuel Kaski, Aalto University School of Science, Department of Computer Science
The thesis is publicly displayed 10 days before the defence in the publication archive Aaltodoc of Aalto University.
Public defence announcement:
Deep learning has been playing an essential role in our daily lives, from machine translation to photo editing. However, in the fields involving crucial decision-making that may cost much, such as biomedicine, more work still needs to be done. Genetics studies how our genes inherited from parents affect our traits. This enables scientists to understand the biological mechanism of diseases and even predict the risk of getting diseases. Genetics has not benefited much from the booming of deep learning, because of the lack of labelled data in genetics and the black-box nature of deep learning. This dissertation proposes novel deep learning approaches with probabilistic and interpretable methods to tackle the challenges of applying deep learning in genetics.
This thesis demonstrates how to improve detecting interactions between genes with deep learning and how to estimate the uncertainty of detected gene-gene interactions. Moreover, the thesis proposes several techniques to incorporate existing knowledge from genetics into deep learning, which is useful when the labelled training data is small. We test the proposed methods with several real-world genetics datasets, and our results show that deep learning can reveal new findings in genetics that current methods may ignore.