Filip Thor
Division of Scientific Computing, IT Department, Uppsala University
Deep learning has shown great performance in various fields of practice. It has, however, seen only limited application in genomic studies. Historically, genotype data has been expensive to generate, but with modern sequencing methods more data can be generated cheaper and faster. This makes data driven approaches interesting to investigate.
I will in this talk give an introduction to my PhD project, starting with a description of the data that is used and some application examples. I will specifically talk about dimensionality reduction using autoencoders, which is work previously done in the research group and serves as the foundation of my work. I will highlight some preliminary results on multi-GPU training and data augmentation techniques, and end with my future plans for my project.