Join the Computer Science Department for this presentation in their colloquium series! Dr. Jiangwen Sun will present, “Impact of different loss functions in training DNN for understanding regulatory roles of DNA sequence in DNA methylation.”
Functonal genomic events, such as DNA methylation, regulate gene expression by moderating chromatin state, playing critical roles in various important biological processes, such as X-chromosome inactivation and aging.
Their aberrant activities have been linked to several diseases such as cancers, immunological disorders, and neurological disorders.
Dr. Sun will present the results from one of his recent efforts in this direction, training deep neural networks to predict DNA methylation from DNA sequence.
We studied the impact of the use of different loss functions, specifically cross-entropy loss and hinge loss in network training on various aspects of the trained network, including the prediction performance, learned sequence patterns, and uncertainly in predictions.
Keep an eye on the Department of Computer Science Colloquium Series here! Free and open to the public.