Recent advances in artificial intelligence have enabled unprecedented success in face and speech recognition, language translation, self-driving vehicles, and game playing. Some examples are FaceID
in iPhone X, automatic face tagging in Facebook
, speech recognition in Siri
, HeyGoogle, and Cortana, language translation, and skin cancer detection using the SkinVision mobile application in GooglePlay. For many of these problems where human-level performance is the benchmark, a wealth of deep learning methods
have been developed and tested. For instance, deep learning methods can detect skin cancer as good as dermatologists
. For many other important scientific problems, however, the full potential of deep learning has not been fully explored yet. In our group, we focus on investigating how such deep neural network methods may be developed to improve human health.
We are currently focused on solving a bioinformatics problem known as 'protein distance prediction'. This problem lies at the heart of a notoriously challenging problem known as 'protein folding' which is selected as one of the top 100 unsolved questions
by the Science magazine. Although the problem remains largely unsolved, deep learning shines another hope to solve it. Many groups, all around the world, are now attacking the problem with new success stories. The field has now become highly competitive. My research background and ever growing interest in deep learning uniquely positions me to research in this area. In a recent publication
, we demonstrated that adding dropout regularizers to residual networks can significantly improve the accuracy of deep learning methods that predict distances. We are currently investigating how deep learning algorithms can be best designed, engineered, and understood for predicting more accurate distances. We also collaborate with faculties and groups in various departments at and outside UMSL.
- University of Missouri-St. Louis Research Award
- The U.S. National Science Foundation