| Abstract: |
|
Over the last few years, rapid progress in deep learning and the availability of high performance GPU-based parallel deep learning frameworks have resulting in a large number of new techniques for supervised learning, regression, semi-supervised learning, and unsupervised learning. Scientific computing itself is a highly diverse field, ranging from numerical solutions to mathematical problems to practical image analysis, engineering and control problems. Scientific computing often also has special requirements for incorporating prior knowledge, as well as validating and explaining solutions. In my talk, I will use a number of successful applications of deep learning in biology, medicine, physics, and ecology to illustrate how these issues are addressed in real-world scientific applications. In the second part of my talk, I will discuss how cutting edge deep learning techniques, including distillation, recurrent neural networks, deep reinforcement learning, and generative models have the potential of greatly simplifying and speeding up many scientific computing tasks. |
|
;){kind=link}