Deep Learning in Computational Science and Engineering
Ferienakademie 2022, Course 4
Dates
- First meeting (virtual) Monday July 18, 1pm
- APPLICATION DEADLINE MAY 10!
Short Description
Deep learning is a game changer in many fields of science and engineering. Deep learning has outperformed classical algorithms in tasks such as Regression, Classification, and Clustering in cases of abundance of data. Prominent examples stem from fields such as image processing or speech analysis. Images can be classified into cats or dogs, traffic lights or hydrants. Twitter feeds can be clustered into similar topics. And via regression, the price of houses can be predicted based on their images and Covid-19 infection rates based on historic data. All these examples have in common that they are only based on data, not on knowledge about the underlying problem. But can deep learning be also used in the field of Computational Science and Engineering? And if, for which tasks? What, if only very little experimental or simulation data is available? And can we include knowledge that we have about the underlying physics of the problem? How does deep learning work, anyways? And how useful is it?
In this course, we will discover the answer to these questions together. To this end, each participant will be given a specific task to work on and to present. But in contrast to classical seminars, we will have plenty of room for discussions and hands-on experience. We will cover basics such as the design of neural networks, regression, learning and automatic differentiation. But we will also address and discuss more advanced topics and high-level applications specific to Computational Science and Engineering. For example, how can we ensure that deep learning algorithms obey basic laws of physics, such as the conservation of mass and momentum so that we can trust in their predictions? How can we efficiently predict stresses from images or the distribution of pollutants in ground water flow based on limited data? We further aim to look at inverse questions such as finding flaws in objects from signals, the ambitious topic of creating meta models with Google's latest graph networks or the exploration of data-driven material models as an alternative to empirical models.
Language
As in the last years, the course language is something similar to English—so don't worry if your English is not perfect, you won't be alone!
Participants
Engineering, Informatics, Simulation Technology, Mathematics, Physics, and related fields of study (Bachelor from 3rd year or Master).
Further Information and Application
... via the Ferienakademie's website www.ferienakademie.de
APPLICATION DEADLINE MAY 10!