Jürgen Hahn has been a research assistant at the Chair of Media Informatics at the University of Regensburg since 2017.
From 2010 to 2015 Jürgen Hahn studied media informatics at the University of regensburg. The title of his bachelor thesis was assembly training via smartglass-augmented reality at the industrial assembly area. In 2016 and 2017, in addition to his master’s degree in media informatics, he worked as a tutor for the courses Entertainment Computing, Game Engineering and Digitalization and Digital Society.
His research interests lie in the topics of human computer interaction, image processing and MR/AR systems.
Chair for Media Informatics
Chair for Media Informatics
Introduction to Computer Science and Media Informatics (BA, Exercise) and Interaction Techniques and Technologies (ITT) (Master). Most of this belongs to HCI
At ITT we use Support Vector Machines (SVM) for Activity Recognition with the Wiimote
Introduction to the basics of SVMs
Differences to non-linear methods (e. g. different types of neural networks)
Use of SVMs
In ITT we use SVMs to implement specific interaction techniques, which is why this makes sense in this seminar.
HCI and Programming language (PL)
Every now and then I wonder whether I am looking for a particular problem, e. g. I need machine learning, but I haven’t actually used AI technologies in my research. My research is more of a basic research and so far only problems or hurdles have arisen that have nothing to do with AI, but HCI and PL
Basics of machine learning (e.g. SVMs)
basics of common deep learning (convolutional NNs, recurrent NNs, feed-forward NNs, gradient descent) but very rusty without active application
Out of interest, I implemented a feed-forward NN myself in C without using libraries other than libc -> The NN was able to learn logical operators (worked very well)
I think you have to differentiate here. It is difficult when you want to understand what you are doing. So real knowledge of the basics of e-g- NNs with all the math and abstract methodology. On the other hand, it is easy to simply search for something on the internet that e.g. was made with PyTorch, Tensorflow, etc. and roughly fits what you want to do yourself. Hack the trial and error-wise so that it works halfway
What interests me most is where and when the current AI hype or AI summer ends. Which problems are so complex that we either don’t have the hardware and /or tha procedures. It wouldn’t be tha first time that an AI summer was followed by an AI winter.
AI is important in higher education and should definitely be addressed in any bachelor’s and master’s degree in computer science. Here a foundation should be built on the basis of fundamentals, the didactic challenge of which is to present complex and abstract facts clearly.
At the moment, in my opinion, it still happens too often that, as I said above, any PyTorch or Tensorflow projecty from the internet are quickly hacked by studnets so that they fit their problem. However, it is also noticeable that there is not much of the basics or §knowing what you are doing”. A CNN is then used and the question of what convolution is or why a CNN is suitable here is rarely answered meaningful.
Since I no longer do any teaching and have not had any AI topics in my research so far, I have nothing with which to contribute to the project “KnowledgeCreatesData”, at least in my opinion.
– I can’t think of anything right now
Mr. Hahn, thank you very much for taking the time to answer these questions. We wish you a nice day!