Machine Learning WS 2021/22 - Lampert, Grillmeyer, Todorkov

In this video the students present their final project for the HCI Machine Learning course. The main task was to teach a Neural Network various movements and then control a slideshow with different gestures. Bonus points could be gained by applying the Neural Network to other applications, like controlling a game, or by leveraging more advanced Data Processing, Data Analysis, Data Visualisation or Machine Learning techniques.

The students were not allowed to use any Machine Learning framework like Tensorflow, Keras or PyTorch and thus had to implement all components needed for classification problems with Neural Networks by themselves: forward propagation with various activation functions, loss functions, backpropagation, optimisation algorithms (Gradient Descent (with Momentum), Adam, etc.). This means they not only had to acquire a deep understanding of the mathematical foundations of Machine Learning, but also had to implement these theories in a computationally efficient way using only Python and Numpy.

Used technologies:

1. Python 3
2. Numpy, Pandas, matplotlib
3. Mediapipe (for keypoint detection) https://mediapipe.dev
4. Reveal.js (for the slideshow) https://github.com/hakimel/reveal.js
5. Sanic (webserver framework for slideshow) https://sanicframework.org/en/

Machine Learning Lecture by Prof. Dr. Marc Erich Latoschik
Machine Learning Exercise supervised by Christian Schell

Видео Machine Learning WS 2021/22 - Lampert, Grillmeyer, Todorkov канала HCI Group Würzburg