This is a machine learning class that consists of normal lectures, tutorials, theory exercises, and a project, you can find a general explanation of each part in general here.
The class itself is in my opinion heavily linear algebra based, but of course, you might feel different, if you have a more math based background like a Bachelor from ETH.
Grading
The final grade is based on the course project (30 %) and a written final exam (70 %).
Topics
- Principal Component Analysis
- Matrix Reconstruction
- Non-Negative Matrix Factorization
- Word embeddings
- Clustering Mixture Models
- Neural Networks
- Generative models
- Sparse Coding
- Dictionary Learning
Project
One can choose between the following four projects:
- Collaborative Filtering
- Text Sentiment Classification
- Road Segmentation
- Galaxy Image Generation
My team and I decided to tackle the problem of road segmentation.
The task here is to do image segmentation on RGB satellite images. So basically we are given some training data and have to build a model that discriminates between two classes road and non-road, but patch-wise, where each patch is of size 16×16, where the training data is of size 400×400 and the validation data is of size 608×608.
Part of the project is of course implement different models and also to write a report.
The project grade is composed of a competitive and a non-competitive part, where the competitive part contributes 30 % and the other 70 % to the grade.