Applied Machine Learning - Week 3

Monday the 4th - Friday the 8th of May 2026

Monday 4th of May (afternoon):
Lectures: More unsupervised learning: Dimensionality reduction with t-SNE and UMAP.
     We will also shortly discuss Cleaning and imputing noisy data.

Exercise: The exercise starts with an introduction to dimensionality reduction.
     The "real" exercise is (preprocessing and) applying dimensionality reduction to the Cosmos2015 data.
     The data for the dimensionality reduction part 2 can be found on the data webpage.


Wednesday 6th of May (morning):
Lectures: An introduction to Convolutional Neural Networks.
     Very cool visualisation of how a CNN works on your handwritten numbers.
     For calculating the output size of convolutional outputs, use the ConvNet Calculator.
     We will also discuss potential datasets for Final Project.
     Potentially also: The ImageNet Competition - a revolution in ML.

Exercise: The exercise is to recognise handwritten numbers with a Convolutional Neural Network: CNN_MNISTdata.ipynb.
     If you want a step up in complexity, consider classifying volcanic ash (socalled 'tephra") from Greenlandic ice cores, based on images of the tephra.
     This exercise also requires two CSV files with metadata: train and test sample.


Wednesday 6th of May (afternoon):
Lectures: Analysing geometric data using Graph Neural Networks (which are fantastic!).
     Example usage of GNN for classification and regression on IceCube data.

Exercise: The exercise is to classify which molecules (which can be represented by a graph) are potentially HIV inhibiting.

---

Last updated: 24th of April 2026 by Troels Petersen.