Applied Machine Learning - Week 4

Monday the 12th - Friday the 16th of May 2025

See Andrej Kaparthy's description of how to reproduce GPT2 from scratch (long but fantastic).


Monday 12th of May (afternoon):
Lectures: We will be discussing (Variational) Auto-Encoders and anomaly detection.

Exercise: The exercise is about classifying digits i.e. the MNIST data, but this time in an unsupervised manner with an autoencoder.
     Following this we will repeat the exercise, this time with a Variational AutoEncoder. These techniques are also useful for anomaly detection in images.


Wednesday 14th of May (morning):
Lectures: Time series, Natural Language Processing (NLP), and Transformers.

Exercise: First, consider the simple example LSTM making FlightPassengerPredictions.ipynb on 1949-1961 airline data.
     Try to train a transformer Language Model for Natural Language Processing (NLP) on Google Colab (45 minutes!).
     Alternatively, use an LSTM to do Natural Language Processing, in this case estimate if a review was good or bad based on IMDB data.


Wednesday 14th of May (afternoon):
Lectures: Final projects kickoff. Introducing Potential Datasets and Projects (TP).

Exercise: The exercise simply to get set up for the final project. It is highly advantages to get hold of your data (or part of it) before this.
     During these exercises you should aim to settle your final project groups and discuss what data to work on.
     Things to consider are data, group, goals of analysis, and methods. We will sit down with you and talk about this.

---

Last updated: 10th of May 2025 by Troels Petersen.