ATML: Advanced Topics in Machine Learning

Welcome

This course is designed for students who aspire to become technical experts, keep up with the changing field, pursue research, or innovate in the field of machine learning with new models and algorithms. Machine learning is evolving continuously and is becoming more complex. At the same time, basic ML tasks of model initialisation, dataset selection, training and evaluation is being automated by powerful AI systems, without need for too much human intervention. It is therefore important that today's programmers, datascientists and engineers go beyond the tools of simply applying machine learning, and develop an understanding of the deeper concepts,  how to improve upon them and keep up with the emerging field. 

Thus, this course will expose students such critical aspects in machine learning that will help students to understand issues not covered in more basic machine learning courses, to read and understand new ideas as they emerge and to develop their own techniques. 

 

Relevant links

Course structure

This is an advanced course to introduce students to several areas of machine learning. There is one track for each area; in 2025-26 there are three such tracks (see below). The exam for the course will have one question (section) for each track, and students can answer any two. 

We recommend that students follow all the tracks during lectures. For the exam, they can choose to prepare all three tracks or any two. 

See introduction slides for details. Come to the first lecture on 12th January!

Supporting materials

As the course proceeds, along with slides, following types of materials will be provided:

  • Examples of questions that can appear in exam
  • Additional exercises and code samples (programming is optional)
  • Exercises for tutorials 

 

Track topics and lecture schedule

The course will cover three tracks corresponding to three subjects in 2025-26. Each track will run independent of others -- think of each as a mini-course. Final exam will require answering questions from two tracks out of three. For now, you should attend all tracks and all lectures. 

Tack 1: Optimisation in machine learning (track page)

Topics: non-convex optimisation and neural networks, loss landscapes in high dimensions, algorithmic stability, overfitting and overparameterisation in neural networks.

  • Mondays 17:10 -- 18:00
  • Lecture Theatre B, 40 George Square
  • Instructor: Rik Sarkar

Track 2: Deep generative modelling (track page)

Topics: learning and inference in families of neural probabilistic models used for high-dimensional data: deep latent variable models (e.g., autoencoders), generative adversarial models, normalising flows, diffusion models. Principles of representation learning, generalisation, model assessment and deployment considerations.

  • Tuesdays 17:10 -- 18:00
  • Room 425 – Anatomy Lecture Theatre, Doorway 3 - Medical School, Teviot
  • Instructor: Nikolay Malkin

Track 3: Geometric learning (track page)

Topics: graph neural networks, models on non-Euclidean domains (e.g., manifolds such as spheres), symmetries, invariances and equivariances.

Auditing the course

Everyone is welcome to attend lectures.

 

Tutorials

Tutorials will start in week 3. Exercises and problems will be provided beforehand. Each week there will be three tutorial slots where different lecturers and tutors will be available to answer student questions and help with exercises. The schedule of availability will be published before tutorials start. 

 

License
All rights reserved The University of Edinburgh