This four‑day tutorial introduces participants to modern methods for analysing high‑dimensional scientific data, with a strong emphasis on autoencoders (AEs) and variational autoencoders (VAEs) as practical tools for discovering low‑dimensional structure in complex systems. The school blends theory, PyTorch‑based hands‑on sessions, and focused case studies drawn from real research problems in materials science, molecular simulation, and computational biology.

The first two days provide the mathematical and computational foundations: a refresher on neural networks, an intuitive and formal introduction to autoencoders and VAEs, and guided PyTorch exercises. We also introduce graph neural networks (GNNs) and graph VAEs at a conceptual level. Because GNNs are substantially more complex than standard neural networks, we do not expect participants to become fully proficient in their use during the school; however, the combination of lectures, examples, and AI‑assisted coding will bring participants close to being able to apply these methods in their own work.

The first day of the tutorial will be online, and days 2, 3 and 4 will have a hybrid onsite and online format with participants encouraged to be onsite if possible due to their and focus on case studies and group‑based problem solving. Participants will work in small teams on representative problems inspired by their own research interests and by the case studies presented during the tutorial, including:

• dimensionality reduction of molecular dynamics trajectories,

• latent‑space regression of physical or biochemical properties,

• stability and sensitivity analysis of learned representations,

• and introductory graph‑based modelling for molecular systems.

Throughout the tutorial, participants will gain practical experience implementing models in PyTorch, interpreting latent spaces, and applying representation‑learning techniques to real scientific datasets. The goal is to equip researchers with a solid working understanding of VAEs and a pathway toward using more advanced architectures such as GNNs and GVAEs in their own research.

Location

The tutorial will take place at University College Dublin

Who Should Apply?

This is an advanced tutorial designed for those with a solid foundation in machine learning. Because we have a full schedule focused on high-level topics, we cannot accept applicants who require training in basic ML principles. Selection is competitive and will be based on the statement of motivation provided during registration. There is no fee to register for this event.

Where to Stay

Accommodations are the responsibility of the participant. However, to help mitigate the high cost of stays in Dublin, we have negotiated a discounted rate for onsite housing at UCD. Rooms can be booked for any window between 22–26 May 2026 (minimum stay of two days).  The code providing a reducion in the  cost of accommodation will be provided to those registered for the tutorial.