Fundamentals of multimodal models

by Rob Firth (STFC)

Thursday 19 Mar 2026, 13:30 → 14:30 Europe/London

Thomson (Daresbury Labs)

This is the last of a series of four, AI themed, talks aimed at researchers who are interested in learning more about AI and its applications.  The talks are offered both online and in person.

1. Concepts and practical implementation using Agentic AI tools, 29 January
2. Ensuring data privacy and security with Federated Learning, 9 February
3. Concepts and application of surrogate models for accelerating simulations and solving complex problems, 17 March
4. Fundamentals of multimodal models, 19 March

This talk will outline what is meant by data modalities, and how they can be handled jointly. This is a capability that many will be familiar with through interaction with Multimodal LLMs using both text and images, but modalities (especially within scientific context) are much more diverse - including Sensor Data, Geospatial Information, Graphs and numerical data. We will cover ways that the data is represented, and how different models learn and combine representations. As well as real-world use cases, limitations will be highlighted..

Materials, including video recordings, will be made available after the event via the Hartree Centre Training Portal