09:00 - 09:30 | Registration, morning refreshments, and poster session 1 |
09:30 - 09:35 | Welcome - Guy Brown, Deputy Director of Centre for Machine Intelligence, University of Sheffield (YouTube Video) |
09:35 - 10:00 | Introduction: exploring multimodal AI beyond vision and language, Haiping Lu (YouTube Video) |
10:00 - 10:40 | Keynote 1: Daniel Zügner, Microsoft Research AI4Science (YouTube Video) |
| Title: MatterGen: a generative model for inorganic materials design Abstract: The design of functional materials with desired properties is essential in driving technological advances in areas like energy storage, catalysis, and carbon capture. Traditionally, materials design is achieved by screening a large database of known materials and filtering down candidates based on the application. Generative models provide a new paradigm for materials design by directly generating entirely novel materials given desired property constraints. In this talk, we present MatterGen, a generative model that generates stable, diverse inorganic materials across the periodic table and can further be fine-tuned to steer the generation towards a broad range of property constraints. To enable this, we introduce a new diffusion-based generative process that produces crystalline structures by gradually refining atom types, coordinates, and the periodic lattice. We further introduce adapter modules to enable fine-tuning towards any given property constraints with a labeled dataset. Compared to prior generative models, structures produced by MatterGen are more than twice as likely to be novel and stable, and more than 15 times closer to the local energy minimum. After fine-tuning, MatterGen successfully generates stable, novel materials with desired chemistry, symmetry, as well as mechanical, electronic and magnetic properties. Finally, we demonstrate multi-property materials design capabilities by proposing structures that have both high magnetic density and a chemical composition with low supply-chain risk. We believe that the quality of generated materials and the breadth of MatterGen’s capabilities represent a major advancement towards creating a universal generative model for materials design. | |
10:40 - 11:20 | Community talks |
11:20 - 11:50 | Break and poster session 2 |
11:50 - 12:30 | Keynote 2: Maria Liakata, Queen Mary University of London (YouTube Video) |
| Title: Longitudinal language processing for dementia Abstract: While the advent of Large Language Modes (LLMs) has brought great promise to the field of AI there are many unresolved challenges especially around appropriate generation, temporal robustness, temporal and other reasoning and privacy concerns especially when working with sensitive content such as mental health data. The programme of work I have been leading consists in three core research directions: (1) data representation and generation (2) methods for personalised longitudinal models and temporal understanding (3) evaluation in real-world settings, with a focus on mental health. I will give an overview of work within my group on these topics and focus on work on longitudinal monitoring for dementia. | |
12:30 - 12:40 | Group Photos |
12:40 - 14:00 | Lunch and poster session 3 |
14:00 - 14:40 | Keynote 3: Nataliya Tkachenko, Lloyds Banking Group (YouTube Video) |
| Title: Ethical challenges for multimodal conversational banking & parametric insurance Abstract: Ever since mass-propagation of generative AI models, multimodal data has been getting increased attention from the customer-focused industries. Multimodal chatbots, which can process and respond to customer queries using enriched context, such as text, voice, and even visual data, offer significant advantages in customer banking and parametric insurance by enhancing user interaction, speed and overall service efficiency. Customers now have an option to choose their preferred mode of communication, whether through typing, speaking, or even using gestures. By analysing customer data from various sources, chatbots can offer personalised financial advice, investment recommendations, and alert about unusual activities. They even can help with the immediate payouts, by promptly verifying predefined parameters, such as weather data for crop insurance for example. However, with enriched context also come multi-dimensional ethical considerations, such as bias, fairness, transparency and confabulations. In this presentation I will cover how these risks emerge and mutually diffuse in highly automated interfaces. | |
14:40 - 15:20 | Community talks |
15:20 - 15:30 | Break |
15:30 - 16:10 | Keynote 4: Adam Steventon, Our Future Health |
| Title: An incredibly detailed picture of human health: the exciting potential of Our Future Health to prevent, detect and treat diseases Abstract: In this presentation, I will detail the groundbreaking efforts of Our Future Health to construct a multimodal dataset encompassing 5 million individuals, representative of the UK’s diverse population. I will explore the transformative potential of this dataset to enhance our capabilities in predicting, detecting, and treating major diseases. Additionally, I will discuss the roles of artificial intelligence in this context, focusing on the opportunities and challenges it presents. This exploration will underscore the potential of AI and large-scale data in shaping the future of healthcare. | |
16:10 - 16:40 | Panel discussion |
| - What are the major barriers to deploying multimodal AI systems in real-world applications?
- How can we best identify and utilise diverse data sources to advance multimodal AI research and applications?
| |
16:40 - 17:00 | Best talk/poster prize winner announcement and closing |
17:00 - 17:30 | Tea/coffee and networking |