Machine learning is no longer restricted to data analysis. It is currently used in theory, experiment and simulation. This is a sign that AI is becoming pervasive in all traditional aspects of research. However, are theorists, experimentalists and computational scientists aware of each other’s problems and the solutions developed to tackle them? Are researchers working in different areas of physics aware of developments in other areas? Industry is creating powerful AI tools that are expected to revolutionise science but how can academic researchers make the most out of them? This session explores how industry-academia collaborations work, how scientists can access industry-developed tools and how they can best benefit from them while feeding back their needs to shape future developments.
Chair - Professor Tony Hey
About the event
In this event we will hear from Professor Anima Anandkumar, Dr. Tie-Yan Liu and Professor Max Welling.
Dr. Tie-Yan Liu: AI-for-Science – the next wave of artificial intelligence
In previous decades, AI has achieved notable success in computer vision, speech recognition and natural language understanding. However, mimicking human vision, speech and language capabilities is just a shallow aspect of AI. It neglects the fact that we, as human beings, are unique because of our courage and ability to discover and change the world. AI-for-Science aims to build powerful tools to help natural scientists to better discover and change the world. Specifically, AI-for-Science assumes that the physical world can be theoretically characterised by fundamental scientific equations, usually at a very large scale. It also acknowledges that there is always a gap between theory and reality, and the evidence of the gap can be found in experimental data. No one has the capability to efficiently solve all those complex scientific equations, analyse huge amounts of experimental data, or create a closed loop between them. This is exactly where AI could play a disruptive role. As a showcase of such disruptions, I will introduce several research projects at MSR AI4Science, including Graphormer, an AI model for molecular dynamics simulation, DeepVortexNet, a neural PDE solver for fluid dynamics, SciGPT, an AI language model to automatically extract knowledge from scientific literature, and LorentzNet, an equivariant AI model to detect new particles from large-scale jet data. After introducing these works, I will also discuss some future trends of AI-for-Science research.
Professor Max Welling: Boosting scientific discovery using machine learning
The tools of AI, ML and DL are starting to transform the way we do science. In particular, we are seeing the start of a new paradigm where data generation through ab initio simulation of physical processes is recycled into machine learning models, which then accelerate these ab initio simulation models. For instance, when performing MD simulations of molecules, one would need quantum mechanical calculations to calculate the contribution of electrons. However, the information from many such simulations can be stored in so-called machine-learned force fields which can in turn accelerate the MD simulation. Analogous paradigm shifts are emerging for DFT calculations and PDE solving to name a few. In my presentation, I will talk about how AI is expected to impact the natural sciences beyond just data analysis and in reverse how the tools from the natural sciences are also making a positive impact on ML research.