A Virtual Workshop

Speakers

Frank Noé, Freie Universität Berlin, Germany

AI for the Sciences [video]

AI, and specifically deep ML methods have a profound impact on industry and information technology. But since recently AI methods are also changing the way we do science. In this talk I will present some of our recent efforts to build machine learning methods that attack fundamental problems in physical and chemical sciences: the sampling problem in physical many-body systems, and the solution of the quantum-chemical electronic Schrödinger equation. Key in making progress in these hard problems with ML is to interrogate the physical system about what the learning problem should be, and to encode physical structures, such as symmetries and conservation laws, into the ML model.

Kyle Cranmer, New York University, USA

How machine learning can help us get the most out of our highest fidelity physical models [video]

The physical sciences are replete with high-fidelity simulators: computational manifestations of physical theories. These simulators often incorporate experimental data or are composed of disparate phenomena that occur at different scales or regimes. Ironically, while these simulators provide our highest-fidelity physical models, they are not well suited for inferring properties of the model from data. I will formulate the emerging area of simulation-based inference and describe how machine learning techniques are well-suited for this task. Finally, I will provide examples of how these techniques can impact physics at the Large Hadron Collider, astroparticle physics, lattice field theory, molecular dynamics, and public health.

Sach Mukherjee, DZNE Bonn, Germany

Towards causal learning in very high dimensions

The biomedical sciences are full of systems that are extremely high dimensional but whose underlying causal structure is not well understood. On the other hand, a series of technological developments over several years has transformed our ability to make measurements on such systems. I will discuss some of our recent efforts to develop scalable methods to exploit high-dimensional data for causal learning. I will also discuss the assessment of causal learning in the scientific context and show results in which causal learners are tested empirically against experimental data from genome-wide molecular biology. Finally, I will discuss some of the many open questions in this area.

Shirley Ho, Flatiron Institute, New York, USA

Discovering Symbolic Models in Physical Systems using Deep Learning[video]

We develop a general approach to distill symbolic representations of a learned deep model by introducing strong inductive biases. We focus on Graph Neural Networks (GNNs). The technique works as follows: we first encourage sparse latent representations when we train a GNN in a supervised setting, then we apply symbolic regression to components of the learned model to extract explicit physical relations. We find the correct known equations, including force laws and Hamiltonians, can be extracted from the neural network. We then apply our method to a non-trivial cosmology example---a detailed dark matter simulation---and discover a new analytic formula that can predict the concentration of dark matter from the mass distribution of nearby cosmic structures. The symbolic expressions extracted from the GNN using our technique also generalized to out-of-distribution-data better than the GNN itself. Our approach offers alternative directions for interpreting neural networks and discovering novel physical principles from the representations they learn.

Gianni De Fabritiis, Barcelona Biomedical Research Park

From symbolic to neural network potentials in molecular simulations [video]

Neural network potentials are energy potentials which usually interpolate a function of the coordinates of the atoms to an energy value which are learned from the data and usually expressed in terms of a neural network. They can help bypassing fixed symbolic forms which have dominated molecular dynamics in the past decades and bring closer molecular mechanics and quantum simulations in a useful way. I will present several applications already available today of this new paradigm and future work in this direction.

Poster presentations

We invite short abstract submissions to be presented as posters at the workshop (no online proceedings). Submissions will be lightly reviewed by the organizers. Both novel works and recently published works that fit within the topic areas of the workshop are acceptable.

The poster session will be organised as follows:

• The posters will be made visible as large thumbnails on a gallary webpage, and can be further enlarged or downloaded. Each poster will be connected to a private breakout room in which participants can ask questions and the presenter may choose to show additional slides.
• The poster should be a one-page figure in PNG format, in portrait orientation, with 9:13 ratio (e.g. 4500x6500 pixels), and preferably less than 10 Mb.
• Accepted poster presenters are invited to submit a 1 minute advertisement video, which will be presented at the start of the poster session.

Registration

Registration is closed

Program

All times are Central European Summer Time (CEST).

Organisers

The AI4Science Kickoff Workshop is organised by:

• Bernd Ensing, University of Amsterdam
• Patrick Forré, University of Amsterdam