Machine Learning and the Physical Sciences, NeurIPS 2024

About

The Machine Learning and the Physical Sciences workshop aims to provide an informal, inclusive, and leading-edge venue for discussing research and challenges at the intersection of machine learning (ML) and the physical sciences (PS). This includes the applications of ML to problems in the physical sciences (ML for PS) as well as developments in ML motivated by physical insights (PS for ML).

Physical sciences are defined inclusively, including but not limited to physics, astronomy, cosmology, chemistry, biophysics, materials science, and Earth science.

Recent years have highlighted unique opportunities as well as challenges in incorporating ML workflows as part of the scientific process in many physical sciences. For example, fields focused on fundamental physics discovery, such as particle physics and cosmology, often have stringent requirements for exactness, robustness, and latency that go beyond those typically encountered in other scientific domains and industry applications. Data preservation and workflow reproducibility are other central challenges that need to be addressed in the era of large experiments, collaborations, and datasets. In these fields and others, simulations play a central role in connecting theoretical models to observations. The ubiquity and increasing complexity of simulators in PS has spurred methodological advances in ML, e.g. in simulation-based inference and differentiable programming, that are finding applications far beyond PS, showcasing the bidirectional nature of the PS-ML intersection.

The breadth of work at the intersection of ML and physical sciences is answering many important questions for both fields while opening up new ones that can only be addressed by a joint effort of both communities. By bringing together ML researchers and physical scientists who apply and study ML, we expect to strengthen the much needed interdisciplinary dialogue, introduce exciting new open problems to the broader community, and stimulate the production of new approaches to solving challenging open problems in the sciences. Invited talks from leading individuals in both communities will cover the state-of-the-art techniques and set the stage for this workshop, which will also include contributed talks selected from submissions.

The invited talks program will showcase unique features of the physical sciences that highlight current challenges and bidirectional opportunities in ML and PS. This includes the central role of simulators in the scientific process, the need for rigorous uncertainty quantification, and the development of hardware-software co-design solutions for real-time inference.

A part of the workshop program will be dedicated to the focus area discussing the role of data-driven vs inductive bias-driven methods in machine learning and the physical sciences, centering the emerging role of foundation models and their complementarity with approaches leveraging physical inductive biases. This will feature an overview talk, followed by a moderated panel discussion.

NeurIPS 2024

The Machine Learning and the Physical Sciences 2024 workshop will be held on December 15, 2024 at the Vancouver Convention Center in Vancouver, BC, Canada as a part of the 38th annual conference on Neural Information Processing Systems (NeurIPS). The workshop is planned to take place in a hybrid format inclusive of virtual participation.

Speakers

Thea Aarrestad
ETH Zurich
Yasaman Bahri
Google DeepMind
Lukas Heinrich
Technical University of Munich

Ann Lee
Carnegie Mellon University
Laurence Perreault-Levasseur
Université de Montréal / Mila

Panelists

Topic: Data-driven vs inductive bias-driven methods in machine learning and the physical sciences.

Anima Anandkumar
Caltech
Johannes Brandstetter
JKU Linz / NXAI
Naoya Takeishi
University of Tokyo / RIKEN

Call for external contributions

In this workshop, we aim to bring together physical scientists and machine learning researchers who work at the intersection of these fields by

applying machine learning to problems in the physical sciences -- physics, chemistry, astronomy, earth science, biophysics, and related sciences; and
using physical insights to understand and/or improve machine learning techniques, for instance building hybrid machine learning algorithms that leverage physical models with machine learning blocks to create interpretable and accurate predictive models.

To this end, we encourage external contributions, which will be presented during in-person poster sessions during the workshop. Selected contributions will be offered 15-minute contributed talks. We invite researchers to submit original work in the following areas or areas related to them:

ML for Physics: Innovative applications of machine learning to the physical sciences; Machine learning model interpretability for obtaining insights into physical systems; Automating/accelerating elements of the scientific process (experimental design, data collection, statistical analysis, etc.).
Physics in ML: Strategies for incorporating scientific knowledge or methods into machine learning models and algorithms; Applications of physical science methods and processes to understand, model, and improve machine learning models and algorithms.
Other areas: Any other area related to the subject of the workshop, including but not limited to probabilistic methods that are relevant to physical systems, such as deep generative models, scientific foundation models, probabilistic programming, simulation-based inference, variational inference, causal inference, etc.

Submission tracks

Research abstract: We invite contributions on either completed or high-quality work-in-progress original research on the topics outlined above.
Datasets & Benchmarks abstract: We invite contributions describing a dataset and/or corresponding benchmarks at the intersection of ML and Physical Sciences, in particular showcasing the unique nature of physical datasets and forward models in the context of ML applications. See submission and review guidelines for specific instructions relating to this track.
Perspectives abstract: This year, we introduce a new Perspectives track, where researchers have the opportunity to present compelling and grounded viewpoints on recent directions and open questions at the intersection of ML and Physical Sciences. This track encourages disseminating perspectives on past, present, or future challenges of interest to scientists working at the intersection between ML and Physical Sciences. The track aims to stimulate productive and respectful conversations on timely topics that will benefit from the ML4PS workshop's attendees' input. Position papers should meet standard scientific rigor, including using evidence and reasoning to support claims, including relevant background and context, and attributing others' work via appropriate citations. Accepted Perspectives will be presented at the workshop during the poster sessions.

Submission instructions

Submissions should be anonymized short papers (extended abstracts) up to 4 pages excluding references. We invite authors to follow the guidelines and best practices from the NeurIPS conference (see also the main conference Datasets & Benchmarks call for guidelines pertaining to the corresponding track). Please ensure that your paper is approachable by someone not an expert in your specific area of physical science. For example, please avoid or at least define jargon. We reserve the right to desk reject any submissions that do not conform to the format. The review process is double-blind (optionally, single-blind for the Datasets & Benchmarks track). All authors must be registered as authors at the time of submission. We will not allow authors to be added after the review process has begun. This workshop is not archival so we will consider papers containing content that is published in an archival venue other than the main NeurIPS conference (e.g. a physics journal). However, such papers will likely need to be rewritten to fit the format and venue.

Submit your work on the submission portal. See submission and review guidelines for important instructions on preparing contributions as well as details on how they will be evaluated.

Submit contributions Submission guidelines

Code sharing & reproducibility guidance: While we do not enforce it, we will highlight submissions containing documented code and reproducible workflows through a Reproducibility Badge.

Double-submission policy: While we primarily encourage the submission of original pieces of work, we also accept submissions that are extended abstract versions of already published work if their topic fits particularly well with the workshop's scope. In contrast, with the objective of respecting the hard work of reviewers and giving equal chances to all submissions, we strictly prohibit submitting to multiple workshops simultaneously. Submissions flagged as coincidentally submitted to multiple NeurIPS workshops will be desk rejected.

Call for reviewers

We need you! Especially if you are considering submitting to the workshop, please also consider helping us with reviews. You can opt-in specifying the maximum number of papers you would like to review:

Volunteer to review