Motivation

The specific goal of the workshop is to bring together AI developments with domain experts and kick-start the implementation of ML solutions to specific computationally expensive astrophysical problems.

The Carl-Zeiss-Stiftung (CZS) summer school on scientific machine learning for Astrophysics aims to provide an informal, inclusive and leading-edge venue for research and discussions at the interface of machine learning (ML) and astrophysical science. This interface spans

1. Applications of ML in astrophysics (ML4astro)
2. Developments in ML motivated by physical insights (physics-informed ML)
3. Convergence of ML and astrophysical modeling.

Scientific usage of ML techniques requires explainability and testability, and an understanding of the model’s operations to ensure the scientific value of the outcome. Thus, the tremendous uptake of ML in astrophysics is questioning what scientific understanding means in the age of complex-AI powered science, and demands to re-evaluate the roles of machines and human scientists in developing scientific understanding in the future.

In this context, recent computer science research such as explainable Artificial Intelligence or physics-informed deep learning have emerged. These approaches present promising avenues to incorporate ML concepts into the scientific reasoning process to accelerate research cycles and help us to confront our theoretical models of the Universe with the stringent constraints obtained from various observational datasets such as galaxy images, stellar spectra or gravitational wave signals.

The scientific need for transparent, testable and explainable algorithms demands the design of new architectures and approaches. Furthermore, the more research is done by complex and highly parameterized ML models the more we have to ask ourselves what scientific understanding actually means. However, the rapid recent developments, with multiple new programming languages and libraries and large number of available techniques in particular in astronomy make adopting these promising methods a daunting challenge for the scientific community.

The purpose of this workshop is to bring together physicists and mathematicians with researchers from the field of ML and computer science to discuss and tackle the unique challenges posed by research problems at the intersection of astrophysics and ML. Many of these problems require the development of novel methods around cutting-edge research challenges which can only be addressed by a joint effort of both communities.

By bringing together an interdisciplinary group of researchers we will enable stimulating discussions, foster new scientific networks to trigger highly interactive work on state-of-the-art research topics with cutting-edge methods and guide the scientific community towards sustainable usage of ML methods.

Thus, we envision a highly collaborative environment for a 5 day workshop in the form of a scientific Hackathon whose outcome will certainly contribute to open-source community software. The morning sessions will feature invited talks from leading individuals in both communities that cover state-of-the-art techniques followed by a panel discussion of pressing open questions that will set the stage for this workshop. The entire afternoon is reserved for interactive and collaborative parallel work sessions in the form of group discussions and tutorials on open problems in scientific ML plus additional time for dedicated collaborative development work on the specific topics mentioned above. The specific aim of the interactive parallel discussion sessions is to facilitate knowledge sharing across disciplines, backgrounds, and career stages and incite new collaborations and formulate strategies for a sustainable adoption of innovative ML methods in science. Thus, at the time of application for this workshop we ask each applicant to propose topics, tutorials and small collaborative research projects for the afternoon sessions.