CS & Physics Meet-Up 2.0 by Lamarr & B3D

5 Dec 2024, 10:00 → 6 Dec 2024, 14:30 Europe/Berlin

Hochschule Bonn-Rhein-Sieg / St. Augustin

Jessica Koch (MPIfR), Jens Buß (Lamarr Institute, TU Dortmund University), Ekaterina Moerova (B3D, MPIfR), Kevin Schmidt

The "CS & Physics Meet-Up by Lamarr & B3D" aims to bring together researchers from the fields of computer science, with a focus on machine learning and artificial intelligence, and physics, specifically particle physics, astroparticle physics, and radio astronomy.

The goal of the meet-up is to provide a mutual overview of each other's research topics and research questions, as well as learning tasks, methods, and data. Our goal is to identify commonalities and develop ideas for future collaborations.

The event will include presentations, poster sessions, and open discussions to develop project ideas for future collaborations between computer scientists and physicists of Lamarr and B3D.

Thursday 5 December

10:00 → 11:00 Arrival

11:00 → 11:10 Organisatorical: Welcome

11:10 → 12:00 Introductions

11:10 Introduction Institute for visual computing

Speaker: Prof. Andre Hinkenjann

11:40 Introduction to the Interdisciplinary Research Area Physics and her Problem Domains

Speaker: Prof. Wolfgang Rhode (TU Dortmund)

12:00 → 13:00 Introductions: Short presentations on discussion groups topics

13:00 → 14:00 Lunch

14:00 → 15:00 Discussion Groups: "A deep learning based radio astronomy data pipeline" (André Hinkenjann); "Scaling Deep Learning for Pulsar Detection in Real-Time Radio Astronomy" (Andrei Kazantsev); "Learning from Time Series for Physics Applications" (Amal Saadallah)

14:00 A deep learning based radio astronomy data pipeline

We would like to discuss neural representation(s) of radio data and algorithms working on this representations, like cleaning, filtering, source finding or visualisation of data cubes e.g. Neural representations could be one solution to bridge the gap between high quality but very slow image generation in the data center and the interactive or even real-time visual analysis of data by humans at their workplace.

Speaker: Prof. André Hinkenjann (Hochschule Bonn-Rhein-Sieg, Institute of Visual Computing)

14:00 Learning from Time Series for Physics Applications

Discussion of the application of time series analysis across diverse physics domains. Participants will explore use cases from areas such as astrophysics, climate science, particle physics, and more, where time series data is critical for capturing temporal dynamics.

Speaker: Amal Saadallah (Lamarr Institute-TU Dortmund)

14:00 Scaling Deep Learning for Pulsar Detection in Real-Time Radio Astronomy

As part of the PUNCH4DFDI project at the Max Planck Institute for Radio Astronomy, a deep learning-based pipeline is being developed for the automatic classification of astronomical radio signals in real-time. A prototype utilizing deep learning techniques has been created to classify emissions from the pulsar in the Crab Nebula. The next step involves expanding the model's capabilities to successfully detect pulses from different pulsars, with other dispersion measures and in other frequency ranges. During the discussion group, a live discussion is planned to explore possible approaches for implementing this scaling.

Speaker: Andrei Kazantsev (Max-Planck-Institut für Radioastronomie)

15:00 → 15:30 Coffee Break + Poster Session

15:30 → 16:45 Tour through the Institute for Virtual Computing

16:45 → 17:00 Organisatorical: Summary and outlook for day 2

19:00 → 21:00 Social Dinner

Details | Directions via GoogleMaps

Friday 6 December

09:45 → 10:00 Organisatorical: Morning Welcome

10:00 → 11:00 Discussion Groups: "Python & Massive Data Processing/Heat" (Claudia Comito); "Long-term Analyses in Astroparticle Physics - Prospects of AI application" (Cyrus Walther); "Efficient Communication between AI and domain experts - Part I" (Ramsés Sanchéz & Jens Buß)

10:00 Efficient Communication between AI and domain experts - Part I

We aim to set up a training in Lamarr to help domain experts (e.g. physicists) and ML experts (e.g. computer scientists) to discuss common research more efficiently in order to identify the machine learning task at hand and set up a fruitful collaboration. In this discussion group we would like to use you as guinea pigs for a first test of our approach, collect your feedback and ideas for future trainings. We will outline some Machine Learning basics that should help domain experts to formulate their machine learning task. Then we dive into one example application and discuss how to present it more efficiently to get to its ML core.

Speakers: Dr Jens Buß (Lamarr Institute, TU Dortmund University), Ramsés Sanchéz

EfficientCommunication.pdf

WhiteBoard.png

10:00 Long-term Analyses in Astroparticle Physics - Prospects of AI application

The time scale of data-taking established large-scale experiments in astroparticle physics can provide allows, by now, for long-term analyses performed on the respective data. Besides known analysis approaches based on statistical methods, we want to explore and discuss openly the prospects and ideas of AI application to these long-term datasets and the new insights possibly to be gained on these physical processes.

Speaker: Cyrus Walther (TU Dortmund University)

Discussion_Round_Pitch.pdf

10:00 Python & Massive Data Processing/Heat

Processing massive scientific datasets is challenging. Avoiding memory bottlenecks without having to rewrite existing software often involves breaking up and analyzing data in smaller chunks, a process both inefficient and unsuitable to exploit the scientific value of the data. To address this problem, at FZJ we co-develop the Python library Heat [https://github.com/helmholtz-analytics/heat]. Heat can be used as a backend to your NumPy/SciPy-based code, to intuitively distribute massive memory-intensive operations to multi-CPU, multi-GPU clusters. In this discussion, I can briefly show an example of Heat usage, but most of all I'm curious about your compute / memory bottlenecks, and about potential collaborations. Feel free to bring your own code!

Speaker: Dr Claudia Comito (Forschungszentrum Jülich)

11:00 → 11:30 Coffee Break + Poster Session

11:30 → 12:30 Discussion Groups: "Deep learning perspectives in radio interferometry" (Kevin Schmitz); "Efficient Communication between AI and domain experts - Part II" (Ramsés Sanchéz & Jens Buß)

11:30 Deep learning perspectives in radio interferometry

Speaker: Kevin Schmidt

11:30 Efficient Communication between AI and domain experts - Part II

We aim to set up a training in Lamarr to help domain experts (e.g. physicists) and ML experts (e.g. computer scientists) to discuss common research more efficiently in order to identify the machine learning task at hand and set up a fruitful collaboration. In this discussion group we would like to use you as guinea pigs for a first test of our approach, collect your feedback and ideas for future trainings. We will outline some Machine Learning basics that should help domain experts to formulate their machine learning task. Then we dive into one example application and discuss how to present it more efficiently to get to its ML core.

Speakers: Ramsés Sanchéz, Dr Jens Buß (Lamarr Institute, TU Dortmund University)

12:30 → 13:00 Organisatorical: Presentation of the "results" from the discussion groups

13:00 → 13:30 Summary & farewell

Lamarr & B3D Mattermost Channel

MailingLists and Mattermost.pdf

13:30 → 14:30 Lunch (Optional)