The third edition of the AstroInformatics thematic school will take place from June 26th to 30th, followed by the AstroInfo AISSAI Hackathon from July 3rd to 7th.

• To make scientists and engineers aware of new technologies in data processing.
• To train scientists and engineers to the computing methods of mass data processing in Astrophysics.
• To reinforce this transverse action of collaboration between computing and astrophysics.
• Demonstrate the possibilities of using these new methods in the case of astrophysical data through practical cases.

The program is available here.

Monday: Introduction to Astro Data, André Schaaff [Centre de Données astronomiques de Strasbourg]

We'll start with an overview of astronomical data. We'll continue with the example of the Strasbourg Astronomical Data Center (50 years old in 2022!), which provides the community services (VizieR, Simbad, Aladin, X-Match) to access, visualize, and manipulate the data it hosts. We'll also take a look at the Virtual Observatory, with illustrations of the interoperability it enables through its standards and protocols.

VizieR, the service that hosts over 20,000 catalogs, will allow us to extract a sample of the data that will be used throughout the rest of the week.

The hands-on sessions will also provide an overview and familiarization with the tools (e.g. visualization tools) that will be used during the week.

Finally, this is an ideal opportunity to check that participants' hardware configurations are operational and to resolve any problems so that everyone can participate in the training under the best possible conditions.

Tuesday: Machine Learning, Valérie Gautard [Commissariat à l'Energie Atomique]

We begin with a brief introduction to machine learning and then cover key concepts such as data preprocessing, supervised and unsupervised learning, and evaluation measures. This overview serves as a foundation for a more in-depth exploration of the field of machine learning.

Mercredi: Introduction au MLOps, Alexandre Boucaud [Laboratoire AstroParticule et Cosmologie]

MLOps stands for ML Operations, which inherited from the DevOps trend, i.e. the ensemble of operations to put algorithms (science) into production (engineering). The goal of this course is to get familiar with a series of tools, principles and good practices that will make your life as a scientific developer way easier, and make you gain confidence in your every day workflow. 
In particular, we will focus on the specific tools designed for machine learning development such as [MLFlow](https://mlflow.org/) or [Hydra](https://hydra.cc/), that will allow you to log and reproduce ML experiments, attach some results (metrics, plots), register ML models, etc. This is very convenient and can be considered as the modern lab notes (cahier de laboratoire).
We will go through practical exemples on get started with these tools and we will try to use them as much as possible during the exercices of the following days.

Jeudi: Deep Learning, Françoise Bouvet [Laboratoire de physique des 2 infinis]

We will describe the main concepts of Deep Learning (DL). The course will focus on Multilayer Neural
Network (MLP) and Convolution Neural Network (CNN).
In particular, we will talk about :

• artificial neurons,
• MLP : structure and how it works,
• CNN : structure and how it works,
• a brief review of other NN structures.

Python and the Keras library will be used for the exercises.

Vendredi: Normalizing flows, Justine Zeghal [Laboratoire AstroParticule et Cosmologie]

Normalizing Flows (NF) is a powerful tool used for modeling complex distributions and used in Bayesian inference in astrophysics for density estimation or sampling.
The key idea of NFs is to transform a simple distribution (such as a Gaussian) into a more complex target distribution through a series of bijective transformations.
By applying these transformations, the data is mapped from a simple distribution to the complex distribution and vice-versa so we can easily generate samples from a complex distribution by using a sampling from the well-known distribution + the bijections, and preserve the probabilities.

In this course we will dive into the theory, applications, and practical implementation of NFs, which have become ubiquitous in Bayesian inference these days.
The ultimate goal of this course is to use Normalizing Flows in the context of a toy cosmological inference problem.

By the end of the program, participants will have a solid foundation in machine learning for their own projects.

The agenda is available here.

This school is open to all :

• doctoral students (this school can be included in the doctoral student training program)
• post-doctoral students
• researchers/teaching researchers (from all institutes)
• engineers from all institutes

Priority will be given to young PhD students, researchers and engineers.

The AstroInfo School, is intended for beginners or people with minimal knowledge of data processing in Machine Learning and Deep Learning.

Necessary prerequisites: Knowledge of Python and Numpy.

Registration is open and can be done on this registration page.  

The registration fees depend on the category of personnel:

• Agent (researcher, ITA, post-doc, PhD student, CDD) employed by the CNRS: registration financed by the CNRS
• PhD student not paid by the CNRS: 350 euros excl. tax
  
• Post-doctoral student not paid by the CNRS: 550 euros excl. tax
  
• Staff from another public institution: 550 euros excl. tax
  
• Staff from a private institution: 1500 euros excl. tax
  

Please note

From June 1st, the rates will be increased:

• PhD student not paid by the CNRS: 455 euros excl. tax
  
• Post-doctoral student not paid by the CNRS: 715 euros excl. tax
  
• Staff from another public institution: 715 euros excl. tax
  
• Staff from a private institution: 1950 euros excl. tax

Registration will close on Friday, June 9th.

The registration fee covers breaks, lunch and dinner, accommodation and participation in social events.
Travel expenses will be paid by your laboratory, from which you will have to request a mission order. 

In the continuity of the school, we are organizing the Astro AISSAI Hackathon. The hackathon is based on a scientific project and will take place throughout the week. This second week is intended for people with some experience, and the number of participants will be limited to 20.

If you are interested in participating, please also register on the Hackathon web site.

Requirements for participants:

• Minimum experience in MLDL or data processing
• Experience in software development

Call for project proposals

We invite senior participants to submit hack proposals in the field of astrophysics and machine learning. We welcome all types of projects, as long as they involve and engage several participants for the entire week on various tasks and lead to a publishable result.
The successful candidates will work with local organizers in advance to prepare hack material (project layout, open data, existing code, starting notebook, etc.).

Prior knowledge of machine learning is preferable but not mandatory, as a local team will be available to work with the participants.

Please submit your proposals on the registration form before May 1st, 23:59 AOE (Anywhere on Earth).

Attention

Registration for the hackathon and submission of topics must be done on the dedicated hackathon website: https://aissai-hackathon.astroinfo.in2p3.fr/.

ASTROINFO is a thematic school which is meant to be recurrent every two years. The pandemic did not allow us to organize this school in 2020. The previous schools took place in Marseille and Séolane and can be found on the site of ASTROINFO 2018 and ASTROINFO 2021