Summary

This workshop is the first edition of a recurring workshop aimed at bringing together the community of users of the CIANNA (Convolutional Interactive Artificial Neural Networks by/for Astrophysicists) software tool. This workshop will provide an opportunity to review recent scientific use cases, present the latest additions, identify development priorities, and formulate a strategy for structuring the community. The workshop will also include demonstration sessions with support for new users in learning how to use the tool.

Scientific justification

CIANNA is a general-purpose development framework for building, training, and deploying artificial neural network models with a wide variety of architectures. This software is primarily designed to meet the needs and constraints of the astrophysics community, whether in terms of the nature of the data (high dynamics, low signal-to-noise ratio, superimposition of sources), the output format of the models (probability, uncertainty), or scientists’ expectations of the models produced (reproducibility, explainability). The framework is also characterized by very low latencies, enabling it to produce lightweight models, particularly used in astrophysics, with numerical efficiency superior to equivalent consumer tools. It thus positions itself as a response to the methodological and technical challenges associated with the explosion in the volume of astronomical data from new large observational infrastructures (e.g., SKA).

CIANNA was used by the MINERVA team in the SKAO Science Data Challenges to build models for detecting and characterizing objects (Cornu et al. 2024). The team’s victory in the second edition of these challenges brought them international visibility, particularly within the radio astronomy community (Hartley et al. 2023) . Today, this software is used by various members of the COSGAL team at LUX at the Paris Observatory, both for detecting galaxies in data from the LoFAR and NenuFAR telescopes (a thesis will begin in September) and for studying the period of reionization of the universe (ANR-TSIA application pending). It is also used at the Strasbourg Astronomical Observatory to detect galaxies in data from the MeerKAT telescope (another thesis will begin in September). The software has been distributed to the Bordeaux Astrophysics Laboratory, where it is used to identify molecular species in the interstellar medium (a thesis in progress, Kessler et al. 2025). The software is also featured in another ANR-TSIA application submitted by CRAL and IRFU-AIM for the detection of galaxies with the MUSE (VLT) instrument. The development of CIANNA in an academic setting and its distribution under a free license also reflects a desire for independence from the predominance of similar tools distributed by industrial players.

Through the organization of this workshop, we hope to begin structuring the community around CIANNA on a French scale, enable new users (particularly doctoral and postdoctoral students) to get started with the tool, and initiate a collective reflection on future priority developments.

Preliminary scientific program

Day 1 – Scientific presentations and feedback from the community

– David Cornu: Presentation of CIANNA and its overall functionalities.
– David Cornu: Automatic detection and characterization of sources using the YOLO-CIANNA method, application to SKAO SDCs.
– Adrien Anthore (guest, doctoral student): Automatic detection of galaxies in SKA precursor surveys: in continuum with LOTSS (LOFAR) and in HI with LADUMA (MeerKAT).
– Nina Kessler (guest, postdoctoral researcher): Identification of complex organic molecular species in star-forming regions.
– Grégory Sainton: Presentation of an “on-the-fly” detection service for integration into astronomical data visualization tools.
– Aristide Doussot: Overview of CIANNA porting to different installation and deployment methods (containers, guix, apt, manual compilation, etc.).
– Round table: Community structuring, feedback on the use of CIANNA, discussion of missing and under development features.

Day 2 – Demonstrations and practical workshops

– CIANNA installation guide and installation assistance on participants’ machines (Aristide Doussot et al.)
– Tutorial 1: Classification of young stars using Spitzer infrared bands
– Tutorial 2: Morphological classification of galaxies using images (Galaxy Zoo)
– Tutorial 3 and demo 1: Object detection using the YOLO-CIANNA method and illustration using various datasets
– Demo 2: Generative diffusion models with deep residual networks, demonstration of the capabilities of the experimental version of CIANNA.
– Round table: Feedback on difficulties encountered during tutorials and demonstrations, discussion on code documentation, identification of channels to facilitate communication between the development team and the community.