2025 Fink Collaboration Meeting

16 juil. 2025, 09:00 → 18 juil. 2025, 17:00 Europe/Paris

Hotel Littéraire Alexandre Vialatte

Anais Moller (Swinburne University), Emille Ishida (CNRS/LPC-Clermont), Julien Peloton (CNRS-IJCLab), Marine HEBERT (LPC)

2025 Fink Collaboration meeting

This meeting will be an opportunity for the Fink collaboration to meet in person, go through hands-on tutorials and explore the scientific potential of the broker, both with ZTF alert data and in preparation for what we intend to do with LSST. 

The program will host a number of contributed talks, several tutorials on Fink capabilities and plenty of time for collaborative work and discussions about our plans for the future.

Note that remote participation is not envisaged.

The meeting will happen in  Clermont Ferrand, between 16 - 18 July 2025.

Registrations are open until 10 June 2025.

Our venue holds a maximum of 25 participants, thus preference will be given for students and those currently working on Fink development. 


Schedule: 
The meeting will start on Wednesday, 16 July at 9:00h and a full program will end on Friday, 18 July, 12:00h.

The management team will be available for discussions with the participants on Friday afternoon until 17h, but there will be no official program.

Conference Dinner

More information will be advertised soon.

 

mer. 16 juillet

1 Introduction: Fink from 2019 to 2025

Orateur: Dr Emille Ishida (CNRS/LPC-Clermont)

Fink_Intro

2 Roundtable

Introduction from participants and their goal in this workshop

10:30 break

Technology: Infrastructure

Graphs

3 Overview of the infrastructure and usage report

Orateur: Dr Julien Peloton (CNRS-IJCLab)

Overview of the infrastructure and usage report.pdf

4 Enhancing Real-Time Astronomical Alert Processing with Scalable Kubernetes-Driven Infrastructure and Observability

Fink is a real-time astronomical alert broker designed to process high-throughput data streams from surveys such as ZTF and the forthcoming LSST. To meet the demands of low-latency, high-reliability data classification at scale, we have developed a production-grade Kubernetes-based deployment architecture that integrates continuous delivery (CI/CD), fine-grained observability, and resource-aware scheduling.

This contribution presents recent work on operationalizing Fink's Spark-based alert pipeline using Kubernetes-native tooling, focusing on standardizing performance monitoring and pipeline introspection. We leverage Prometheus and JMX exporters for service-level metrics, integrate sparkmeasure to capture stage-level execution details, and develop automated profiling strategies to detect bottlenecks and regressions. Our goal is to elevate Spark observability to first-class status within real-time scientific data infrastructures.

We also report on collaborations with CERN (Luca Canali), Kubeflow, and Stackable.tech to prototype next-generation workflows for distributed data science, bridging operational excellence with scientific reproducibility. These efforts aim to position Fink as a model for scalable, maintainable, and transparent alert processing in the upcoming data-intensive era of time-domain astronomy.

Orateur: Fabrice Jammes (CNRS)

Fink-k8s-july2025.pdf

12:30 Lunch

Technology: Science Portal

Graphs

5 Science portal session

New functionalities and discussion on evolution (aka what worked and did not work for ZTF, and what is missing for Rubin)

Orateur: Dr Julien Peloton (CNRS-IJCLab)

15:00 Break

Technology: Preparation to LSST

Graphs

6 LSST simulations: what to expect for alerts and data

Orateur: Anais Moller (Swinburne University)

Sims_notebook

7 Manipulating Rubin 1st alerts

Orateur: Dr Julien Peloton (CNRS-IJCLab)

or5p2

jeu. 17 juillet

Science: Extra-galactic

8 Early identification of optical tidal disruption events with the Fink broker

Detecting Tidal Disruption Events (TDEs) candidates early is essential for follow-up observations at peak brightness, enabling confirmation of their nature and a deeper understanding of their complex multi-wavelength behavior. TDEs are rare events, and their detection is one of the key goals of large ground optical facilities, such as the Zwicky Transient Facility (ZTF) and the upcoming Vera C. Rubin Observatory. However, managing their vast alert streams requires automated pipelines that identify the nature of the detected events. We introduce a module developed within the Fink alert broker, designed to enable the early identification of TDEs observed by ZTF.
A first step consists in the automatic selection of light curves compatible with a rising transient scenario. It is followed by a tailored feature extraction based on a multi-band fit of the rising part of the light curves. It enables the computation of physically-motivated features, such as temperature and rise time, which are essential to distinguish TDEs from other long-lasting transients. Finally, a machine learning classifiers trained on a sample of high quality TDEs to distinguish similar events, with and without additional features related to the distance to host galaxy. Despite the challenge offered by the sparse and highly imbalanced training dataset, the classifiers provides reasonably completeness and purity metrics.

The talk describes the training of the classifier on archival data, implementation of the module inside Fink broker that reports the candidates in nearly real time, and presents several promising candidates identified during the development, as well as the perspectives of adaptation of this approach for the upcoming Rubin Observatory' Legacy Survey of Space and Time data stream.

Orateur: Sergey Karpov (Institute of Physics, Czech Academy of Sciences)

early_tdes.pdf

9 Hostless transients detection in Fink

Hostless transients refer to transients for which we cannot find an associated host galaxy. This could happen for the following possible reasons: 1. Mostly due to the limitations of telescopes, we are unable to detect the associated host; 2. In very rare cases, supernovae have escaped their hosts due to dynamical ejection, tidal stripping or other astrophysical phenomena. Such candidates are interesting for many reasons such as: to understand why some transients occur in very faint galaxies or to find new low surface brightness galaxies. We developed and integrated a statistical pipeline for ZTF alerts in Fink broker that detects potential hostless transients. By applying image processing and frequency domain analysis to ZTF science and template alert stamps, our pipeline (ELEPHANT) is discovering interesting potential hostless transients. ∼19% of the reported transients on our Fink Telegram bot from April 2024 to March 2025 have spectroscopic classification on the Transient Name Server portal.

Moving forward, we aim to update and integrate the methodology for the Vera C. Rubin Observatory data in Fink broker. Since the LSST can produce up to 10 million alerts each night, it would be beneficial to have a real-time pipeline that can flag potential hostless transients. Our pipeline is independent of other catalogues, operates solely on stamp images and can process alerts at a faster rate. Therefore, it would be an ideal tool for the LSST to create a potential hostless transients catalogue.

Orateur: M. Rupesh Durgesh (COIN)

Hostless transient detection.pdf

10 Discussion: Extra-galactic science

10:30 Break

Science: Galactic

11 Finding magnetic cataclysmic variable stars with Fink

Magnetic cataclysmic variable stars are rare close binary systems in which matter from a main sequence star accretes on a strongly magnetized white dwarf. This presentation will introduce the characteristics of these objects and show the development of a module within Fink designed to find magnetic cataclysmic variable stars in the Zwicky Transient Facility alert stream, in collaboration with the Fink Cataclysmic Variable team from Brazil. The module is based on a nearest neighbors algorithm using lightcurve statistical features which characterize luminosity variability of celestial bodies. Accuracy tests and first results show promising performances for future detections. Difficulties encountered in real-time implementation, as well as changes and thoughts on future LSST implementation will also be discussed.

Orateur: Clement MUR

Fink Collaboration Meeting_compressed.pdf

12 Search for hidden baryons through interstellar scintillation

Cool molecular hydrogen H2 may be the ultimate possible constituent to the Milky-Way baryonic hidden matter. I will describe a new way to search for such transparent matter in the Galactic discs and halo, through its diffractive and refractive effects on the light of background stars. The light of a background star can be subject to stochastic fluctuations on the order of a few percent at a characteristic time scale of a few minutes. Results from simulations and from a feasibility test performed with the ESO-NTT telescope will be presented.
I will show that a mini-survey of a few hours with the telescope of the Vera Rubin Observatory, consisting in filming a single field (in the LMC or the SMC) at the rate of a few images per minute has the unique potential to discover (or exclude) interstellar scintillation due to transparent turbulent clouds in the halo.
Complementary observations triggered by the fink broker will then enable us to monitor the sources that undergo scintillation over a long period of time and determine their size, distance and spectrum.

Orateur: marc moniez (LAL-IN2P3)

FINK-workshop-Moniez.pptx

13 Discussion: Galactic science

12:30 Lunch

Science: First alerts

14 What will you do and what do you need for 1st alerts?

What will you do and what do you need for 1st alerts?

Technology: Classification Graph

Président de session: Julius HRIVNAC

Graphs

15:00 Break

Science: Classification

15 Brainstorming on classification

Defining what classification means in Fink, and its impact

16 Sightseeing & conference dinner

ven. 18 juillet

Science: Solar System

17 Rotational properties of asteroids from large sky surveys

Orateur: Konstantinos Xenos (OCA)

presentation.pdf

18 Discussion

Technology: LSST preparation

Graphs

19 Hands on: From ZTF to LSST

Hands-on examples!

Adapting your current science module and filter to LSST:
- change in schema
- difference of cutouts
- retraining ML models
- scaling for LSST

Orateurs: Dr Julien Peloton (CNRS-IJCLab), Anais Moller (Swinburne University), Dr Emille Ishida (CNRS/LPC-Clermont)

10:30 Break

Technology: Visualisation

Graphs

20 Contest: Science portal visualisation

How to visualise LSST light-curves efficiently?

Orateur: Dr Anais Möller (Swinburne University)

Notebook on visualisation

21 Wrap-up

WarpUp