Connectionists: ECML-PKDD 2019 Discovery Challenge - Correcting Transiting Exoplanet Light Curves for Stellar Spots

[Nikolaou, Nikos]

[Apologies for multiple postings]

Call for Participation

ECML-PKDD 2019 Discovery Challenge - Correcting Transiting Exoplanet Light Curves for Stellar Spots (Ariel Machine Learning & Stellar Activity Challenge)

Organised and Sponsored by the European Space Agency (ESA) M4 ARIEL Mission Consortium and the University College London
Hosted by the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML-PKDD) 2019

*** Motivation & Goals***

The field of exoplanet discovery and characterisation has been growing rapidly in the last decade. However, several big challenges remain, many of which could be addressed using machine learning and data mining methodology. For instance, the most successful method for detecting exoplanets, transit photometry –measuring the faint decrease in incoming stellar light as an exoplanet passes between the Earth and a target star– is very sensitive to the presence of stellar spots, i.e. areas of the star that are colder and emit fewer light. The current approach is to identify the effects of such spots visually and correct for them manually or discard the data. As a first step to automate this process, we propose a regular competition on data generated by a simulator of the European Space Agency’s upcoming Ariel mission, whose objective is to characterise the atmosphere of 1000 exoplanets. The data consist of light curves (i.e. time series of the light received by the observation instrument from the target star-planet system) corrupted by stellar spots, along with auxiliary observation information. The goal is to correct the light curves for the presence of stellar spots, by predicting the relative planet-to-star radius ratio. This is a yet unsolved problem in the community. Solving it will mean improving our understanding of the characteristics of currently confirmed exoplanets, potentially recognising false positive / false negative detections and improving our ability to analyse new observations – primarily but not limited to those expected from Ariel– without the need to equip new telescopes with additional instruments with all the extra costs this implies.

*** Dataset & Task Details***

Task: Supervised learning, multi-target regression; all variables are continuous.
Features: Each training datapoint consists of a set of 55 noisy light curves (one per wavelength, each being a timeseries of 300 timesteps) and a set of 6 additional stellar and planetary parameters.
Targets: The goal is to predict a set of 55 values (relative radii, one per wavelength) for any datapoint.
Dataset size: The size of the dataset is ~20Gb.

For a more detailed description, please visit the competition website at:

https://ariel-datachallenge.azurewebsites.net/ML

*** Participation ***

Participants of this challenge will submit the predictions of their models on the provided test dataset. The ground truth will be released to the participants after the end of the competition.
The solutions will be automatically ranked and the participants will obtain immediate feedback in the leaderboard maintained in the site. The 2 top-ranked participants will be invited to provide a brief description of their solution (describing data preprocessing steps, models and algorithms used) in the week after the competition closes. The 2 top-ranked participants will then be eligible to prizes, provided they beat the baseline, and no plagiarism or test set leakage has occured.

To participate, please visit the competition website at:

https://ariel-datachallenge.azurewebsites.net/ML

and follow the instructions to learn details about the problem, data, submission format, evaluation protocol and the baseline solution.

*** Important dates *** (All times are in AoE time)

* April 15th, 2019: Beginning of the competition, release of training dataset and test dataset (w/o ground truth). Participants can start submitting model predictions on test set and obtaining immediate feedback in the leaderboard.
* Aug 15th, 2019: Competition closes, release of ground truth. Top-ranked participants are invited to submit a brief description of their solution.
* Aug 22nd, 2019: Deadline for submitting description of solution. Organisers start checking solutions for plagiarism and test set leakage.
* Aug 25th, 2019: Announcement of the competition winners. (Tentative, subject to successful checks and collaboration from participants)
* Sept 16th - 20th, 2019: ECML-PKDD 2019.

*** Prizes ***

The 2 top-ranked participants (provided they beat the baseline) will be awarded with a free registration to the ECML-PKDD 2019, to be held in Würzburg, Germany, from September 16 - 20, 2019.

*** Dissemination of results ***

The 5 top-ranked participants (provided they beat the baseline) will be invited to present their solutions at ECML-PKDD 2019, to be held in Würzburg, Germany, from September 16 - 20, 2019. The authors of solutions that are of interest to the organisers will be invited to participate to larger scale collaborations in the context of the Ariel mission and beyond.

*** Organizing team  ***

Nikolaos Nikolaou - UCL, England - n.nikolaou at ucl.ac.uk<mailto:n.nikolaou at ucl.ac.uk> - (Main organizer)
Ingo P. Waldmann - UCL, England
Subhajit Sarkar - University of Cardiff, Wales
Angelos Tsiaras - UCL, England
Billy Edwards - UCL, England
Mario Morvan - UCL, England
Kai Hou Yip - UCL, England
Giovanna Tinetti - UCL, England

*** Contact Email  ***

For technical issues or questions, please contact us at:

exoai.ucl at gmail.com<mailto:exoai.ucl at gmail.com>

Nikolaos Nikolaou
Postdoctoral Researcher
Department of Physics & Astronomy, UCL
https://nnikolaou.github.io<https://nnikolaou.github.io/>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20190417/2d9079f0/attachment.html>