A collaborative workshop focused on developing a benchmark dataset for machine learning classification of astronomical transients
Our workshop is structured around seven key objectives that will guide our collaborative efforts
Establish a comprehensive and scientifically rigorous taxonomy for astronomical transient classification that serves as the foundation for benchmark development.
Compile and validate a diverse dataset of well-characterized transients with real (i.e., messy and heterogeneous) multi-band light curves and spectroscopic classifications that is made freely available to the community.
Develop and agree upon standardized metrics and evaluation protocols for comparing machine learning classification methods.
Document the main challenges in transient classification and establish research priorities for the community to address.
Create sustainable infrastructure for dataset hosting, model evaluation, and community engagement using standard tools in the machine learning community.
Establish working groups and communication channels to continue development and maintenance of the benchmark dataset.
Set the standard for the first TDAbench release and set the foundation for future releases that include LSST and Roman data..
The TDAbench dataset will serve as the definitive benchmark for evaluating machine learning approaches to astronomical transient classification. Our goal is to create a dataset that is:
The dataset will include photometric light curves, host galaxy properties, contextual information, and where available, spectroscopic classifications from major transient surveys. The recently complete Bright Transient Survey (BTS) conducted by the Zwicky Transient Facility (ZTF), which has spectroscopically classified more than 10k transients will serve as the preliminary basis for the benchmark dataset. Relevant observations from other surveys will be added during the course of the workshop.
Key deadlines and milestones for TDAbench 2026
Online registration portal opens for all participants. Early registration is encouraged due to limited capacity.
There will be a very limited number of slots for (brief) contributed talks on topics relevant to building the benchmark dataset. There will be multiple (extended) poster sessions where participants can highlight their ongoing work on transient classification. All submissions will be reviewed by the scientific organizing committee.
Authors will be notified of abstract acceptance and presentation format (oral or poster).
Last day for early bird registration rates. Hotel block reservations recommended by this date.
Final deadline for workshop registration. Late registrations may be accepted on a space-available basis.
Four days of collaborative sessions, presentations, and working group meetings at SkAI Institute, Chicago.