BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.adass.org//adass2026//talk//UYVFZS
BEGIN:VTIMEZONE
TZID:AWST
BEGIN:STANDARD
DTSTART:20000101T000000
RRULE:FREQ=YEARLY;BYMONTH=1;UNTIL=20051231T160000Z
TZNAME:AWST
TZOFFSETFROM:+0800
TZOFFSETTO:+0800
END:STANDARD
BEGIN:STANDARD
DTSTART:20070325T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:AWST
TZOFFSETFROM:+0900
TZOFFSETTO:+0800
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20071028T030000
RRULE:FREQ=YEARLY;BYDAY=4SU;BYMONTH=10;UNTIL=20081025T190000Z
TZNAME:AWDT
TZOFFSETFROM:+0800
TZOFFSETTO:+0900
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-adass2026-UYVFZS@pretalx.adass.org
DTSTART;TZID=AWST:20261104T164500
DTEND;TZID=AWST:20261104T170000
DESCRIPTION:Most astronomical machine learning workflows begin only after a
 stronomy specific processing has already taken place: selecting archive pr
 oducts\, interpreting FITS metadata\, transforming coordinates\, matching 
 catalogues\, resampling exposures\, and combining observations across filt
 ers. These operations usually sit outside the model\, making them difficul
 t to optimize jointly\, test within the same computation\, or execute effi
 ciently on GPUs.\n\nWe present torchfits and torchsky\, a tensor native so
 ftware stack that connects FITS data with differentiable models of the obs
 erved sky. torchfits provides selective access to local and remote FITS im
 ages and tables\, exposing tensors and columnar batches for analysis and t
 raining. torchsky provides celestial geometry\, catalogue association\, ma
 pmaking\, spectral response\, PSF convolution\, and other observation oper
 ators on a common PyTorch runtime\, with execution on CPUs and GPUs.\n\nTo
 gether\, these tools support multispectral representations of the sky as a
  function of position and wavelength\, fitted to heterogeneous observation
 s rather than restricted to precomputed coadds or catalogues. The same com
 ponents also support conventional survey processing and large-scale cross-
 matching.\n\nWe will describe the software architecture\, numerical valida
 tion\, performance\, and initial applications to multispectral mapmaking a
 nd survey-scale catalogue association. We will also show how torchsky’s 
 geometric primitives are being reused in a new cross-matching system. More
  broadly\, placing data access\, celestial geometry\, and observation oper
 ators on a common tensor runtime opens new possibilities for joint inferen
 ce across surveys and wavelengths\, differentiable calibration and mapmaki
 ng\, and foundation models that learn from astronomical observations rathe
 r than only from preprocessed products.
DTSTAMP:20261001T111330Z
LOCATION:Banquet Hall
SUMMARY:From FITS archives to differentiable multispectral sky models - Sé
 bastien Fabbro
URL:https://pretalx.adass.org/adass2026/talk/UYVFZS/
END:VEVENT
END:VCALENDAR