BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.adass.org//adass2026//talk//WWKV97
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-WWKV97@pretalx.adass.org
DTSTART;TZID=AWST:20261102T153600
DTEND;TZID=AWST:20261102T154800
DESCRIPTION:In astrophysics\, Deep neural networks have achieved state-of-t
 he-art performance in a variety of spectral analysis tasks. However\, thes
 e models are often considered "black boxes"\, providing limited insight in
 to predictions and requiring post-hoc tools for explanations. This lack of
  transparency can limit their adoption in scientific contexts\, as in thes
 e environments interpretability is essential. This is one of the challenge
 s in the development of RHEA\, the Rare-earth \\& Heavy Elements Analyzer 
 of the HARLOCK project within the PRORIS program. It combines gamma-ray sp
 ectroscopy\, multispectral imaging\, ultraviolet fluorescence screening an
 d X-ray fluorescence analysis to identify Rare-Earth Elements (REE) deposi
 ts and associated elemental and mineralogical signatures across different 
 spatial scales. \n\nIn this work\, we propose a model-agnostic framework t
 hat converts any Convolutional spectral classifier into an explainable pre
 dictor without hindering its performance on multiple metrics\, demonstrati
 ng it on a complete model working with attention masks for additional insi
 ght and training stability. Our approach exploits the Neural Network's lat
 ent representation and preservation of spatial features: after training\, 
 the final regression layer of the backbone model is removed and replaced b
 y a K-Nearest Neighbors (KNN) regressor operating directly on the learned 
 embedding space\; as a result\, every prediction is accompanied by an expl
 anation in the form of the most similar spectra observed during training. 
 This works particularly well thanks to a custom loss function designed to 
 optimize the quality of the learned embedding space\, without hindering pe
 rformances. \n\nFurthermore\, in our model\, for each predicted element an
  associated attention mask is learned at training time starting from a phy
 sics-based initialization. This way\, the framework provides\, alongside i
 nstance-level explanations\, easily readable heatmaps in the form of a mas
 k-by-weights product\, allowing users to trace model outputs back to repre
 sentative examples in the training data and the relevant parts of the inpu
 t used. Experimental results show that\, in a regression task estimating t
 he composition of soil samples\, the average R2 score over the predicted e
 lements stays around 0.9 for both the original approach and the explainabl
 e Deep kNN model (depending on the complexity of the task).\n\nOur proposa
 l offers a practical path toward trustworthy AI for spectroscopy\, combini
 ng the representational power of deep learning with the transparency of ex
 ample-based reasoning.
DTSTAMP:20261116T154905Z
LOCATION:Seminar Room 2
SUMMARY:Explainable XRF Element Detection with Embedding-Space KNN and Phys
 ics-Based Masks for the HARLOCK Lunar Rover - Alberto Luise\, Ghulam Musta
 fa
URL:https://pretalx.adass.org/adass2026/talk/WWKV97/
END:VEVENT
END:VCALENDAR
