Artificial neural networks (not only) in spectroscopy suffer from poor interpretability due to an extensive number of parameters in the model. We propose an entirely new perspective on interpretability through lottery tickets (i.e., tiny, sparse networks with high performance). We will exploit lottery tickets for interpretability as they dramatically reduce the number of parameters and preserve the spatial structure of the data. Also, we will study a potential of lottery tickets for embedded spectroscopic systems (e.g., satellites, rovers).

CEITEC VUT-J-23-8332

Czech

Vrábel Jakub, Ing., Ph.D. - principal person responsible
Pořízka Pavel, doc. Ing., Ph.D. - fellow researcher

Advanced instrumentation and methods for material characterization
- responsible department (6.3.2023 - not assigned)
Central European Institute of Technology BUT
- responsible department (29.1.2023 - 6.3.2023)
Advanced instrumentation and methods for material characterization
- beneficiary (1.1.2023 - 31.12.2023)

SAEIDFIROUZEH, H.; KUBELÍK, P.; LAITL, V.; KŘIVKOVÁ, A.; VRÁBEL, J.; RAMMELKAMP, K.; SCHRÖDER, S.; GORNUSHKIN, I.; KÉPEŠ, E.; ŽABKA, J.; FERUS, M.; POŘÍZKA, P.; KAISER, J. Laser-induced breakdown spectroscopy in space applications: Review and prospects. TRAC-TRENDS IN ANALYTICAL CHEMISTRY, 2024, vol. 181, no. B, p. 1-22. ISSN: 1879-3142.
Detail

Responsibility: Vrábel Jakub, Ing., Ph.D.