Sirvi Autor "Kuslap, Hannes" järgi

Nüüd näidatakse 1 - 2 2

Eesti tulevikukliima temperatuuriekstreemumid
(Tartu Ülikool, 2023) Kuslap, Hannes; Toll, Velle, juhendaja; Luhamaa, Andres, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Füüsika instituut
Masinõppe algoritmide rakendamine füüsikast lähtuvate fotomeetriliste punanihete täpsustamisel
(Tartu Ülikool, 2025) Kuslap, Hannes; Tempel, Elmo, juhendaja; Tuvikene, Taavi, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
This thesis explores the application of machine learning methods to improve the accuracy of photometric redshift (photo-z) estimates by refining the outputs of the physics-based model TOPz. Using data from the WAVES survey, several regression based machine learning models were trained to reproduce and enhance TOPz outputs. The best-performing model was XGBoost, which with an optimal configuration significantly reduced prediction errors and the proportion of outliers compared to the original TOPz estimates. Two enhancement models were developed: one directly predicted the logarithmic redshift transformation ζ = ln(1 + z), while the other estimated the full ζ probability distribution. Both models were combined with the original TOPz outputs, with the direct ζ prediction model achieving the lowest error (MAE = 0,0265), and the probability distribution model providing better interpretability and the ability to handle ambiguous solutions. The best results were achieved by linearly or geometrically combining model outputs, optimizing the weight between TOPz and XGBoost contributions. The study demonstrates that a hybrid approach combining physical modeling and machine learning enables significantly more accurate and robust photometric redshift estimates, which are essential for large-scale astronomical surveys such as J-PAS and WAVES.