Specific Star Formation Rate Estimation and Classification Using Adaptive Learning Techniques

MSc thesis defense by Marcho Markov and Traian Popa


In this project, in the first part, we investigate the performance of several features extracted on datasets of images of galaxies. We construct regression and classfication models with the best performing of those features in order to predict the specific star formation rate of a galaxy as an alternative to the more used and slower method of spectroscopic analysis.

The second part of the project deals with investigating methods of adapting our models from an older dataset to a more recent one. We present an algorithm that uses a form of semi-supervised machine learning called active learning as an approach to domain adaptation and we test different scenarios for determining the proper balance between the source and target models for a successful and accurate model shift.

Supervisors: Kim Steenstrup Pedersen

Censor: Ole Winther, IMM DTU