Non-stationary and unpredictable data distributions in classification and quantification

Abstract

This work is dedicated to all people who are unable to dedicate something. I also thank my advisor Prof. Batista, for whom, at times, I have been a heavier burden than he could have anticipated. He did not drop the weight, and for that he has my gratitude. Once Prof. Batista moved overseas, Prof. Parros welcomed me with open arms, and I cannot thank her enough for personally supporting my final months in this journey. Finally, I'm immensely thankful that I could collaborate with André Maletzke, who rocketed many of my ideas off the ground. Additional thanks go to FAPESP and CAPES, that financially supported this thesis (grants 2017/22896-7 and PROEX-6909543/D, respectively). "'All conditioned things are impermanent'-when one sees this with wisdom, one turns away from suffering." (The Buddha) Palavras-chave: dados não estacionários, aprendizado com dados positivos e não-rotulados, quantificação, classificação, mudança de conceito.