My work is mostly concerned with applications in astro-particle physics, a field that is characterized by extreme class imbalances, by a domain-specific post-processing of predictions, and by the fact that all training data is simulated while the learned models must be valid in practice. My current focuses are the aggregation of predictions in terms of ordinal quantification (a.k.a. unfolding), learning under class-conditional label noise, and the smart control of simulations through active class selection.
mirkobunse / ordinal_quantification Goto Github PK
View Code? Open in Web Editor NEWThis project forked from bertocast/ordinal_quantification
Home Page: https://mirkobunse.github.io/ordinal_quantification