Bayesian Active Learning for Classification and Preference Learning - 2011

Title : Bayesian Active Learning for Classification and Preference Learning Author(s): Houlsby, Neil and Huszár, Ferenc and Ghahramani, Zoubin and Lengyel, Máté Link(s) : http://arxiv.org/abs/1112.5745

Introduces an active learning approach using predictive entropies applied to Gaussian Process (GP) classification (and GP preference learning).

The authors note two approaches to active learning:

• Decision theoretic, (, ), minimizes expected loss after decisions are made based on collected data.
• Information theoretic, (, a), aim to reduce model space as quickly as possible, often using concepts from information theory.