The 36th conference on Neural Information Processing Systems (NeurIPS) will be held in New Orleans from November 28th to December 9th.
Our group will be presenting an interesting paper entitled “Bayesian Active Learning with Fully Bayesian Gaussian Processes“. In this paper, we exploit the bias-variance trade-off inherent in a Gaussian process’ hyperparameters to design two new active learning acquisition functions: Bayesian Query-by-Committee and Query-by Mixture of Gaussian Processes. For six simulators, we improve the active learning performance by 12%! We will use these insights to efficiently analyze the effects of policies in traffic simulators, leading to greener aviation traffic 🌱✈.
Check below for more details!
- Session: Poster Session 5
- Time: Thursday, 1 December, 18:30
- Title: Bayesian Active Learning with Fully Bayesian Gaussian Processes [Paper] [Code]
- Authors: Christoffer Riis, Francisco Antunes, Frederik Boe Hüttel, Carlos Lima Azevedo, Francisco Camara Pereira