In IEEE transactions on pattern analysis and machine intelligence ; h5-index 127.0
In the last years, crowdsourcing is transforming the way classification sets are obtained. Instead of relying on a single expert, crowdsourcing shares the effort among a large number of collaborators. This is being applied in the laureate Laser Interferometer Gravitational Waves Observatory (LIGO) in order to detect glitches which might hinder the identification of gravitational-waves. Probabilistic methods, such as Gaussian Processes (GP), have proven successful in crowdsourcing. However, GPs do not scale well to large sets (such as LIGO), which hampers their broad adoption. This has led to the recent introduction of deep learning based crowdsourcing methods, which have become the state-of-the-art. However, the accurate uncertainty quantification of GPs has been sacrificed. In this work, we first leverage a standard sparse GP approximation (SVGP) to develop a GP-based crowdsourcing method that factorizes into mini-batches. This makes it able to cope with previously-prohibitive sets. This first approach, Scalable Variational Gaussian Processes for Crowdsourcing (SVGPCR), brings back GP-based methods to the state-of-the-art, and excels at uncertainty quantification. SVGPCR outperforms deep learning methods and previous probabilistic ones on LIGO data. Its behavior is analyzed in a controlled experiment on MNIST. Moreover, recent GP inference techniques are also adapted to crowdsourcing and evaluated experimentally.
Morales-Alvarez Pablo, Ruiz Pablo, Coughlin Scotty, Molina Soriano Rafael, Katsaggelos Aggelos