Chengjiang Long
Kitware Inc, USA
Title: Collaborative Active Learning from Crowds for Visual Recognition
Biography
Biography: Chengjiang Long
Abstract
Active learning is an effective way to relieve the tedious work of manual annotation in many applications of visual recognition. The vast majority of previous works, if not all of them, focused on active learning with a single human oracle. The problem of active learning with multiple oracles in a collaborative setting has not been well explored. Moreover, most of the previous works assume that the labels provided by the human oracles are noise free, which may often be violated in reality.
To solve the above-mentioned issues, we proposed two models (i.e., a distributed multi-labeler active learning model and a centralized multi-labeler active learning model) for collaborative active visual recognition from the crowds, where we explore how we can effectively model the labelers’ expertise in a crowdsourcing labeling system to build better visual recognition models. Both two models are not only robust to label noises, but also a principled label quality measure to online detect irresponsible labelers. We also extended the centralized multi-labeler active learning model from binary cases to multi-class cases and also incorporate the idea of reinforcement learning to actively select both the informative samples and the high-quality annotators, which better explores the trade-off between exploitation and exploration.
Our collaborative active learning models have been validated in the real-world visual recognition benchmark datasets. The experimental results strongly show the validity and efficiency of the two proposed models.