Unsupervised representation learning by discovering reliable image relations
作者：
Highlights：
• Identify reliable image relations to drive unsupervised representation learning.
• Divide-and-conquer strategy to retain only reliable relations between images.
• Learning an ensemble of local representations by matching to random target embeddings.
• Iterative extension of the set of reliable relations by inference.
• Consolidate the local representations into one global representation by transitivity.
摘要
•Identify reliable image relations to drive unsupervised representation learning.•Divide-and-conquer strategy to retain only reliable relations between images.•Learning an ensemble of local representations by matching to random target embeddings.•Iterative extension of the set of reliable relations by inference.•Consolidate the local representations into one global representation by transitivity.
论文关键词：Unsupervised learning,Visual representation learning,Unsupervised image classification,Mining reliable relations,Divide-and-conquer
论文评审过程：Received 13 May 2019, Revised 16 September 2019, Accepted 10 November 2019, Available online 17 January 2020, Version of Record 30 January 2020.
论文官网地址：https://doi.org/10.1016/j.patcog.2019.107107
