Learning from majority label: A novel problem in multi-class multiple-instance learning

Open Webpage

Kaito Shiku, Shinnosuke Matsuo, Daiki Suehiro, Ryoma Bise

Published: 01 Jan 2026, Last Modified: 13 Nov 2025Pattern Recognit. 2026EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•We proposed “Learning from Majority Label” (LML) as one of the most practical settings in multi-class Multiple-Instance Learning.•We proposed a Counting Network trained to produce bag-level majority labels, estimated by counting the number of instances in each class.•Based on our detailed analysis, we developed a Majority Proportion Enhancement Module (MPEM).•We demonstrated the superiority of the proposed method on four datasets compared to conventional MIL methods.