Counting Network for Learning from Majority Label

TL;DR

The paper tackles Learning from the Majority Label (LML), a mono-label multi-class MIL task in which bag labels encode the majority class among contained instances. It introduces a Counting Network that uses a softmax with temperature to produce near-binary instance predictions, enabling differentiable counting so that bag-level counts $\hat N^i_c$ can be obtained and the bag majority $\hat{\bm Y}^i = s(\bm N^i, T)$ can be predicted. Training minimizes cross-entropy between the estimated majority $\hat{\bm Y}^i$ and the ground-truth $\bm Y^i$ to align instance counts with bag labels. Experiments on four datasets demonstrate superior instance-level accuracy over strong MIL baselines, with ablations showing the counting operation and the argmax-based selection reduce overestimation and improve consistency. The approach offers a principled path to recover per-instance labels from bag-level majority supervision, with potential impact in domains like medical imaging where majority diagnoses guide decisions.

Abstract

The paper proposes a novel problem in multi-class Multiple-Instance Learning (MIL) called Learning from the Majority Label (LML). In LML, the majority class of instances in a bag is assigned as the bag's label. LML aims to classify instances using bag-level majority classes. This problem is valuable in various applications. Existing MIL methods are unsuitable for LML due to aggregating confidences, which may lead to inconsistency between the bag-level label and the label obtained by counting the number of instances for each class. This may lead to incorrect instance-level classification. We propose a counting network trained to produce the bag-level majority labels estimated by counting the number of instances for each class. This led to the consistency of the majority class between the network outputs and one obtained by counting the number of instances. Experimental results show that our counting network outperforms conventional MIL methods on four datasets The code is publicly available at https://github.com/Shiku-Kaito/Counting-Network-for-Learning-from-Majority-Label.

Figures (4)

• Figure 1: A novel problem setting: Learning from Majority.
• Figure 2: Counting approach.
• Figure 3: Overview of the proposed method.
• Figure 4: Effect of $\arg\max$ operation in each scenario. The subtraction value is average on all datasets.