Fast learning from label proportions with small bags
This work addresses the challenge of learning instance classifiers when only bag-level label proportions are available, which is useful for scenarios where individual labeling is costly or impossible, but it is incremental as it builds on existing LLP methods.
The paper tackled the problem of learning from label proportions with small bags by proposing an EM algorithm that alternates between optimizing a neural network classifier and incorporating bag-level annotations. The results showed that this approach converges faster to a comparable or better solution compared to existing methods on two image datasets.
In learning from label proportions (LLP), the instances are grouped into bags, and the task is to learn an instance classifier given relative class proportions in training bags. LLP is useful when obtaining individual instance labels is impossible or costly. In this work, we focus on the case of small bags, which allows to design an algorithm that explicitly considers all consistent instance label combinations. In particular, we propose an EM algorithm alternating between optimizing a general neural network instance classifier and incorporating bag-level annotations. Using two different image datasets, we experimentally compare this method with an approach based on normal approximation and two existing LLP methods. The results show that our approach converges faster to a comparable or better solution.