Preserving Fine-Grain Feature Information in Classification via Entropic Regularization
This addresses the issue of information loss in classification when using coarse labels, which is incremental as it builds on existing regularization techniques.
The paper tackles the problem of training models on coarse labels for fine-grained classification or regression, showing that standard cross-entropy can overfit to coarse features. It introduces an entropy-based regularization to promote feature diversity, empirically demonstrating improved performance on fine-grained tasks.
Labeling a classification dataset implies to define classes and associated coarse labels, that may approximate a smoother and more complicated ground truth. For example, natural images may contain multiple objects, only one of which is labeled in many vision datasets, or classes may result from the discretization of a regression problem. Using cross-entropy to train classification models on such coarse labels is likely to roughly cut through the feature space, potentially disregarding the most meaningful such features, in particular losing information on the underlying fine-grain task. In this paper we are interested in the problem of solving fine-grain classification or regression, using a model trained on coarse-grain labels only. We show that standard cross-entropy can lead to overfitting to coarse-related features. We introduce an entropy-based regularization to promote more diversity in the feature space of trained models, and empirically demonstrate the efficacy of this methodology to reach better performance on the fine-grain problems. Our results are supported through theoretical developments and empirical validation.