Hierarchical Class-Based Curriculum Loss

TL;DR

This paper introduces a hierarchical curriculum loss function for classification tasks that respects label hierarchies and assigns different weights to labels based on their levels, improving accuracy and interpretability.

Contribution

It proposes a novel loss function that enforces hierarchical constraints and learns non-uniform label weights, with theoretical guarantees and empirical validation.

Findings

Outperforms baseline methods on real-world image datasets

Provides a tighter bound of 0-1 loss under hierarchical constraints

Enhances model interpretability and accuracy

Abstract

Classification algorithms in machine learning often assume a flat label space. However, most real world data have dependencies between the labels, which can often be captured by using a hierarchy. Utilizing this relation can help develop a model capable of satisfying the dependencies and improving model accuracy and interpretability. Further, as different levels in the hierarchy correspond to different granularities, penalizing each label equally can be detrimental to model learning. In this paper, we propose a loss function, hierarchical curriculum loss, with two properties: (i) satisfy hierarchical constraints present in the label space, and (ii) provide non-uniform weights to labels based on their levels in the hierarchy, learned implicitly by the training paradigm. We theoretically show that the proposed loss function is a tighter bound of 0-1 loss compared to any other loss satisfying the hierarchical constraints. We test our loss function on real world image data sets, and show that it significantly substantially outperforms multiple baselines.