Leveraging Class Hierarchies with Metric-Guided Prototype Learning
This work addresses classification tasks where classes have hierarchical relationships, offering incremental improvements over existing prototype-based methods.
The paper tackled the problem of classification with hierarchical class structures by integrating a metric derived from the hierarchy into a prototypical network, resulting in improved error rates weighted by the cost matrix and overall precision across four public datasets.
In many classification tasks, the set of target classes can be organized into a hierarchy. This structure induces a semantic distance between classes, and can be summarised under the form of a cost matrix, which defines a finite metric on the class set. In this paper, we propose to model the hierarchical class structure by integrating this metric in the supervision of a prototypical network. Our method relies on jointly learning a feature-extracting network and a set of class prototypes whose relative arrangement in the embedding space follows an hierarchical metric. We show that this approach allows for a consistent improvement of the error rate weighted by the cost matrix when compared to traditional methods and other prototype-based strategies. Furthermore, when the induced metric contains insight on the data structure, our method improves the overall precision as well. Experiments on four different public datasets - from agricultural time series classification to depth image semantic segmentation - validate our approach.