Deep learning diagnostic algorithms are proving comparable results with human experts in a wide variety of tasks, they still require a huge amount of well annotated data for training which is often non-affordable.
Metric learning techniques have allowed a reduction on the required annotated data allowing few-shot learning. Existing deep metric learning loss functions have made possible generating models capable of tackling complex scenarios with the presence of many classes and scarcity on the number of images per class, not only work for classification tasks, but to many other clinical applications where measuring similarity is the key.
Currently used state-of-the-art loss functions still suffer from slow convergence due to the selection of effective training samples that has been partially solved by the multi-class N-pair loss by simultaneously adding additional samples from the different classes.
The constellation loss goes one step further by simultaneously learning distances among all class combinations. The application of the constellation loss for visual class embedding for Hematoxylin-Eosin microscopy images class characterization shows that constellation loss function over-performs the other methods by obtaining more compact clusters while achieving better classification results for cancer detection on histopathology images.