2d6cc4b2d139a53512fb8cbb3086ae2e-Reviews.html

First provide a summary of the paper, and then address the following criteria: Quality, clarity, originality and significance. This paper proposes a model for labeling images with classes for which no example appear in the training set, which is based on a combination of word and image embeddings and novelty detection. Using distances in th embedding space between test images and unseen and seen class labels, the approach is able to assign a probability for a new image to be from an unseen class. This is later used to decide which classifier to use (one designed for seen classes the other for unknown ones). Results on CIFAR10 are provided.