Although face recognition has achieved fairly remarkable results in recent years, it heavily relies on large-scale labeled face datasets to train the high-capacity deep convolutional neural networks. However, it is unrealistic to collect larger
labeled datasets to further boost the performance, which requires burdensome and expensive annotation efforts. Meanwhile, we have easy access to abundant unlabeled face data. It is a natural idea to jointly utilize limited labeled and abundant unlabeled data to obtain higher performance gain, which is the
target of semi-supervised learning. In this paper, we propose a unified Neighborhood-Aware Attention Network (NAAN) for semi-supervised face recognition, where the neighborhood is defined as a k-hop ego network centered in the given sample called “ego”. Considering the different importance of neighbors, we employ the graph attention network to learn the ego’s representation, which selectively attends to informative nodes in the neighborhood. With the neighborhood-aware embeddings, NAAN infers pairwise relations of unlabeled face images to cluster them. We evaluate our model on two face recognition datasets MegaFace and IJB-A, and it yields favorably comparable performance to the fully-supervised results.