This paper presents two metric learning approaches for face identification: Logistic Discriminant Metric Learning (LDML) and Marginalised kNN (MkNN). Face identification is the task of determining whether two face images depict the same person. The challenge lies in variations in scale, pose, lighting, background, expression, hairstyle, and glasses. The authors evaluate their methods on the Labeled Faces in the Wild (LFW) dataset, a large and challenging dataset of faces from Yahoo! News. The evaluation protocol defines two settings: restricted, where a fixed set of positive and negative image pairs is given, and unrestricted, where faces are labelled by their identity. The authors are the first to present results for the unrestricted setting, showing that their methods benefit from this richer training data, much more so than the current state-of-the-art method. Their results of 79.3% and 87.5% correct for the restricted and unrestricted settings, respectively, significantly improve over the current state-of-the-art result of 78.5%. Confidence scores obtained for face identification can be used for many applications, such as clustering or recognition from a single training example. The authors show that their learned metrics also improve performance for these tasks. The paper also discusses the application of learned face metrics for unsupervised clustering of face images and face recognition from a single exemplar. The results show that LDML achieves a classification accuracy of 79.3% on the restricted setting of the LFW dataset, and LDML and MkNN achieve comparable accuracies on the unrestricted setting, around 83%. The authors conclude that their methods significantly improve performance for face identification tasks, especially when using more training data. They also note that pose changes remain a major challenge for future work.This paper presents two metric learning approaches for face identification: Logistic Discriminant Metric Learning (LDML) and Marginalised kNN (MkNN). Face identification is the task of determining whether two face images depict the same person. The challenge lies in variations in scale, pose, lighting, background, expression, hairstyle, and glasses. The authors evaluate their methods on the Labeled Faces in the Wild (LFW) dataset, a large and challenging dataset of faces from Yahoo! News. The evaluation protocol defines two settings: restricted, where a fixed set of positive and negative image pairs is given, and unrestricted, where faces are labelled by their identity. The authors are the first to present results for the unrestricted setting, showing that their methods benefit from this richer training data, much more so than the current state-of-the-art method. Their results of 79.3% and 87.5% correct for the restricted and unrestricted settings, respectively, significantly improve over the current state-of-the-art result of 78.5%. Confidence scores obtained for face identification can be used for many applications, such as clustering or recognition from a single training example. The authors show that their learned metrics also improve performance for these tasks. The paper also discusses the application of learned face metrics for unsupervised clustering of face images and face recognition from a single exemplar. The results show that LDML achieves a classification accuracy of 79.3% on the restricted setting of the LFW dataset, and LDML and MkNN achieve comparable accuracies on the unrestricted setting, around 83%. The authors conclude that their methods significantly improve performance for face identification tasks, especially when using more training data. They also note that pose changes remain a major challenge for future work.