The paper "Is that you? Metric learning approaches for face identification" by Matthieu Guillaumin, Jakob Verbeek, and Cordelia Schmid presents two methods for learning robust distance measures for face identification: Logistic Discriminant Metric Learning (LDML) and Marginalised kNN classification (MkNN). The authors evaluate these methods on the Labeled Faces in the Wild (LFW) dataset, a challenging dataset of faces from Yahoo! News, which includes variations in scale, pose, lighting, background, expression, and more. LDML learns a metric from a set of labeled image pairs, aiming to minimize distances between positive pairs while maximizing distances between negative pairs. MkNN, on the other hand, computes the probability that two images belong to the same class using a k-nearest-neighbour classifier. The authors compare their methods with state-of-the-art techniques such as Large Margin Nearest Neighbour (LMNN) and Information Theoretic Metric Learning (ITML). In the restricted setting, where only a subset of positive and negative pairs are labeled, LDML achieves an accuracy of 79.3%, outperforming the current best method (78.5%). In the unrestricted setting, where all available face labels are used, LDML and MkNN achieve an accuracy of 87.5%, significantly improving over the state-of-the-art. The authors also demonstrate that their learned metrics improve performance in clustering and recognition from a single example. The paper highlights the effectiveness of metric learning in handling the challenges of face identification, particularly in leveraging more training data to improve performance.The paper "Is that you? Metric learning approaches for face identification" by Matthieu Guillaumin, Jakob Verbeek, and Cordelia Schmid presents two methods for learning robust distance measures for face identification: Logistic Discriminant Metric Learning (LDML) and Marginalised kNN classification (MkNN). The authors evaluate these methods on the Labeled Faces in the Wild (LFW) dataset, a challenging dataset of faces from Yahoo! News, which includes variations in scale, pose, lighting, background, expression, and more. LDML learns a metric from a set of labeled image pairs, aiming to minimize distances between positive pairs while maximizing distances between negative pairs. MkNN, on the other hand, computes the probability that two images belong to the same class using a k-nearest-neighbour classifier. The authors compare their methods with state-of-the-art techniques such as Large Margin Nearest Neighbour (LMNN) and Information Theoretic Metric Learning (ITML). In the restricted setting, where only a subset of positive and negative pairs are labeled, LDML achieves an accuracy of 79.3%, outperforming the current best method (78.5%). In the unrestricted setting, where all available face labels are used, LDML and MkNN achieve an accuracy of 87.5%, significantly improving over the state-of-the-art. The authors also demonstrate that their learned metrics improve performance in clustering and recognition from a single example. The paper highlights the effectiveness of metric learning in handling the challenges of face identification, particularly in leveraging more training data to improve performance.