Reviews: Fast Low-rank Metric Learning for Large-scale and High-dimensional Data

However, it still encounters scalability problem when handling large data. This work gives a new formulation that learns the low-rank cosine similarity metric by embedding the triplet constraints into a matrix to further reduce the complexity and the size of involved matrices. The idea of embedding the evaluation of loss functions into matrices is interesting. For Stiefel manifolds, rather than following the projection and retraction convention, it adopts the optimization algorithm proposed by Wen et al. (Ref. Generally, this paper is well-written with promising results.