Abstract

Embedding representation learning via neural networks is at the corefoundation of modern similarity based search. While much effort has been put indeveloping algorithms for learning binary hamming code representations forsearch efficiency, this still requires a linear scan of the entire dataset pereach query and trades off the search accuracy through binarization. To thisend, we consider the problem of directly learning a quantizable embeddingrepresentation and the sparse binary hash code end-to-end which can be used toconstruct an efficient hash table not only providing significant searchreduction in the number of data but also achieving the state of the art searchaccuracy outperforming previous state of the art deep metric learning methods.We also show that finding the optimal sparse binary hash code in a mini-batchcan be computed exactly in polynomial time by solving a minimum cost flowproblem. Our results on Cifar-100 and on ImageNet datasets show the state ofthe art search accuracy in [email protected] and NMI metrics while providing up to98X and 478X search speedup respectively over exhaustive linear search.

 

Introduction (beta)

None

 

Conclusion (beta)

None