Unsupervised Projection Models
Unsupervised Projection Models these methods fracture the input feature space using randomly drawn hyperplanes independently of the data distribution. Given the random nature of the learning procedure they are the fastest models at training time but suffer from the disadvantage of requiring long hashcodes and many hashtables to attain a reasonable level of retrieval effectiveness.
Name | Codebook | Architecture | Optimisation |
A. Andoni, P. Indyk, 2006. Near-Optimal Hashing Algorithms for Approximate Nearest Neighbor in High Dimensions |
Binary |
Shallow |
Leech Lattice |
A. Andoni, P. Indyk, T. Laarhoven, 2015. Practical and Optimal LSH for Angular Distance |
Binary |
Shallow |
Randomly Rotated Cross-Polytopes |
M. Datar, N. Immorlica, P. Indyk, V. Mirrokni, 2004. Locality-sensitive hashing scheme based on p-stable distributions |
Binary |
Shallow |
Random Hyperplanes |
A. Gionis, P. Indyk, R. Motwani, 1999. Similarity Search in High Dimensions via Hashing |
Binary |
Shallow |
Bit Sampling |
S. Moran, R. McCreadie, C. Macdonald, I. Ounis, 2016. Enhancing First Story Detection using Word Embeddings |
Binary |
Shallow |
Random Hyperplanes |
S. Petrovic, M. Osborne, V. Lavrenko, 2010. Streaming First Story Detection with application to Twitter |
Binary |
Shallow |
Random Hyperplanes |
S. Petrovic, M. Osborne, V. Lavrenko, 2012. Using paraphrases for improving first story detection in news and Twitter |
Binary |
Shallow |
Random Hyperplanes |
M. Raginsky, S. Lazebnik, 2009. Locality-sensitive binary codes from shift-invariant kernels |
Binary |
Shallow |
Random Fourier Features |
A. Shrivastava, P. Li, 2014. Densifying One Permutation Hashing via Rotation for Fast Near Neighbor Search |
Binary |
Shallow |
Random Permutations |