Semantic Hashing
Published:
Semantic hashing uses a word-count vectors obtained from a large set of documents and traines a deep graphical model using semantic information and maps up to the 4th layer where are represented addresses space. Documents are mapped to memory addresses in such a way that semantically similar documents are located at nearby addresses. Documents similar to a query document can then be found by simply accessing all the addresses that differ by only a few bits from the address of the query document. Unlike sparse distributed memory which operates on 1000-bit addresses, semantic hashing works on 32 or 64-bit addresses found in a conventional computer architecture.
See also
Papers
- Salakhutdinov, R., & Hinton, G. (2007). Semantic hashing. RBM, 500(3), 500.