Vector Search Using Adapted Embeddings
AssigneeOracle International Corporation
InventorsMichael Louis Wick, Jeffrey H. Alexander
Life of this patent
Estimating the term…
Abstract
Methods, techniques and systems for providing improved vector search using adapted embedding are disclosed. Vector representations of documents may be stored for searching, the vectors generated according to a pre-trained embedding model. A de-embedding head of the model may be repurposed as an inverse embedding model for determining similarity during search, the inverse embedding model trained on the generated vectors and the resulting model weights used to precompute an inverse hessian matrix. The precomputed matrix may then be used to identify vectors according to search query predicates.
Claims20
A computer-implemented method, comprising:
- training an inverse embedding model according to a plurality of vectors generated according to an embedding model and respective input data, the plurality of vectors individually comprising one or more dimensions;
- and
- identifying at least one vector of the plurality of vectors based at least in part on a similarity to at least one query vector generated from a query predicate according to the embedding model, wherein the similarity is determined according to a similarity function based at least in part on the inverse embedding model.
One or more non-transitory computer-accessible storage media storing program instructions that when executed on or across one or more processors cause the one or more processors to perform:
- training an inverse embedding model according to a plurality of vectors generated according to an embedding model and respective input data, the plurality of vectors individually comprising one or more dimensions;
- and
- identifying at least one vector of the plurality of vectors based at least in part on a similarity to at least one query vector generated from a query predicate according to the embedding model, wherein the similarity is determined according to a similarity function based at least in part on the inverse embedding model.
A system, comprising:
- a vector store comprising at least one processor and a memory, configured to:
- store a plurality of vectors generated according to an embedding model and respective input data, the plurality of vectors individually comprising one or more dimensions;
- train an inverse embedding model according to the plurality of and the respective input data;
- and
- identify, responsive to a query received at the vector store, at least one vector of the plurality of vectors based at least in part on a similarity to at least one query vector generated from a predicate of the query according to the embedding model, wherein the similarity is determined according to a similarity function based at least in part on the inverse embedding model.
Dependencies are inferred from the claim text.