| CPC G06F 16/906 (2019.01) [G06F 16/24578 (2019.01); G06F 16/258 (2019.01); H03M 7/30 (2013.01)] | 19 Claims |
|
1. A computer-implemented method, the method comprising:
receiving, by a computing system comprising one or more processors, a search query;
determining, by the computing system, query inner products between the search query and a plurality of quantized elements that correspond to a plurality of data elements, wherein the query inner products approximate reference inner products between the search query and the plurality of data elements, wherein the plurality of quantized elements were determined based on an inner-product weighted quantization error; and
providing, by the computing system, a search result responsive to the search query based on the determined query inner products, wherein the search result corresponds to a data element from among the plurality of data elements that is associated with an inner product between a corresponding quantized element and the search query having a magnitude which is greater than magnitudes of inner products between the search query and other quantized elements among the plurality of quantized elements,
wherein
the inner-product weighted quantization error for each quantized element among the plurality of quantized elements comprises:
a parallel component that is parallel to a corresponding data element, and
an orthogonal component that is orthogonal to the corresponding data element; and
the inner-product weighted quantization error includes a weighted sum of the parallel component and the orthogonal component.
|