Distributed similarity learning for high-dimensional image features

A system comprising:

• a plurality of processors;
• computer storage media storing computer-usable instructions that, when used by the plurality of processors, causes the plurality of processors to:
• access a set of data features corresponding to an image;
• determine a plurality of orthogonal subspaces from a space formed by the set of data features using a set of projection matrices, each subspace having a set of lower dimension data features than a dimension of the set of data features;
• learning, in parallel, an optimal matrix for each of a plurality of similarity functions for the plurality of orthogonal subspaces that generates a similarity measure consistent with a respective set of lower dimension data features as constraints, wherein, each similarity function is based on a dimension of a corresponding subspace:
• solve the combined similarity function, using a support vector machine, the combined similarity function based on a combination of each learned optimal matrix;
• and
• based on the solving, associate a tag, title, description, or metadata to the image.

A non-transitory machine-readable storage medium storing instructions which, when executed by at least one processor, performs operations comprising:

• accessing a set of data features corresponding to an image;
• determining a plurality of orthogonal subspaces from a space formed by the set of data features using a set of projection matrices, each subspace having a set of lower dimension data features than a dimension of the set of data features;
• learning, in parallel, an optimal matrix for each of a plurality of similarity functions for the plurality of orthogonal subspaces that generates a similarity measure consistent with a respective set of lower dimension data features as constraints, wherein each similarity function is based on a dimension of a corresponding subspace;
• solving the combined similarity function, using a support vector machine, the combined similarity function based on a combination of each learned optimal matrix;
• and
• based on the solving, associating a tag, title, description, or metadata to the image.