Transformation of embedding matrices for compression, storage, and reuse

A system comprising:

• at least one processor;
• and
• a non-transitory computer readable medium comprising instructions that, when executed by the at least one processor, cause the system to:
• generate an embedding matrix comprising a plurality of embedding vectors representing a text content item, the embedding matrix having values arranged along a first dimension expected to exhibit locality among the plurality of embedding vectors;
• apply a decorrelating transform along the first dimension of the embedding matrix to generate a transformed embedding matrix, the decorrelating transform applied to compact energy distributed across the embedding matrix in the first dimension into a sub-portion of the transformed embedding matrix;
• compress the transformed embedding matrix by performing a compression operation to generate a compressed embedding matrix;
• and
• store the compressed embedding matrix for the text content item.

A non-transitory computer-readable medium comprising instructions that, when executed by at least one processor, cause the at least one processor to:

• generate an embedding matrix comprising a plurality of embedding vectors representing a text content item;
• select a first dimension of the embedding matrix for application of a decorrelating transform based on an expected locality among values of the embedding matrix along the first dimension;
• apply the decorrelating transform along the first dimension of the embedding matrix to generate a transformed embedding matrix by compacting energy distributed across the embedding matrix in the first dimension into a sub-portion of the transformed embedding matrix;
• and
• compress the transformed embedding matrix by performing a compression operation to generate a compressed embedding matrix.