Abstract
The present disclosure provides a method, a system, and a computer program product for routing machine learning inference requests based on properties of a format-preserving encrypted identifier without performing decryption. The method comprises receiving a transaction request containing an identifier encrypted using format-preserving encryption, wherein the encrypted identifier preserves the format and length of the original identifier. The method includes extracting a prefix portion of the encrypted identifier without performing decryption and without accessing cryptographic key material. The method includes matching the extracted prefix against a pre-computed routing table that maps encrypted prefixes to machine learning model versions, wherein the routing table is generated by applying the same format-preserving encryption to known identifier prefixes. The method also includes routing the inference request to the matched model version for scoring. The encrypted routing table provides a deterministic mapping between an encrypted prefix and a corresponding model version while maintaining encryption of the identifier throughout the routing and scoring pipeline.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Qin, Hao and Xia, Ruiyang, "SYSTEM, METHOD, AND COMPUTER PROGRAM PRODUCT FOR ROUTING A MACHINE LEARNING INFERENCE REQUEST TO A MODEL VERSION BASED ON PROPERTIES OF A FORMAT-PRESERVING ENCRYPTED IDENTIFIER WITHOUT DECRYPTION", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/12404