# US 12,120,216 B2 · Secure process for validating machine learning models using homomorphic encryption techniques > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,120,216 B2 - Document ID: US12120216B2 - Status: Granted - Kind: B2 - Application number: 17/747,031 - Filed: 2022-05-18 - Published: 2022-12-01 - Granted: 2024-10-15 - Earliest claimed: 2021-05-25 - Assignees: Unify Patente GmbH & Co. KG - Applicants: Unify Patente GmbH & Co. KG - Inventors: Michael Brochonski - Web page: https://patentgrep.com/patent/US12120216B2 ## CPC - H04L 9/008: {involving homomorphic encryption} - G06N 20/00: Machine learning - H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract Methods and systems for secure validation of a machine learning model can facilitate encryption of a machine learning model and evaluating the performance of the model. Validation data can be encrypted and processed by the machine learning model so that results of the model's processing can be received and evaluated to evaluate how the machine learning model functioned. ## Claims ### Claim 1 A method for secure validation of machine learning models and parallel validation data using homomorphic encryption, comprising: providing, by a provider, a machine learning model and providing, by a user, validation data; encrypting, by the provider, the machine learning model; sending, by the provider, a public encryption parameter to the user; selecting, by the user and provider, a unifying encoding method; encrypting, by the user, the validation data; sending, by the user, the encrypted validation data; processing, the encrypted validation data with the encrypted machine learning model; providing encrypted results of said processing to the provider and the user; decrypting the results and evaluating whether the performance of the machine learning model is satisfactory with the given valuation data of the user; and wherein the step of processing, the encrypted validation data with the encrypted machine learning model is repeated with another unifying coding method in case the result of the machine learning model does not meet the requirements for accuracy and efficiency. ### Claim 2 (depends on 1) The method according to claim 1, wherein: the step of encrypting, by the provider, the machine learning model further comprises generating, by the provider, public, secure and/or functionality homomorphic encryption parameters; and the method further comprises sending, by provider, the homomorphic encryption parameters to user. ### Claim 3 (depends on 2) The method according to claim 2, wherein the public homomorphic encryption parameters comprising a scheme defining the precision and efficiency of the subsequent processing of the encrypted validation data with the encrypted machine learning model, wherein the scheme is a Brakerski-Fan-Vercauteren (BFV) or a Cheon-Kim-Kim-Song (CKKS) scheme. ### Claim 4 (depends on 3) The method according to claim 3, wherein the functionality homomorphic encryption parameters comprising one of cyclotonic ring, modulus (modulo) and/or level depth. ### Claim 5 (depends on 1) The method according to claim 1, wherein the unifying encoding method use a block size of n=256 bits or a block size of n=128 bits. ### Claim 6 (depends on 1) The method according to claim 1, wherein the step of encrypting, by the user, the validation data (further comprises generating, by user, public keys that are going to be used during the transaction process; and wherein the method further comprising the steps: sending, by user, the encrypted validation data and the generated public keys to the provider; and processing, the encrypted validation data with the encrypted machine learning model, wherein the public keys of user are used. ### Claim 7 (depends on 1) The method according to claim 1, wherein the number of repeated processing of the machine learning model is limited to a predetermined threshold n that is to avoid the risk of extraction attacks. ### Claim 8 A method for secure validation of machine learning models and parallel validation data using homomorphic encryption, comprising: providing, by a provider, a machine learning model and providing, by a user, validation data; encrypting, by the provider, the machine learning model; sending, by the provider, a public encryption parameter to the user; selecting, by the user and provider, a unifying encoding method; encrypting, by the user, the validation data; sending, by the user, the encrypted validation data; processing, the encrypted validation data with the encrypted machine learning model; providing encrypted results of said processing to the provider and the user; decrypting the results and evaluating whether the performance of the machine learning model is satisfactory with the given valuation data of the user; and wherein a neural network watermarking is used to trace the machine learning model upon a determination that a redistribution of the provided machine learning model is not to occur. ### Claim 9 (depends on 1) The method according to claim 1, wherein the method is executed on an online external system, a public cloud solution, and/or a private offline system. ### Claim 10 A system for secure validation of machine learning models and parallel validation data using homomorphic encryption, the system comprising a provider and a user, the provider including a computer device having a processor connected to a non-transitory computer readable medium, the user including a computer device having a processor connected to a non-transitory medium; the user being communicatively connectable to the provider, wherein the system is configured such that a machine learning model is providable via the provider and validation data is providable via the user; the provider configured to encrypt the machine learning model and send a public encryption parameter to the user; the user and the provider configured to select a unifying encoding method; the user configured to encrypt the validation data and send the validation data so that the validation data is processable by the encrypted machine learning model to output encryption results that are providable to the user and the provider, the user and the provider being configured to decrypt the encryption results and evaluate whether the machine learning model performed satisfactorily with the valuation data, and in response to the machine learning model performing unsatisfactorily with the valuation data and does not meet requirements for accuracy and/or efficiency, the encrypted machine learning model processing of the validation data is repeatable with another unifying coding method. ### Claim 11 (depends on 10) The system according to claim 10, wherein the system comprising at least one of online external system, public cloud solution system and/or private offline system. ### Claim 12 (depends on 10) The system of claim 10, wherein the user is a component of an online external system, a public cloud computer system, a private offline system, a local system, a network system and/or cloud system. ### Claim 13 (depends on 10) The system of claim 10, wherein the provider is a component of an online external system, a public cloud computer system, a private offline system, a local system, a network system and/or cloud system. ### Claim 14 (depends on 1) The method of claim 1, wherein the provider includes a computer device having a processor connected to a non-transitory computer readable medium and the user includes a computer device having a processor connected to a non-transitory medium. ### Claim 15 (depends on 14) The method of claim 14, wherein the user is a component of an online external system, a public cloud computer system, a private offline system, a local system, a network system and/or cloud system. ### Claim 16 (depends on 13) The method of claim 13, wherein the provider is a component of an online external system, a public cloud computer system, a private offline system, a local system, a network system and/or cloud system. ### Claim 17 (depends on 8) The method of claim 8, wherein the provider includes a computer device having a processor connected to a non-transitory computer readable medium and the user includes a computer device having a processor connected to a non-transitory medium. ### Claim 18 (depends on 17) The method of claim 17, wherein the user is a component of an online external system, a public cloud computer system, a private offline system, a local system, a network system and/or cloud system. ### Claim 19 (depends on 17) The method of claim 17, wherein the provider is a component of an online external system, a public cloud computer system, a private offline system, a local system, a network system and/or cloud system.