IP Library › Granted Patent US 12,333,393
Granted Patent B2
US 12,333,393 · App. 17/985,777 · Granted Jun 17, 2025

Systems and methods for adaptively improving the performance of locked machine learning programs

Inventor: Murali Aravamudan (Andover, MA)
Assignee: nference, Inc.
G06N20/00G06F21/6245G16H10/60
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,333,393
App. No.
17/985,777
Granted
Jun 17, 2025
Kind
B2
Abstract

Techniques for adaptively improving the performance of a locked machine learning program have been disclosed. In one particular embodiment, the techniques may be realized as a method for enabling a first party to provide a trained machine learning model to a second party, the method comprising receiving the trained machine learning model from the first party, the trained machine learning model being associated with one or more policies defining permissible operations; and constraining the second party to operate the trained machine learning model in a manner that is consistent with said one or more policies.

Claims (43)

1. A method for enabling a first party to provide a trained machine learning model to a second party, the method comprising:

receiving the trained machine learning model from the first party, the trained machine learning model being associated with one or more policies defining permissible operations;

constraining the second party to operate the trained machine learning model in a manner that is consistent with said one or more policies;

collecting, by a program operating in a secure enclave serving as a nth-stage of a secure pipeline, a dataset;

transmitting said dataset to a (n−1) stage of said secure pipeline; and

re-training the trained machine learning model using the dataset, wherein prior to re-training the trained machine learning model is locked with respect to a first demographic, and wherein after re-training the trained machine learning model is locked with respect to a heuristically generated different demographic associated with the dataset.

2. The method of claim 1 , further comprising encapsulating the trained machine learning model within a secure enclave.

3. The method of claim 2 , wherein the secure enclave is connected to a first stage of a secure data pipeline.

4. The method of claim 3 , wherein the first stage accepts an output of another data pipeline.

5. The method of claim 4 , wherein the first stage is connected to an input of a second stage data pipeline.

6. The method of claim 1 , further comprising:

collecting operational data associated with the second party operating the trained machine learning model; and

storing the operational data under control of a secure enclave thereby restricting access to the operational data to one or more authenticated parties.

7. The method of claim 6 , further comprising permitting additional data to be added to the operational data by the one or more authenticated parties.

8. The method of claim 7 , further comprising attaching at least one digital certificate to the additional data.

9. The method of claim 1 , wherein the (n−1) stage of said secure pipeline processes the received dataset in a secure enclave.

10. The method of claim 1 , further comprising de-identifying the dataset in a stage of the secure pipeline.

11. A system configured to allow a first party to provide a trained machine learning model to a second party, the system comprising:

a non-transitory memory; and

one or more processors configured to read instructions from the non-transitory memory, the instructions causing the one or more processors to perform operations comprising:

receiving the trained machine learning model from the first party, the trained machine learning model being associated with one or more policies defining permissible operations;

constraining the second party to operate the trained machine learning model in a manner that is consistent with said one or more policies;

collecting, by a program operating in a secure enclave serving as a nth-stage of a secure pipeline, a dataset;

transmitting said dataset to a (n−1) stage of said secure pipeline; and

re-training the trained machine learning model using the dataset, wherein prior to re-training the trained machine learning model is locked with respect to a first demographic, and wherein after re-training the trained machine learning model is locked with respect to a heuristically generated different demographic associated with the dataset.

12. The system of claim 11 , wherein the operations further comprise encapsulating the trained machine learning model within a secure enclave.

13. The system of claim 11 , wherein the operations further comprise:

collecting operational data associated with the second party operating the trained machine learning model; and

storing the operational data under control of a secure enclave thereby restricting access to the operational data to one or more authenticated parties.

14. The system of claim 13 , wherein the operations further comprise permitting additional data to be added to the operational data by the one or more authenticated.

15. The system of claim 11 , wherein the operations further comprise de-identifying the dataset in a stage of the secure pipeline.

16. A non-transitory computer readable medium storing instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:

receiving a trained machine learning model from a first party, the trained machine learning model being associated with one or more policies defining permissible operations;

constraining a second party to operate the trained machine learning model in a manner that is consistent with said one or more policies;

collecting, by a program operating in a secure enclave serving as a nth-stage of a secure pipeline, a dataset;

transmitting said dataset to a (n−1) stage of said secure pipeline; and

re-training the trained machine learning model using the dataset, wherein prior to re-training the trained machine learning model is locked with respect to a first demographic, and wherein after re-training the trained machine learning model is locked with respect to a heuristically generated different demographic associated with the dataset.

17. The non-transitory computer readable medium of claim 16 , wherein the operations further comprise encapsulating the trained machine learning model within a secure enclave.

18. The non-transitory computer readable medium of claim 16 , wherein the operations further comprise:

collecting operational data associated with the second party operating the trained machine learning model; and

storing the operational data under control of a secure enclave thereby restricting access to the operational data to one or more authenticated parties.

19. The non-transitory computer readable medium of claim 18 , wherein the operations further comprise permitting additional data to be added to the operational data by the one or more authenticated parties.

20. The non-transitory computer readable medium of claim 17 , wherein the operations further comprise de-identifying the dataset in a stage of the secure pipeline.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Feb 8, 2023
From: ARAVAMUDAN, MURALI
To: NFERENCE, INC.
Reel/Frame 062629/0387 →
Continuity (7)
Continuation In Part 16908520 · Jun 22, 2020
Provisional Application 63012738 · Apr 20, 2020
Provisional Application 62984989 · Mar 4, 2020
Provisional Application 62985003 · Mar 4, 2020
Provisional Application 62962146 · Jan 16, 2020
Provisional Application 62865030 · Jun 21, 2019
Related Publication 20230071353A1 · Mar 9, 2023
References Cited (93)
US 7542969B1 · Rappaport et al. · 2009 [cited by applicant]
US 9514405B2 · Chen et al. · 2016 [cited by applicant]
US 9953095B1 · Scott et al. · 2018 [cited by applicant]
US 10360507B2 · Aravamudan et al. · 2019 [cited by applicant]
US 11062218B2 · Aravamudan et al. · 2021 [cited by applicant]
US 11487902B2 · Ardhanari et al. · 2022 [cited by applicant]
US 11545242B2 · Aravamudan · 2023 [cited by applicant]
US 20040013302A1 · Ma et al. · 2004 [cited by applicant]
US 20070039046A1 · Van Dijk et al. · 2007 [cited by applicant]
US 20080118150A1 · Balakrishnan et al. · 2008 [cited by applicant]
US 20080243825A1 · Staddon et al. · 2008 [cited by applicant]
US 20090116736A1 · Neogi et al. · 2009 [cited by applicant]
US 20110255788A1 · Duggan et al. · 2011 [cited by applicant]
US 20110307460A1 · Vadlamani et al. · 2011 [cited by applicant]
US 20130132331A1 · Kowalczyk et al. · 2013 [cited by applicant]
US 20150161413A1 · Calem et al. · 2015 [cited by applicant]
US 20150254555A1 · Williams, Jr. et al. · 2015 [cited by applicant]
US 20160105402A1 · Soon-Shiong et al. · 2016 [cited by applicant]
US 20160247307A1 · Stoop et al. · 2016 [cited by applicant]
US 20170032243A1 · Corrado et al. · 2017 [cited by applicant]
US 20170061326A1 · Talathi et al. · 2017 [cited by applicant]
US 20170091391A1 · LePendu · 2017 [cited by applicant]
US 20180060282A1 · Kaljurand · 2018 [cited by applicant]
US 20180212971A1 · Costa · 2018 [cited by applicant]
US 20190319982A1 · Durand et al. · 2019 [cited by applicant]
US 20190354883A1 · Aravamudan et al. · 2019 [cited by applicant]
US 20200082270A1 · Gu · 2020 [cited by examiner]
US 20200402625A1 · Aravamudan et al. · 2020 [cited by applicant]
US 20210019287A1 · Prasad et al. · 2021 [cited by applicant]
US 20210224264A1 · Barve et al. · 2021 [cited by applicant]
US 20210248268A1 · Ardhanari et al. · 2021 [cited by applicant]
US 20220050921A1 · LaFever et al. · 2022 [cited by applicant]
US 20220138599A1 · Aravamudan et al. · 2022 [cited by applicant]
US 20240320348A1 · Williams · 2024 [cited by examiner]
CN 105938495A · 2016 [cited by applicant]
JP 2019536178A · 2019 [cited by applicant]
WO WO2015084759A1 · 2015 [cited by applicant]
WO WO2015149114 · 2015 [cited by applicant]
WO WO2018057945 · 2018 [cited by applicant]
WO WO2020257783 · 2020 [cited by applicant]
WO WO2021011776 · 2021 [cited by applicant]
WO WO2021146694A1 · 2021 [cited by applicant]
WO WO2021178689 · 2021 [cited by applicant]
AMD Secure Encrypted Virtualization (SEV), https://developer.amd.com/sev/, accessed Sep. 23, 2020 (5 pages). [cited by applicant]
Arora, S. et al., “A Simple But Tough-To-Beat Baseline for Sentence Embeddings”, ICLR, 2017 (16 pages). [cited by applicant]
AWS Key Management Service (KMS), https://aws.amazon.com/kms, accessed Jan. 20, 2021 (3 pages). [cited by applicant]
AWS Key Management Service (KMS), https://aws.amazon.com/kms, accessed Sep. 23, 2020 (6 pages). [cited by applicant]
Bartunov, S. et al., “Breaking Sticks And Ambiguities With Adaptive Skip-Gram”, retrieved online from URL :< https://arxiv.org/pdf/1502.07257.pdf>, [cs.CL], Nov. 15, 2015 (15 pages). [cited by applicant]
Bojanowski, P. et al., “Enriching Word Vectors with Subword Information”, retrieved online from URL :< https://arxiv.org/pdf/1607.04606.pdf>, [cs.CL], Jun. 19, 2017 (12 pages). [cited by applicant]
Confidential Computing Consortium, “What is the Confidential Computing Consortium?”, https://confidentialcomputing.io, accessed Sep. 24, 2020 (2 pages). [cited by applicant]
De Guzman, C.G. et al., “Hematopoietic Stem Cell Expansion and Distinct Myeloid Developmental Abnormalities in a Murine Model of the [cited by applicant]
Desagulier, G., “ A lesson from associative learning: asymmetry and productivity in multiple-slot constructions”, Corpus Linguistic and Linguistic Theory, 12(2): 173-219, 2016, submitted Aug. 13, 2015, <http://www.degru… [cited by applicant]
Devlin, J. et al., “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding”, arXiv:1810.04805v2 [cs.CL], May 24, 2019 (16 pages). [cited by applicant]
Divatia, A., “The Fact and Fiction of Homomorphic Encryption”, Dark Reading, www.darkreading.com/attacks-breaches/the-fact-and-fiction-of-homomorphic-encryption/a/d-id/1333691, Jan. 22, 2019 (3 pages). [cited by applicant]
Dwork, C., “Differential Privacy: A Survey of Results”, Lecture Notes in Computer Science, vol. 4978, pp. 1-19, 2008 (19 pages). [cited by applicant]
Ferraiuolo, A. et al., “Komodo: Using verification to disentangle secure-enclave hardware from software”, SOSP '17, Shanghai, China, pp. 287-305, Oct. 28, 2017 (19 pages). [cited by applicant]
Garten, Y. et al., “Pharmspresso: a text mining tool for extraction of pharmacogenomic concepts and relationships from full text”, BMC Bioinformatics, 10(Suppl. 2):S6, Feb. 5, 2009 (9 pages). [cited by applicant]
Genkin, D. et al., “Privacy in Decentralized Cryptocurrencies”, Communications of the ACM, 61(6):78-88, Jun. 2018 (11 pages). [cited by applicant]
Hageman, G.S. et al., “A common haplotype in the complement regulatory gene factor H ( [cited by applicant]
Ikeda, T. et al., “Anticorresponding mutations of the [cited by applicant]
Intel, “What is Intel® SGX?”, http://www.intel.com/content/www/us/en/architecture-and-technology/software-guard-extensions.html, accessed Sep. 23, 2020 (8 pages). [cited by applicant]
International Preliminary Report on Patentability issued in International Application No. PCT/US2021/020906, Sep. 15, 2022 (9 pages). [cited by applicant]
International Search Report and Written Opinion issued by the European Patent Office as International Searching Authority in International Application PCT/US2017/053039, dated Dec. 20, 2017 (15 pages). [cited by applicant]
International Search Report and Written Opinion issued by the U.S. Patent and Trademark Office as International Searching Authority issued in International Application No. PCT/US21/20906, dated May 19, 2021 (10 pages). [cited by applicant]
International Search Report and Written Opinion issued by U.S. Patent and Trademark Office as International Searching Authority in International Application No. PCT/US20/42336, dated Sep. 30, 2020 (10 pages). [cited by applicant]
International Search Report and Written Opinion issued by U.S. Patent and Trademark Office as International Searching Authority, for International Application No. PCT/US20/38987, dated Nov. 9, 2020 (26 pages). [cited by applicant]
Joulin, A. et al., “Bag of Tricks for Efficient Text Classification”, retrieved online from URL:<https://arXiv.org/pdf/1607.01759v3.pdf>, [cs.CL], Aug. 9, 2016 (5 pages). [cited by applicant]
Kiros, R. et al., “Skip-Thought Vectors”, retrieved online from URL:<https://arXiv.org/abs/1506.06726v.1>, [cs.CL], Jun. 22, 2015 (11 pages). [cited by applicant]
Kolte, P. “Why Is Homomorphic Encryption Not Ready For Primetime?”, Baffle, https://baffle.io/blog/why-is-homomorphic-encryption-not-ready-for-primetime/, Mar. 17, 2017 (4 pages). [cited by applicant]
Korger, C., “Clustering of Distributed Word Representations and its Applicability for Enterprise Search”, Doctoral Thesis, Dresden University of Technology, Faculty of Computer Science, Institute of Software and Multime… [cited by applicant]
Kutuzov, A. et al., “Cross-lingual Trends Detection for Named Entities in News Texts with Dynamic Neural Embedding Models”, Proceedings of the NewsIR'16 Workshop at ECIR, Padua, Italy, Mar. 20, 2016 (6 pages). [cited by applicant]
Le, Q. et al., “Distributed Representations of Sentences and Documents”, Proceedings of the 31st International Conference of Machine Learning, Beijing, China, vol. 32, 2014 (9 pages). [cited by applicant]
Li, H. et al., “Cheaper and Better: Selecting Good Workers for Crowdsourcing,” retrieved online from URL: https://arXiv.org/abs/1502.00725v.1, pp. 1-16, Feb. 3, 2015 (16 pages). [cited by applicant]
Ling, W et al., “Two/Too Simple Adaptations of Word2Vec for Syntax Problems”, retrieved online from URL:<https://cs.cmu.edu/˜lingwang/papers/naacl2015.pdf>, 2015 (6 pages). [cited by applicant]
Maxwell, K.N. et al., “Adenoviral-mediated expression of Pcsk9 in mice results in a low-density lipoprotein receptor knockout phenotype”, PNAS, 101(18):7100-7105, May 4, 2004 (6 pages). [cited by applicant]
Mikolov, T. et al., “Distributed Representations for Words and Phrases and their Compositionality”, retrieved online from URL:https://arXiv.org/abs/1310.4546.v1 [cs.CL], Oct. 16, 2013 (9 pages). [cited by applicant]
Mikolov, T. et al., “Efficient Estimation of Word Representations in Vector Space”, retrieved online from URL: https://arXiv.org/abs/1301.3781v3 [cs.CL] Sep. 7, 2013 (12 pages). [cited by applicant]
Murray, K., “A Semantic Scan Statistic for Novel Disease Outbreak Detection”, Master's Thesis, Carnegie Mellon University, Aug. 16, 2013 (68 pages). [cited by applicant]
Neelakantan, A., et al., “Efficient Non-parametric Estimation of Multiple Embeddings per Word in Vector Space,” Department of Computer Science, University of Massachusetts, (2015) (11 pages). [cited by applicant]
Pennington, J. et al., “GloVe: Global Vectors for Word Representation”, retrieved online from URL:<https://nlp.stanford.edu/projects/glove.pdf>, 2014 (12 pages). [cited by applicant]
Rajagopalan, H., et al., “Tumorigenesis: RAF/RAS oncogenes and mismatch-repair status”, Nature, 418:934, Aug. 29, 2002 (1 page). [cited by applicant]
Rajasekharan, “Unsupervised NER using BERT”, [accessed Jun. 10, 2021], Toward Data Science, <URL: https://towardsdatascience.com/unsupervised-ner-using-bert-2d7af5f90b8a>, Feb. 28, 2020 (27 pages). [cited by applicant]
Shamir, A. “How to Share a Secret”, Communications of the ACM, 22(11):612-613, Nov. 1979 (2 pages). [cited by applicant]
Shweta, Fnu et al., “Augmented Curation of Unstructured Clinical Notes from a Massive EHR System Reveals Specific Phenotypic Signature of Impending COVID-19 Diagnosis”, https://www.medrxiv.org/content/10.1101/2020.04.19… [cited by applicant]
Van Mulligen, E.M. et al., “The EU-ADR corpus: Annotated drugs, diseases, targets, and their relationships”, Journal of Biomedical Informatics, 45:879-884, published online Apr. 25, 2012 (6 pages). [cited by applicant]
Wieting, J. et al., “Revisiting Recurrent Networks for Paraphrastic Sentence Embeddings”, retrieved online from URL:<https://arXiv.org/pdf/1705.00364v1.pdf>, [cs.CL], Apr. 30, 2017 (12 pages). [cited by applicant]
Yao, Z. et al., “Dynamic Word Embeddings for Evolving Semantic Discovery”, WSDM 2018, Marina Del Rey, CA, USA, Feb. 5-9, 2018 (9 pages). [cited by applicant]
Zuccon, G., et al., “Integrating and Evaluating Neural Word Embeddings in Information Retrieval”, ADCS, Parramatta, NSW, Australia, Dec. 8-9, 2015 (8 pages). [cited by applicant]
Daniel Genkin et al: “Privacy in decentralized cryptocurrencies”, Communications of the ACM, Association for Computing Machinery, Inc, United States, vol. 61, No. 6, May 23, 2018 (May 23, 2018), pp. 78-88, XP058407634, … [cited by applicant]
Bradley Malin: “Guidance Regarding Methods for De-identification of Protected Health Information in Accordance with the Health Insurance Portability and Accountability Act (HIPAA) Privacy Rule”, Nov. 26, 2012 (Nov. 26, … [cited by applicant]
Devlin Jacob et al: “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding”, Oct. 11, 2018 (Oct. 11, 2018), XP055968792, Retrieved from the Internet: URL:https://www.arxiv.org/pdf/1810.04805vl… [cited by applicant]
Rajasekharan Ajit: “Unsupervised NER using BERT”, Towards Data Science, Feb. 28, 2020 (Feb. 28, 2020), XP093054321, Retrieved from the Internet: URL:https://towardsdatascience.com/unsuper vised-ner-using-bert-2d7 af5f90… [cited by applicant]
European Search Report; EP 24 17 9988; By: Maenpaa, Jari; Sep. 2, 2024. [cited by applicant]