US 12,278,907 B2
Apparatus for secure multiparty computations for machine-learning
Shriphani Palakodety, San Jose, CA (US); Patrick Grinaway, Brooklyn, NY (US); Galana Gebisa, Mountain View, CA (US); Volkmar Frinken, San Jose, CA (US); Jayavanth Shenoy, Mountain View, CA (US); and Guha Jayachandran, Cupertino, CA (US)
Assigned to Onai Inc., San Jose, CA (US)
Filed by Onai Inc., San Jose, CA (US)
Filed on Feb. 18, 2022, as Appl. No. 17/675,663.
Prior Publication US 2023/0269090 A1, Aug. 24, 2023
Int. Cl. H04L 9/32 (2006.01); G06N 20/00 (2019.01)
CPC H04L 9/3247 (2013.01) [G06N 20/00 (2019.01); H04L 2209/46 (2013.01)] 20 Claims
OG exemplary drawing
 
1. An apparatus for secure multiparty computations for machine-learning,
the apparatus comprising at least a processor; and
a memory communicatively connected to the at least a processor, the memory containing instructions configured the at least a processor to:
submit a secure multiparty computation request onto an immutable sequential listing, wherein the secure multiparty computation request comprises:
a contingent payment; and
an authenticity commitment of a first private dataset, wherein the authenticity commitment further comprises a cryptographic commitment to data indicating reliability of the first private dataset;
receive at least a participant commitment from each participating device of a quorum of participating devices, wherein each at least a participant commitment further comprises a cryptographic commitment representing an intention of the participating device to partake in a secure multiparty computation corresponding to the secure multiparty computation request;
generate a first localized model as a function of the first private dataset wherein the first localized model is further generated using a localized machine learning model trained with inputs of exemplary first private datasets correlated to outputs of exemplary consolidated private datasets, wherein the machine learning model further comprises generating a local gradient on the first private dataset to output the first localized model;
perform a joint training protocol as a function of the first localized model and a second localized model from the quorum of participating devices, wherein the joint training protocol comprises generating a joint training datum, wherein performing the joint training protocol comprises generating a multiparty dataset machine learning model configured to utilize an ensemble method to combine the first localized model and the second localized model; and
verify the joint training datum, wherein verifying the joint training datum includes receiving a plurality of proofs from each participating device of the quorum of participating devices, wherein verifying the joint training datum comprises validating the generated multiparty dataset machine learning model as a function of an accuracy threshold.