Systems and methods of a novel self-serve, customer driven data platform that can automatically unify and structure the data that comes from different sources, in order to provide well defined data for any federated learning task. This platform solves a critical problem for federated learning which usually requires multiple different data sources jointly learning one model. In the real-world scenario, the assumption that most existing federated learning frameworks have, that different data owners follow the same rule or structure to save the data, usually does not hold. Our data platform is the one of the novel and heuristic ways to solve this practical problem and makes larger scale and automated industrial level federated learning achievable.

According to one embodiment, a secure computing method includes setting a coefficient selected from a ring of integers Q based on first data X, generating n pieces of first fragment data from the first data X based on the coefficient, causing a learning model held in the computing device to learn the first fragment data, generating n pieces of second fragment data from second data Z based on the coefficient, performing, by each of the n computing devices, inference based on the second fragment data using the learning model, and obtaining decoded data dec by decoding k pieces of inference result data. The coefficient is set to make each of the n pieces of first fragment data less than a maximum value of the ring of integers Q.

The disclosed systems, and methods are directed to a method for Privacy Preserving Inference (PPI) comprising receiving a first set of matrix information from a client device, generating k.sub.c−1 matrices by operating a first CSPRNG associated with the server with k.sub.c−1 seeds, computing inferences from the set of k.sub.c matrices, generating a matrix S.sub.s, generating k.sub.s−1 random matrices, computing a matrix Y.sub.k.sub.s in accordance with the inference matrix Y, the matrix S.sub.s and the k.sub.s−1 random matrices, transmitting a second set of matrix information to the client device, the second set of matrix information includes k.sub.s−1 seeds corresponding to the k.sub.s−1 random matrices and the matrix Y.sub.k.sub.s, receiving a matrix U from the client device, and computing an inference value y from the matrix U.

Some implementations disclosed herein enable matching identifiers across multiple sources. This may involve adding a unique attribute (e.g., anonymous unique homomorphic identifiers) and/or using randomization to enable comparing data from multiple sources, while also maintaining data privacy. In one example, matches across multiple sources are identified, for example, identifying that there are 100 user identifiers that are in private data sets of three different sources. Such matching may be used to enable private, multi-touch attribution. In another example, techniques are used to determine that data maintained by one source is not also within other sources (e.g., identifying that there are 200 user identifiers that are in data from a first source but not in data from a second source and not in data from a third source. Such determinations may be used to generate control group data that does not match data from other sources.

A method of performing cryptographic key exchange while overcoming a malicious adversary party using a multi-party computation (MPC) process performed by the multiple parties, where the parties hold initial shares of a secret used an as exponentiation of the key exchange, where the parties do not reveal the initial shares during the entire process, and where arithmetical computations are performed on the initial shares and on random values outputted during MPC processes.

A method may include receiving, from a first trusted authority, a secret key specific to a party for use in posting to a blockchain. The method may also include receiving, from a second trusted authority, a correlated randomness component specific to the party and associated with a given temporal segment. The method may additionally include generating a party-generated randomized mask, and computing, using an input from the party, the correlated randomness component, and the party-generated randomized mask in a non-interactive multi-party computation (NIMPC), an NIMPC-encrypted input associated with the party for the given temporal segment. The method may also include encrypting the NIMPC-encrypted input according to a blockchain encryption algorithm to yield a ciphertext, and submitting the ciphertext to a block associated with the given temporal segment in a blockchain.

This document describes techniques that prevent the sharing or leakage of user information. In one aspect, a method includes receiving, by a first MPC server, a request for a selection criterion of at least one interest group to which a user of a client device belongs. The received request does not reveal an identifier of the client device to the first MPC server. In response to receiving the request, the first MPC server determines a set of ordered selection criterion of the at least one interest group retrieved from a cache of the first MPC server. The set of ordered selection criterion is transformed into a set of key/value pairs secured from being revealed by the second MPC server. The first MPC server transmits the set of key/value pairs to the second MPC server with data that enables the second MPC server to identify a key having a highest value.

Systems and methods for improved distributed symmetric cryptography are disclosed. A client computer may communicate with a number of cryptographic devices in order to encrypt or decrypt data. Each cryptographic device may possess a secret share and a verification share, which may be used in the process of encrypting or decrypting data. The client computer may generate a commitment and transmit the commitment to the cryptographic devices. Each cryptographic device may generate a partial computation based on the commitment and their respective secret share, and likewise generate a partial signature based on the commitment and their respective verification share. The partial computations and partial signatures may be transmitted to the client computer. The client computer may use the partial computations and partial signatures to generate a cryptographic key and verification signature respectively. The client computer may use the cryptographic key to encrypt or decrypt a message.

A device participates in secret sharing-based MPC. Original data can be restored by combining a share of the device with a corresponding share of another device. The device includes means for acquiring random number and means for updating a share of the device on the basis of the acquired random number. A method for updating by the updating means is designed to perform update in a manner that a share of the device updated on the basis of the acquired random number is combined with the corresponding share of the other device updated on the basis of the random number to cancel an influence of the random number and restore the original data.

A method is disclosed. An authentication node may receive a plurality of encrypted match values, wherein the plurality of encrypted match values were formed by a plurality of worker nodes that compare a plurality of encrypted second biometric template parts derived from a second biometric template to a plurality of encrypted first biometric template parts derived from a first biometric template. The authentication node may decrypt the plurality of encrypted match values resulting in a plurality of decrypted match values. The authentication node may then determine if a first biometric template matches the second biometric template using the plurality of decrypted match values. An enrollment node may be capable of enrolling a biometric template and storing encrypted biometric template parts at worker nodes.