Performing privacy-preserving multi-party analytics on vertically partitioned local data

1. A method comprising:

computing, by a computing device at a party among a plurality of parties, a sum of local data owned by the party, wherein the local data is vertically partitioned into a plurality of data segments, and wherein each data segment comprises a non-overlapping subset of data features;

transforming, by the computing device, a cost function of a data analytics task to a gradient descent function, wherein the cost function comprises a summation of a plurality of cost function values;

initializing, by the computing device, each data segment;

decrypting, by the computing device, a collection of data shards received from the mediator using the private key corresponding to the first party;

shuffling, by the computing device, the collection of data shards that are decrypted using the private key corresponding to the first party;

transmitting, by the computing device, the shuffled collection of data shards to one of the plurality of parties and the mediator;

anonymizing, by the computing device, aggregated data shards received from a mediator, wherein the aggregated data shards comprise the plurality of cost function values for data features owned by the plurality of parties;

updating, by the computing device, local model parameters based on the aggregated data shards;

learning, by the computing device, a global analytic model based on the updated local parameters and cost function values; and

performing, by the computing device, privacy-preserving multi-party analytics on the vertically partitioned local data based on the learned global analytic model.

2. The method of claim 1 , wherein initializing each data segment comprises:

dividing, by the computing device, each data segment into at least two data shards;

recursively encrypting, by the computing device, each data shard using a plurality of public keys corresponding to the plurality of parties and a mediator; and

sharing, by the computing device, each encrypted data shard with the mediator.

3. The method of claim 1 , wherein the global analytic model comprises a plurality of features, each feature being associated with a particular party, and wherein a set of features associated with the particular party has no interdependency with features associated with another party.

4. The method of claim 3 , wherein the global analytic model comprises a linear relationship between a model parameter and a feature vector.

5. The method of claim 3 , wherein the global analytic model comprises one of: a linear regression cost function, a logistic regression cost function, a ridge regression cost function, a support vector machine (SVM) function, and a least absolute shrinkage and selection operator (LASSO) function.

6. The method of claim 1 , wherein the plurality of parties participate in data sharing under a semi-honest adversary model in which each party follows an agreed-upon protocol while attempting to acquire information about private data during computation.

7. The method of claim 1 , wherein the collection of data shards is further decrypted using other private keys corresponding to other parties in the plurality of parties following a predetermined priority order.

8. A system comprising:

a plurality of computing devices associated with a plurality of parties participating in privacy-preserving data sharing under a semi-honest adversary model, and

a mediator coupled to the plurality of parties,

wherein each of the plurality of computing devices comprises a hardware processor to:

compute, at a party among a plurality of parties, a sum of local data owned by the party, wherein the local data is vertically partitioned into a plurality of data segments, and wherein each data segment comprises a non-overlapping subset of data features;

transform a cost function of a data analytics task to a gradient descent function, wherein the cost function comprises a summation of a plurality of cost function values;

initialize each data segment;

decrypt a collection of data shards received from the mediator using the private key corresponding to the first party;

shuffle the collection of data shards that are decrypted using the private key corresponding to the first party;

transmit the shuffled collection of data shards to one of the plurality of parties and the mediator;

anonymize aggregated data shards received from the mediator, wherein the aggregated data shards comprise the plurality of cost function values for data features owned by the plurality of parties;

update local model parameters based on the aggregated data shards;

learn a global analytic model based on the updated local model parameters; and

wherein the mediator computes a global gradient for the global analytic model used by the plurality of parties to perform privacy-preserving analytics on vertically partitioned local data.

9. The system of claim 8 , wherein the hardware processor in each of the plurality of computing devices is further to:

divide each data segment into at least two data shards;

recursively encrypt each data shard using a plurality of public keys corresponding to the plurality of parties and a mediator; and

share each encrypted data shard with the mediator.

10. The system of claim 8 , wherein the global analytic model comprises a plurality of features, each feature being associated with a particular party, and wherein a set of features associated with the particular party has no interdependency with features associated with another party.

11. The system of claim 10 , wherein the global analytic model comprises a linear relationship between a model parameter and a feature vector.

12. The system of claim 10 , wherein the global analytic model comprises one of: a linear regression cost function, a logistic regression cost function, a ridge regression cost function, a support vector machine (SVM) function, and a least absolute shrinkage and selection operator (LASSO) function.

13. The system of claim 8 , wherein the plurality of parties participate in data sharing under a semi-honest adversary model in which each party follows an agreed-upon protocol while attempting to acquire information about private data during computation.

14. A non-transitory machine-readable storage medium encoded with instructions executable by at least one processor of a computing device associated with a mediator in a multi-party environment, the machine-readable storage medium comprising instructions to:

receive, from a plurality of parties, sums of local data owned by the party, wherein the data is vertically partitioned into a plurality of data segments, and wherein each data segment comprises a non-overlapping subset of data features;

transform a cost function of a data analytics task to a gradient descent function, wherein the cost function comprises a summation of a plurality of cost function values;

aggregate the sums received from the plurality of parties into aggregated data shards;

decrypt aggregated data shards using the private key corresponding to a first party;

shuffle the aggregated data shards that are decrypted using the private key corresponding to the first party;

transmit the shuffled collection of data shards to one of the plurality of parties and a mediator;

share the aggregated data shards with the plurality of parties to update local model parameters at each party such that each party learns a global analytic model, wherein the aggregated data shards comprise the plurality of cost function values for data features owned by the plurality of parties; and

compute a global gradient for the global analytic model used by the plurality of parties to perform privacy-preserving multi-party analytics on the vertically partitioned local data.

15. The non-transitory machine-readable storage medium of claim 14 , wherein the global analytic model comprises a plurality of features, each feature being associated with a particular party, and wherein a set of features associated with the particular party has no interdependency with features associated with another party.

16. The non-transitory machine-readable storage medium of claim 15 , wherein the global analytic model comprises a linear relationship between a model parameter and a feature vector.

17. The non-transitory machine-readable storage medium of claim 15 , wherein the global analytic model comprises one of: a linear regression cost function, a logistic regression cost function, a ridge regression cost function, a support vector machine (SVM) function, and a least absolute shrinkage and selection operator (LASSO) function.

18. The non-transitory machine-readable storage medium of claim 14 , wherein the plurality of parties participate in data sharing under a semi-honest adversary model in which each party follows an agreed-upon protocol while attempting to acquire information from private data during computation.