Systems and methods for federated learning using distributed messaging with entitlements for anonymous computation and secure delivery of model

1. A method for federated learning using distributed messaging, comprising:

generating, by an aggregator node in a distributed computer network, an aggregator node public/private key pair;

communicating, by the aggregator node, the aggregator node public key to a plurality of participant nodes in the distributed computer network;

receiving, by the aggregator node and from each of the participant nodes, a message comprising a local machine learning (ML) model encrypted with a participant node private key and with the aggregator node public key, and a participant node public key corresponding to the participant node private key, the participant node public key encrypted with the aggregator node public key;

decrypting, by the aggregator node, the local ML models and the participant node public keys using the aggregator node public key;

decrypting, by the aggregator node, the local ML models using the participant node public keys;

generating, by the aggregator node, an aggregated ML model based on the local ML models;

encrypting, by the aggregator node and with each participant node public key, the aggregated ML model; and

communicating, by the aggregator node, the encrypted ML models to all participant nodes;

wherein each participant node decrypts one of the encrypted ML models using its participant node private key, and modifies its local ML model with the aggregated ML model.

2. The method of claim 1 , wherein the aggregator node communicates the aggregator node public key to the plurality of participant nodes in the distributed computer network using a first plain communication channel, wherein the first plain communication channel is configured to allow the aggregator node to write and to allow the participant nodes to read.

3. The method of claim 1 , wherein the aggregator node further communicates a request for the local ML models to the participant nodes.

4. The method of claim 1 , wherein the participant nodes remove their respective participant node public and private keys after decrypting the aggregated ML model.

5. The method of claim 1 , wherein the encrypted local ML models are received over a first encrypted communication channel, wherein the first encrypted communication channel is configured to allow the participant nodes to write and to allow the aggregator node to read.

6. The method of claim 1 , wherein the aggregator node communicates the encrypted ML models to all participant nodes over a second encrypted communication channel, wherein the second encrypted communication channel is configured to allow the aggregator node to write and to allow the participant nodes to read.

7. A method for federated learning using distributed messaging, comprising:

receiving, by a participant node in a distributed computer network comprising a plurality of participant nodes and from an aggregator node in the distributed computer network, an aggregator node public key;

generating, by the participant node, a participant node public/private key pair;

encrypting, by the participant node, a local machine learning (ML) model encrypted with the participant node private key;

encrypting, by the participant node, the encrypted local ML model and the participant node public key with the aggregator node public key;

communicating, by the participant node, the encrypted ML model and the encrypted participant node public key to the aggregator node;

receiving, from the aggregator node, a plurality of encrypted aggregated ML models, each aggregated ML models encrypted with a private key for one of the plurality of participant nodes;

decrypting, by the participant node, one of the encrypted aggregated ML model using the participant node private key; and

updating, by the participant nodes, the local ML model based on the aggregated ML model.

8. The method of claim 7 , wherein the aggregator node public key is received using a first plain communication channel, wherein the first plain communication channel is configured to allow the aggregator node to write and to allow the participant nodes to read.

9. The method of claim 7 , further comprising:

receiving, by the participant node, a request for the local ML models from the aggregator node.

10. The method of claim 7 , further comprising:

removing, by the participant node, the participant node public and private keys after decrypting the aggregated ML model.

11. The method of claim 7 , wherein the encrypted local ML model is communicated over a first encrypted communication channel, wherein the first encrypted communication channel is configured to allow the participant nodes to write and to allow the aggregator node to read.

12. The method of claim 7 , wherein the encrypted ML model is received over a second encrypted communication channel, wherein the second encrypted communication channel is configured to allow the aggregator node to write and to allow the participant nodes to read.

13. A distributed computing system, comprising:

an aggregator node;

a plurality of participant nodes;

a first plain communication channel that is configured to allow the aggregator node to write and to allow the participant nodes to read;

a first encrypted communication channel that is configured to allow the participant nodes to write and to allow the aggregator node to read; and

a second encrypted communication channel that is configured to allow the aggregator node to write and to allow the participant nodes to read;

wherein:

the aggregator node generates an aggregator node public/private key pair;

the aggregator node communicates the aggregator node public key to the plurality of participant nodes over a first plain communication channel;

each of the participant nodes generates a participant node public/private key pair;

each of the participant nodes encrypts a local machine learning (ML) model encrypted with the participant node private key;

each of the participant nodes encrypts the encrypted local ML model and the participant node public key with the aggregator node public key;

each of the participant nodes communicates the encrypted ML model and the encrypted participant node public key to the aggregator node over the first encrypted communication channel;

the aggregator node decrypts the encrypted local ML models and the participant node public keys using the aggregator node public key;

the aggregator node decrypts the local ML models using the participant node public keys;

the aggregator node generates an aggregated ML model based on the local ML models;

the aggregator node encrypts the aggregated ML model with each participant node public key;

the aggregated ML model communities the encrypted ML models to all participant nodes over the second encrypted communication channel;

each of the participant nodes receives the plurality of encrypted aggregated ML models from the aggregator node;

each of the participant nodes decrypts one of the encrypted aggregated ML model using the participant node private key; and

each of the participant nodes updates its local ML model based on the aggregated ML model.

14. The system of claim 13 , wherein each of the participant nodes removes its participant node public and private keys after decrypting the aggregated ML model.

15. The system of claim 13 , wherein the aggregator node further communicates a request for the local ML models to the participant nodes.