Method for verifying model update
The federated learning system includes a moderator and client devices. Each client device performs a method for verifying model update as follows: receiving a hash function and a general model; training a client model according to the general model and raw data; calculating a difference as an update parameter between the general model and the client model, sending the update parameter to the moderator; inputting the update parameter to the hash function to generate a hash value; sending the hash value to other client devices, and receiving other hash values; summing all the hash values to generate a trust value; receiving an aggregation parameter calculated according to the update parameters; inputting the aggregation parameter to the hash function to generate a to-be-verified value; and updating the client model according to the aggregation parameter when the to-be-verified value equals the trust value.
1. A method for verifying model update applicable to a federated learning system having a moderator and a plurality of client devices, wherein the method comprises a plurality of steps performed by each of the plurality of client devices and the plurality of steps comprises:
receiving a hash function and a general model from the moderator;
training a client model according to the general model and raw data;
calculating a difference between the general model and the trained client model as an update parameter;
sending the update parameter to the moderator;
inputting the update parameter to the hash function to generate a hash value;
sending the hash value to the plurality of client devices other than the client device performing the method and receiving a plurality of other hash values;
summing the hash value and the plurality of other hash values to generate a trust value;
receiving an aggregation parameter from the moderator, wherein the aggregation parameter is calculated by the moderator according to the update parameter from each of the plurality of client devices;
inputting the aggregation parameter to the hash function to generate a to-be-verified value;
comparing the to-be-verified value and the trust value, and updating the trained client model according to the aggregation parameter when the to-be-verified value equals the trust value.
2. The method for verifying model update of claim 1 , wherein the hash function is additively homomorphic hash function.
3. The method for verifying model update of claim 1 , wherein calculating the difference between the general model and the trained client model as the update parameter by each of the plurality of client devices comprises:
calculating the difference between the general model and the trained client model; and
performing a quantization procedure to convert the difference from a floating-point type to an integer type.
4. The method for verifying model update of claim 1 , wherein:
calculating the difference between the general model and the trained client model as the update parameter by each of the plurality of client devices comprises: encrypting the difference according to a public key to generate the update parameter by each of the plurality of client devices; and
updating the trained client model according to the aggregation parameter comprises: decrypting the aggregation parameter according to a private key and updating the trained client model according to the decrypted aggregation parameter by each of the plurality of client devices.
5. The method for verifying model update of claim 4 , further comprising: performing a KeyGen protocol of a threshold additive homomorphic encryption to generate the public key and the private key by each of the plurality of client devices.
6. A method for verifying model update applicable to a federated learning system having a plurality of moderators and a plurality of client devices, wherein the method comprises a plurality of steps performed by each of the plurality of client devices and the plurality of steps comprises:
receiving a general model from the federated learning system;
training a client model according to the general model and raw data;
calculating a difference between the general model and the trained client model;
generating an update parameter by encrypting the difference with a public key;
sending the update parameter to the plurality of moderators;
receiving a plurality of aggregation parameters from the plurality of moderators, wherein each of the plurality of aggregation parameters is calculated by each of the plurality of moderators according to the update parameter;
searching a mode in the plurality of aggregation parameters;
generating a decrypted result according to the mode and a private key when a number of modes exceeds half of a number of aggregation parameters; and
updating the trained client model according to the decrypted result.
7. The method for verifying model update of claim 6 , wherein each of the plurality of client devices performs a KeyGen protocol of a threshold additive homomorphic encryption to generate the public key and the private key.
8. A method for verifying model update applicable to a federated learning system having a plurality of moderators and a plurality of client devices, wherein the method comprises a plurality of steps performed by each of the plurality of client devices and the plurality of steps comprises:
receiving a hash function and a general model from one of the plurality of moderators;
training a client model according to the general model and raw data;
calculating a difference between the general model and the trained client model;
generating an update parameter by encrypting the difference with a public key;
sending the update parameter to the plurality of moderators;
inputting the update parameter to the hash function to generate a hash value;
sending the hash value to the plurality of client devices other than the client device performing the method and, receiving a plurality of other hash values;
summing the hash value and the plurality of other hash values to generate a trust value;
receiving a plurality of aggregation parameters from the plurality of moderators, wherein each of the plurality of aggregation parameters is calculated by each of the plurality of moderators according to the update parameter;
inputting each of the plurality of aggregation parameters to the hash function to generate a plurality of to-be-verified values;
comparing each of the plurality of to-be-verified values and the trust value sequentially; and
when one of the plurality of to-be-verified values equal to the trust value is found, decrypting one of the plurality of aggregation parameters corresponding to the one of the plurality of to-be-verified values equal to the trust value according to a private key and updating the trained client model according to the decrypted one of the plurality of aggregation parameters.
9. The method for verifying model update of claim 8 , wherein the hash function is additively homomorphic hash function.
10. The method for verifying model update of claim 8 , wherein each of the plurality of client devices performs a KeyGen protocol of a threshold additive homomorphic encryption to generate the public key and the private key.