Tokenized federated learning
One embodiment of the invention provides a method for federated learning (FL) comprising training a machine learning (ML) model collaboratively by initiating a round of FL across data parties. Each data party is allocated tokens to utilize during the training. The method further comprises maintaining, for each data party, a corresponding data usage profile indicative of an amount of data the data party consumed during the training and a corresponding participation profile indicative of an amount of data the data party provided during the training. The method further comprises selectively allocating new tokens to the data parties based on each participation profile maintained, selectively allocating additional new tokens to the data parties based on each data usage profile maintained, and reimbursing one or more tokens utilized during the training to the data parties based on one or more measurements of accuracy of the ML model.
1 . A method for federated learning, comprising:
initiating the federated learning across a plurality of data parties, wherein the federated learning comprises a plurality of epochs;
for a given epoch of the plurality of epochs:
ranking one or more provider data parties comprising the data parties based on a plurality of data received during the given epoch, a resource usage of the one or more provider data parties, and a reputation of the one or more provider data parties;
allocating one or more new tokens for use during training to the one or more provider data parties based on the ranking;
training, by a plurality of consumer data parties comprising the data parties, a machine learning model on the plurality of data received during the given epoch from the one or more provider data parties; and
reimbursing one or more tokens utilized during the training to the one or more provider data parties based on an accuracy of the machine learning model in the given epoch relative to an accuracy of the machine learning model in an earlier epoch.
2 . The method of claim 1 , further comprising:
counting, for each of the consumer data parties, a corresponding number of tokens utilized by the consumer data party during the training, wherein the corresponding number of tokens is indicative of an amount of data the consumer data party consumed to train the machine learning model.
3 . The method of claim 2 , further comprising:
sorting the plurality of consumer data parties into a ranked list based on, for each data party, the corresponding number of tokens utilized by the consumer data party during the training; and
allocating the new tokens based on the ranked list, wherein a consumer data party ranked first in the ranked list is allocated a most amount of the new tokens, and a consumer data party ranked last in the ranked list is allocated a fewest amount of the new tokens.
4 . The method of claim 1 , further comprising:
counting, for each provider data party, a corresponding number of tokens utilized by the provider data party during the training, wherein the corresponding number of tokens is indicative of an amount of data the provider data party provided to train the machine learning model.
5 . The method of claim 1 , further comprising:
for each epoch of the federated learning:
tracking a first measurement representing a maximum accuracy of the machine learning model until the epoch of the federated learning;
tracking a second measurement representing the accuracy of the machine learning model in the epoch of the federated learning; and
determining an improvement in the accuracy of the machine learning model based on the first measurement and the second measurement.
6 . The method of claim 5 , wherein reimbursing one or more tokens utilized during the training to one or more of the plurality of data parties based on one or more measurements of accuracy of the machine learning model comprises:
determining an amount of tokens utilized during the training to reimburse based on the improvement in the accuracy of the machine learning model.
7 . A system for federated learning, comprising:
at least one processor; and
a non-transitory computer readable memory device storing instructions that when executed by the at least one processor causes the at least one processor to perform operations including:
initiating the federated learning across a plurality of data parties, wherein the federated learning comprises a plurality of epochs;
for a given epoch of the plurality of epochs:
ranking one or more provider data parties comprising the data parties based on a plurality of data received during the given epoch, a resource usage of the one or more provider data parties, and a reputation of the one or more provider data parties;
allocating one or more new tokens for use during training to the one or more provider data parties based on the ranking;
training, by a plurality of consumer data parties comprising the data parties, a machine learning model on the plurality of data received during the given epoch from the one or more provider data parties; and
reimbursing one or more tokens utilized during the training to the one or more provider data parties based on an accuracy of the machine learning model in the given epoch relative to an accuracy of the machine learning model in an earlier epoch.
8 . The system of claim 7 , wherein the operations further comprise:
counting, for each of the consumer data parties, a corresponding number of tokens utilized by the consumer data party during the training, wherein the corresponding number of tokens is indicative of an amount of data the consumer data party consumed to train the machine learning model.
9 . The system of claim 8 , further comprising:
sorting the plurality of consumer data parties into a ranked list based on, for each data party, the corresponding number of tokens utilized by the consumer data party during the training; and
allocating the new tokens based on the ranked list, wherein a consumer data party ranked first in the ranked list is allocated a most amount of the new tokens, and a consumer data party ranked last in the ranked list is allocated a fewest amount of the new tokens.
10 . The system of claim 7 , wherein the operations further comprise:
counting, for each provider data party, a corresponding number of tokens utilized by the provider data party during the training, wherein the corresponding number of tokens is indicative of an amount of data the provider data party provided to train the machine learning model.
11 . The system of claim 7 , wherein the operations further comprise:
for each epoch of the federated learning:
tracking a first measurement representing a maximum accuracy of the machine learning model until the epoch of the federated learning;
tracking a second measurement representing the accuracy of the machine learning model in the epoch of the federated learning; and
determining an improvement in the accuracy of the machine learning model based on the first measurement and the second measurement.
12 . The system of claim 11 , wherein reimbursing the one or more tokens utilized during the training to one or more of the plurality of data parties based on one or more measurements of accuracy of the machine learning model comprises:
determining an amount of tokens utilized during the training to reimburse based on the improvement in the accuracy of the machine learning model.
13 . A computer program product for federated learning, the computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor to cause the processor to:
initiate the federated learning across a plurality of data parties, wherein the federated learning comprises a plurality of epochs;
for a given epoch of the plurality of epochs:
rank one or more provider data parties comprising the data parties based on a plurality of data received during the given epoch, a resource usage of the one or more provider data parties, and a reputation of the one or more provider data parties;
allocate one or more new tokens for use during training to the one or more provider data parties based on the ranking;
train, by a plurality of consumer data parties comprising the data parties, a machine learning model on the plurality of data received during the given epoch from the one or more provider data parties; and
reimburse one or more tokens utilized during the training to the one or more provider data parties based on an accuracy of the machine learning model in the given epoch relative to an accuracy of the machine learning model in an earlier epoch.
14 . The computer program product of claim 13 , wherein the program instructions executable by the processor further cause the processor to:
count, for each of the consumer data parties, a corresponding number of tokens utilized by the consumer data party during the training, wherein the corresponding number of tokens is indicative of an amount of data the consumer data party consumed to train the machine learning model.
15 . The computer program product of claim 14 , further comprising:
sorting the plurality of consumer data parties into a ranked list based on, for each data party, the corresponding number of tokens utilized by the consumer data party during the training; and
allocating the new tokens based on the ranked list, wherein a consumer data party ranked first in the ranked list is allocated a most amount of the new tokens, and a consumer data party ranked last in the ranked list is allocated a fewest amount of the new tokens.
16 . The computer program product of claim 13 , wherein the program instructions executable by the processor further cause the processor to:
count, for each provider data party, a corresponding number of tokens utilized by the provider data party during the training, wherein the corresponding number of tokens is indicative of an amount of data the provider data party provided to train the machine learning model.
17 . The computer program product of claim 13 , wherein the program instructions executable by the processor further cause the processor to:
for each epoch of the federated learning:
track a first measurement representing a maximum accuracy of the machine learning model until the epoch of the federated learning;
track a second measurement representing the accuracy of the machine learning model in the epoch of the federated learning;
determine an improvement in the accuracy of the machine learning model based on the first measurement and the second measurement; and
determine an amount of tokens utilized during the training to reimburse based on the improvement in the accuracy of the machine learning model.