Blockchain-based verification of machine learning
Disclosed are systems and method for machine learning and blockchain-based anti-discrimination validation. The described techniques uses a machine learning model to generate a numerical determination associated with a first person based on an input data set associated with the first person. The numerical determination is further based on a corrective module configured to compensate for prohibited discrimination by the machine learning model. The technique generates a blockchain transaction data structure comprising a state of the machine learning model at the time of generating the numerical determination, a copy of the input data set associated with the person, and an indication of a correction by the machine learning model. The blockchain transaction data structure is recorded or published in a blockchain network.
1. A computer-implemented method for performing machine learning, comprising:
generating, using a machine learning model, a numerical determination associated with a first person based on an input data set associated with the first person,
wherein the generating comprises compensating for prohibited discrimination by the machine learning model using a corrective module that compensates by:
determining whether an initial output by the machine learning model violates a nondiscrimination policy based on an evaluation data set; and
modifying the initial output to generate the numerical determination using a corrective algorithm in response to determining that the initial output violates the nondiscrimination policy;
generating a blockchain transaction data structure comprising a state of the machine learning model at the time of generating the numerical determination, a copy of the input data set associated with the first person, and an indication of a correction by the machine learning model; and
recording the blockchain transaction data structure in a blockchain network.
2. The method of claim 1 , further comprising:
receiving a second blockchain transaction data structure from the blockchain network; and
validating compliance of a nondiscrimination policy based on the received second transaction.
3. The method of claim 1 , wherein the indication of the correction comprises a copy of the initial output by the machine learning model prior to correction, a state of the corrective algorithm, and a result of the corrective algorithm.
4. The method of claim 1 , wherein the machine learning module is trained using a corrective input data set to achieve a requested level of non-discrimination by the machine learning model.
5. The method of claim 4 , wherein the indication of the correction comprises an indication that the machine learning model has been verified using an evaluation data set subsequent to training using the corrective input data set.
6. A system for performing machine learning, comprising:
a memory configured to store computer-executable instructions; and
a hardware processor coupled to the memory and configured to execute the instructions to:
generate, using a machine learning model, a numerical determination associated with a first person based on an input data set associated with the first person,
wherein the generating comprises compensating for prohibited discrimination by the machine learning model using a corrective module that compensates by:
determining whether an initial output by the machine learning model violates a nondiscrimination policy based on an evaluation data set; and
modifying the initial output to generate the numerical determination using a corrective algorithm in response to determining that the initial output violates the nondiscrimination policy;
generate a blockchain transaction data structure comprising a state of the machine learning model at the time of generating the numerical determination, a copy of the input data set associated with the first person, and an indication of a correction by the machine learning model; and
record the blockchain transaction data structure in a blockchain network.
7. The system of claim 6 , wherein the hardware processor is further configured to:
receive a second blockchain transaction data structure from the blockchain network; and
validate compliance of a nondiscrimination policy based on the received second transaction.
8. The system of claim 6 , wherein the indication of the correction comprises a copy of the initial output by the machine learning model prior to correction, a state of the corrective algorithm, and a result of the corrective algorithm.
9. The system of claim 6 , wherein the machine learning module is trained using a corrective input data set to achieve a requested level of non-discrimination by the machine learning model.
10. The system of claim 9 , wherein the indication of the correction comprises an indication that the machine learning model has been verified using an evaluation data set subsequent to training using the corrective input data set.
11. A non-transitory computer readable medium comprising computer executable instructions for performing machine learning, including instructions for:
generating, using a machine learning model, a numerical determination associated with a first person based on an input data set associated with the first person,
wherein the generating comprises compensating for prohibited discrimination by the machine learning model using a corrective module that compensates by:
determining whether an initial output by the machine learning model violates a nondiscrimination policy based on an evaluation data set; and
modifying the initial output to generate the numerical determination using a corrective algorithm in response to determining that the initial output violates the nondiscrimination policy;
generating a blockchain transaction data structure comprising a state of the machine learning model at the time of generating the numerical determination, a copy of the input data set associated with the first person, and an indication of a correction by the machine learning model; and
recording the blockchain transaction data structure in a blockchain network.
12. The non-transitory computer readable medium of claim 11 , further comprising instructions for:
receiving a second blockchain transaction data structure from the blockchain network; and
validating compliance of a nondiscrimination policy based on the received second transaction.
13. The non-transitory computer readable medium of claim 11 , wherein the indication of the correction comprises a copy of the initial output by the machine learning model prior to correction, a state of the corrective algorithm, and a result of the corrective algorithm.
14. The non-transitory computer readable medium of claim 11 , wherein the machine learning module is trained using a corrective input data set to achieve a requested level of non-discrimination by the machine learning model.
15. The non-transitory computer readable medium of claim 14 , wherein the indication of the correction comprises an indication that the machine learning model has been verified using an evaluation data set subsequent to training using the corrective input data set.