Verifying machine learning model integrity by hashing
A system may include a plurality of subsystems, such as an identity management system configured to store information associated with digital identities, a resource management system configured to store information associated with resources of a community, a financial system configured to manage financial transactions, an infrastructure system configured to manage a digital representation of the community, a transportation system configured to manage a plurality of transportation devices of the community, an education system configured to manage the education of the community, a security system configured to secure information of the plurality of subsystems, an energy system configured to manage an electrical transmission system and an electrical generation system of the community, and a central hub system configured to manage other subsystems.
1 . A computer-implemented method, comprising:
receiving a request to generate a model token;
accessing a machine learning model;
generating a hash based on the machine learning model;
generating a token based on the machine learning model, wherein the token comprises the hash;
storing the token on a blockchain storage system;
receiving a verification request for the machine learning model comprising a user hash;
determining whether the user hash matches the hash; and
based on determining the user hash is a same as the hash, transmitting an indication that integrity of the machine learning model is verified.
2 . The computer-implemented method of claim 1 further comprising receiving a first smart contract attribute, and wherein generating the token comprises:
generating a smart contract based on the first smart contract attribute;
tokenizing the smart contract and the machine learning model to generate a model use token; and
appending the hash to the model use token, wherein the token is the model use token.
3 . The computer-implemented method of claim 2 , wherein the verification request comprises a notification indicating the first smart contract attribute is satisfied, and wherein determining the user hash matches the hash occurs automatically in response to the notification.
4 . The computer-implemented method of claim 1 , wherein generating the hash based on the machine learning model comprises:
accessing a model weight associated with the machine learning model; and
hashing the model weight.
5 . The computer-implemented method of claim 4 , wherein hashing the model weight comprises applying the model weight as input to a hash function.
6 . The computer-implemented method of claim 5 , wherein the hash function is one of: MD5, SHA-1, or SHA-256.
7 . The computer-implemented method of claim 1 , wherein generating the hash based on the machine learning model comprises applying source code of the machine learning model as input to a hash function.
8 . The computer-implemented method of claim 7 further comprising:
transmitting a source code request to a machine learning model storage location; and
receiving the source code of the machine learning model from the machine learning model storage location.
9 . The computer-implemented method of claim 1 further comprising:
generating a public listing for the machine learning model; and
transmitting user interface information comprising the public listing for display in a graphical user interface, and
wherein the verification request is received via the graphical user interface.
10 . The computer-implemented method of claim 1 , wherein generating the token comprises:
accessing a model weight associated with the machine learning model; and
tokenizing the model weight.
11 . The computer-implemented method of claim 1 , wherein generating the token comprises:
accessing source code of the machine learning model; and
tokenizing at least a portion of the source code.
12 . The computer-implemented method of claim 1 , wherein the token is generated according to a tokenization standard, and wherein the tokenization standard is selected based in part on the blockchain storage system.
13 . The computer-implemented method of claim 12 , wherein the blockchain storage system is Ethereum, and wherein the tokenization standard is one of: ERC-721, or ERC-1155.
14 . The computer-implemented method of claim 1 , wherein the blockchain storage system is selected from among a plurality of blockchain storage systems based on the request to generate the model token.
15 . The computer-implemented method of claim 1 , wherein the hash is generated using a hashing function, and wherein the hashing function is selected based in part on the blockchain storage system.
16 . The computer-implemented method of claim 1 further comprising:
transmitting user interface information to a user computing system to cause the user computing system to display a graphical user interface, wherein the request is received via the graphical user interface.
17 . The computer-implemented method of claim 1 further comprising:
receiving an access request for the machine learning model; and
in response to receiving the access request, transmitting the hash to a user computing system.
18 . The computer-implemented method of claim 17 further comprising:
receiving a second user request for the machine learning model comprising a second user hash;
determining the second user hash does not match the hash; and
based on determining the user hash does not match the hash, transmitting an alert to the user computing system indicating the integrity of the machine learning model is not verified.
19 . The computer-implemented method of claim 1 further comprising:
receiving an update notice associated with the machine learning model indicating a change to at least one weight value associated with the machine learning model;
generating an updated hash for the machine learning model based in part on the changed at least one weight value;
generating an updated token comprising the updated hash; and
storing the updated token on the blockchain storage system.
20 . The computer-implemented method of claim 19 further comprising transmitting an update notice to a user of the machine learning model indicating the change to the at least one weight value.