Zero-Knowledge Machine Learning (ZKML) is concept that marries the realms of privacy-focused cryptographic techniques and machine learning. Aleo, a zero-knowledge blockchain, contributes to the ZKML space by providing a platform to develop private applications. With its unique programming model, developers can write privacy-preserving, verifiable computations with ease, which are then compiled into zero-knowledge proofs. In the context of ZKML, Aleo can enable the training and execution of machine learning models in a zero-knowledge environment, ensuring that no information about the data, apart from the output, is revealed. This is a big step towards privacy-preserving AI/ML solutions.
STAKEME took part in the zkml-initiative-1 hackathon and the work was done in two sections: zkML Algorithms and Python Developer Tooling
In the zkML Algorithms section: The Linear Regression ML algorithm was developed using the Leo language. More detailed description is in Google Colaboratory (you can use online) or you can download .ipynb (Jupyter Notebook)
Google Colaboratory: https://colab.research.google.com/drive/1c3_Pu7yELRymKxCocp4PKbJ3uGVsXTq7?usp=sharing
In the Python Developer Tooling section: the following modules have been developed
Running Leo from Python
Usage
from interp_leo.leo_uitls import convert_from_leo_type
from interp_leo.leo_program import LeoProgram
leo_program = LeoProgram(path=os.getcwd() + '/path_to_aleo_application')
leo_program.func(...)leo_program will automatically parse all functions and you will be able to access these functions directly in python. When passing arguments, they will be automatically converted to Leo type
Convert types Python/Leo
convert_from_leo_type('1u128') #128
convert_to_leo_type(10, 'u128') #10i128Note: In case of exceeding the range of available numbers of a certain type, an error will be issued