Description

Data Privacy is a big concern. People want good recommendation systems but don't want to give their data. Also Getting Data for making a medical Machine Learning model is a big deal as the patient's data needs to be kept private and very less medical data is available . DeAI solves this problem by giving an architecture where you provide data to train machine learning models but no body owns it nor it is exposed anywhere. The data you provide goes to our smart contract where the model is trained and then discarded. Only an irreversible hash of the data provided is stored in the the web3storage for discarding duplicating data. After the model has been trained the model is saved on the web3storage and can be used for getting predictions. For getting predictions the user has to buy a subscription whose money will be later provided to the code and data contributors.

DeAI showcase

How it's made

Training code of the Machine learning model is written inside our smart contract which interacts with our web app. The client form the web app can give training to the smart contract whose hash is matched from the file stored in the web3storage if the hash exists the data is discarded and the model is not trained. Else the the hash of the current data is appended in the hash file which is then uploaded back to web3storage. Then the model is retrained and the new model is then pushed to the web3storage. This model then can be used for getting predictions. For getting predictions the user has to buy membership (unlock protocol) for a minimal fee (this fee is then used to provide incentives to the contributors).

Technologies used

IPFS