Is it possible somehow to train a machine learnig model inside the Cartesi Machine? I believe, if models are trained outside Cartesi, its not possible to audit the results integrity, if it is biased or not. If everything is within Cartesi, I think this would be possible.
projects that I saw:
https://github.com/souzavinny/rollups-examples/tree/main/biometrics
https://medium.com/cartesi/ecosystem-update-mainstream-developers-on-the-blockchain-os-e7210b381ca4
CodePudding user response:
It's possible to have two approaches: just run the trained model inside the Cartesi Machine or train it inside. If your DApp doesn't benefit from proving the model itself, you can port just the model to the Cartesi Machine. If allowing others to replicate the training to obtain exactly the same model is important to your DApp, you can have the training set available along with the Cartesi Machine to train it so others can reproduce it.
Bare in mind that if you want to train the model inside the Cartesi Machine, you'll have to port all the dependencies needed to train it while if you just want to run the model you can take a similar approach to the one on the biometric example you provided, needing the dependencies only on your native machine to generate the model and not having to worry about porting them to the Cartesi Machine RISC-V based ISA.