Publish a model on Github

If your dapp needs a particular ML model to be run to identify face from a video, count the number of reps of pushups from a video etc. Create a trained model and upload to a public repository.

Servers will run these models

Any server can join a network of ML nodes.

When a ML task request comes in - e.g. Does this video have a real person in it? - the node will download the model from github, and execute the model on the input provided

POST /ml-verify

DATA
{
  "model": "https://github.com/my-repo/my-model/branch/main/model.tar",
  "input": "ipfs://12..34",
  "payment_sig" : "123...456"
}

Commit reveal

A commit reveal game is played.

ML model commits to an output - e.g. “Yes this video has a person”.

Once most of the nodes have committed, a reveal happens.

Majority wins.

All the nodes in the majority will get a share of fees.

Slash?

Maybe there’s a partial slashing for the nodes that was wrong - basis a stake to participate in the network.