A simple example demonstrating how to run an AI-based prover on W3bstream.

The AI model is designed to predict whether a user is walking or running. It operates under the assumption that the user wears 'smart shoes' equipped with two smart sensors. These sensors are connected via Bluetooth to each other and to a mobile app. As the user walks, the main sensor measures the vertical force exerted by the feet and records the interval in milliseconds between two consecutive steps. It then calculates average values across several steps and sends this information to the mobile app, which in turn sends it to the W3bstream project. The W3bstream network 'proves' whether the user is currently walking or running through AI prediction and sends the proof to the project's contract for tokenization.

This demo is based on the Smartcore AI library for Rust, specifically version 0.3.2.

It utilizes the KNN Classifier for simplicity, and no special optimization is included in this example. The dataset, training process, classifier, and optimizations can be easily adapted to create a model for other use cases, but such customizations are left as an exercise for the reader.

The AI model file generated by this tool must then be imported into the W3bstream provere. Check this repo for how to do it (WIP)

cd gen_model
cargo run

You should see an output similar to:

Finished dev [unoptimized + debuginfo] target(s) in 0.15s
     Running `target/debug/runner_ai`
********* Training  Model **********
100 samples. Accuracy: 1
********** Model trained **********
Prediction: [92.0, 76.0, 400.0] -> Running

The model file is located in the modeldirectory. Feel free to customize the training data in the data/ directory and modify src/main.rs as needed.

By now see https://github.com/simonerom/depin-zk-test/tree/ai-test