Secure machine learning on Internet Computer
The goal is to create a prototype of a secure and decentralized machine learning application (train and predict) using Internet Computer blockchain and compare similar solution locally and with traditional centralized approach.
cycles-cost-tested/ic-mnist folder contains trials including tests with canbench on the cycles costs of loading MNIST training dataset (50k records for ~200B cycles).
cycles-cost-tested/ic_djstorch folder contains test with instruction overflow (more than 40B instructions) with only 10 MNIST records using Azle.
ic_linfa is a first working machine learning (train/test) prototype using Linfa ML toolkit on the Internet Computer and is secured with encrypted message to and from the canister.
- You can try the prototype here https://zvrui-xqaaa-aaaao-a3pbq-cai.icp0.io/
ic_split-learning a prototype of neural net using split learning between client wasm and canister.
I'm new to Rust language, so this is also a learning project for me.