I am building a VRF Validator Selection Simulation in Rust

Problem

The current Decentralized Machine Learning relies on static leader election or fastest node win strategy, which poses significant threats, allowing nodes with better hardware to monopolize gradient validation.

Solution

My work proposes replacing predictable leaders with a Verifiable Random Function (VRF) combined with a dynamic trust scoring system. This ensures that validator selection remains cryptographically unpredictable to attackers while favoring honest nodes.

Proof

This project is a Rust simulation of the same concept providing experimental proof that trust score along with VRF selection reduces the probability of selecting a malicious node in the validator committee from 30% random baseline to approximately 16.5%.

Terminology

• VRF - Verifiable Random Function
• Validator - A node that validates gradients submitted by other nodes and then helps compute the global aggregate.
• Simulation - Representation of something that could exist in real life.
• Node - One system in a distributed environment that has identity, keypair, and flag.
• VRF Output - A pseudorandom value computed per node, per round, using that node's secret key and the round number.

NPM Package

This engine is compiled to WebAssembly (WASM). You can find the public release here: https://www.npmjs.com/package/vrf-bft-simulator

Run locally

1. Fork the repo
2. Clone the repo
3. Open in your IDE
4. Run cargo build && cargo run (or wasm-pack build for WebAssembly)

Made by Ayush Saksena