Decentralised Federated Machine Learning on Polkadot

A protocol for training AI models across distributed nodes without centralising data. Governed by collective consensus.

Centralised AI Limits Access and Control

Today’s dominant AI models (OpenAI, Meta, Microsoft, etc.) require massive centralised compute and data. This limits access, raises safety and control concerns, and concentrates power in the hands of a few.

DecentralML aims to democratise AI through decentralised federated machine learning with transparent on-chain governance.

A Polkadot Protocol for DFML

DecentralML is a Polkadot protocol for decentralised federated machine learning (DFML) with collective safety consensus. Model weights, licensing, and jurisdiction rules are controlled transparently on-chain.

Privacy-preserving

Node or on-device training — data never leaves the source.

BYO Token Staking

Flexible economy using Substrate pallets (Balances, Grandpa, Ink!).

Collective Consensus

Model weights, licensing, and safety controls on-chain.

Built on Substrate

DecentralML leverages Substrate’s modular pallet architecture for seamless integration with the Polkadot ecosystem.

Substrate / Polkadot / Kusama Rust pallet-decentralml Python client TensorFlow Federated Docker IPFS (Infura)

What You Can Do

Decentralised Federated ML

Train across nodes without centralising data.

Collective Governance

On-chain controls for model weights, licensing, safety, jurisdiction.

Node or On-Device Training

Privacy by design — data stays local.

Collaborative AI Training

Share training between organisations and nodes.

Polkadot-JS Apps

Connect via ws://localhost:9944.

Where It Fits

Privacy-preserving ML for healthcare, finance, or regulated sectors. Collaborative model training across organisations. Transparent, on-chain governance of AI models. Reducing centralisation and improving power and compute efficiency.

Run the Node

Clone the repo and follow the GitHub README for setup.

cargo build -p pallet-decentralml
./target/release/node-decentralml --dev

Use the Python client and Docker instructions in the repo. Connect via Polkadot-JS Apps.

Apache 2.0

DecentralML is fully open source. Contributions, forks, and ecosystem integration are encouraged.

View on GitHub