C

Cerebrum

An Open-Source Distributed AI Infrastructure built using the Substrate Blockchain.

The problem Cerebrum solves

Cerebrum is an Open-Source Distributed AI Infrastructure on Blockchain that brings machine learning models to organic data sources, allowing anyone to train and rewards nodes (Neurons) participating in the network by training models on their local data.

Cerebrum helps you simplify the process of training a Machine Learning model by taking a four-pronged approach.

  • Good organic data is at the crux of improving the results of a Model. Cerebrum brings the ML model to the neurons and trains on their local data without uploading any personal data to the cloud.
  • Users own their data and get rewarded for participating proportionally to their contribution to the network powered by the blockchain allowing participants to monetize their data and establish trust by seeing who uses their data on chain.
  • Accessible Model States make it easy to build applications on top of Cerebrum and Substrate as the ML models are available directly via a static link can be integrated into any application.
  • An Intuitive Interface phases out the complexities with a simple drag and drop interface to get started in a few clicks.

Additionally with these features, Cerebrum offers,

⚡ Lightning fast transactions powered by a custom blockchain built using Substrate and Frontier.

⛽ Gasless transactions powered by Meta-Transactions.

📌 Distributed Storage powered by IPFS, Textile and Fleek.

📁 Model States backed up via NFT.storage to Filecoin.

🌐 Live on The Internet Computer at https://mzyxx-jqaaa-aaaad-qaama-cai.raw.ic0.app

The Substrate node is accessible at

https://anudit-cerebrum-6hp8-9933.githubpreview.dev

but sleeps after a certain duration so let me know if it isn't accessible.

Challenges we ran into

Tons, this was our first foray into the Substrate and Polkadot ecosytem. Building and running a full node that is EVM compatible took some time but the frontier architecture made it simpler.
Additionally, the substrate evm-pallet has lots of gotchas with address conversions and integrations, but the documentation helped

Discussion