Lab Name

Flavum

Short Description

Flavum is an experimental chain-agnostic machine learning L2 rollup project for creating a platform that can integrate machine learning with different blockchain solutions. It offers the possibility of bringing trustless machine learning (ML) into different decentralized ecosystems. ML algorithms are executed in a rollup system, providing a transparent and traceable way of ML inference or even training. Trustless execution, inference, or training are guaranteed by zero-knowledge proofs.

Scope of Lab

The lab implements a prototype project for a chain-agnostic machine learning L2 rollup system. Instead of classical transactional execution, the rollup executes inference of different machine learning (ML) models on different data points. Inference on data points is stored in the rollup system as a trace in a transparent way. Correctness of execution is guaranteed by zero-knowledge proofs and Merkle trees that are written into possible different distributed ledger systems. Data points and ML algorithms can be presented in different ways, like owned or common objects providing the way of modelling them as special NFT objects. Execution traces among these objects are stored on the rollup. Blockchain integration is built up in a modularized way, providing the possibility to integrate with different distributed ledger technologies.

Blockchain integration provides use cases for:

  • Transferring tokens or information from a blockchain platform to a data point of a rollup.
  • Transferring rollup data points back to a blockchain platform token or other information.
  • Deploying machine learning (ML) models to the rollup system. The deployment is integrated with the given blockchain system by zero-knowledge proofs.
  • Executing inference of an ML on different data points is guaranteed by zero-knowledge proofs and Merkle trees for state transition. Roots of the Merkle trees are written into the ledger with the help of different smart contracts.

Additional features might include:

  • Storage of the execution trace can be configured in different ways, like storing it in special “cheap” fields of a ledger technology or storing it in a distributed storage system.
  • Integrating machine learning training as well into the rollup system.

Initial Committers

  • Daniel Szego https://github.com/Daniel-Szego

Sponsor

Pre-existing repository

none

This site uses Just the Docs, a documentation theme for Jekyll.