Transaction Lifecycle: proposers & voters

FLock System Design

A transaction lifecycle in blockchain is the stages of sending via blockchain, be it a message, cryptocurrency, or another type of digital asset.

1. User stakes tokens to participate

2. Trainers submits the training result

3. Validation: checked that it’s legitimate and follows the rules

4. Confirmation

5. Repeat steps 2 - 4 until training fully completed

6. Log the finalised data on the chain.

In the blockchain, the computer nodes are divided into two roles: proposers and voters.

Proposers

‘Proposers’ are blockchain nodes that teach the network new things.

Their selection is based on a special system called proof-of-stake (PoS). They train ML models using their local private datasets and subsequently report updates to the blockchain.

Proposers’ responsibilities:

• Utilising local data insights to refine the global model's accuracy.

• Implementing gradient-based algorithms or other suitable techniques for weight optimisation.

• Collaborating in a decentralised manner through secure multi-party computation (SMPC).

• Contributing to continuous model improvement and facilitating real-time updates to achieve localised accuracy.

Voters

Voters are a select group of blockchain nodes that form an auditing committee for quality control. During the training phase, they are tasked with scrutinising suggested changes to the network’s knowledge base to ensure that they are accurate and fair.

Voters form a consensus committee within the FLock ecosystem, selected based on specific criteria, such as reputation or staking.

During the training process, they are entrusted with:

• Reviewing and validating aggregated model weights.

• Employing cryptographic verification techniques to ensure the integrity of weight updates.

• Participating in a decentralised consensus protocol to maintain learning process fairness.

• Identifying and reporting inconsistencies or fraudulent activities.

• Upholding transparency and integrity.

FLock's system of proposers and voters brings a new level of decentralisation, security, and efficiency to machine learning.

By combining blockchain's immutable and transparent nature with state-of-the-art machine learning techniques, FLock is setting new standards for federated learning with blockchain.