GBC.AI

Dec 2, 2021

4 min read

Guardians of the Blockchain (GBC.AI) wins at the Blockchain and Internet of Things Conference 2021 (BIOTC) with "Machine Learning View on Blockchain Parameter Adjustment."

ROBERT VASILIEV, GBC.AI Chief AI Officer, Vice-President of Artificial Intelligence Laboratory Association

In July 2021, the GBC.AI framework development team participated in the BIOTC-2021 conference http://www.biotc.net with an applied report "Machine Learning View on Blockchain Parameter Adjustment", which won in the published section.

Please find the essence of the report and the issues raised in it, for more information, don't hesitate to contact us.

What is the main problem that we know how to complete?

To get consensus, a fundamental problem in distributed ledger computing is achieving an agreement among many parties for a single data value in the presence of faulty processes. The consensus mechanism is an underlying part of blockchain design and commits new blocks and changes protocol itself. In addition to classic correctness requirements, blockchains need specific ones: high performance regarding transactions per second, fast transaction confirmation, and so on.

Blockchains control requirements with parameters, but how does the network meet qualitative and optimize quantitative requirements? Typically we have the main blockchain network (mainnet) without access to try different parameters to test network parameter optimizations. In our paper, we have provided a machine learning view on blockchain parameter adjustments. We list blockchain parameters for the Solana blockchain and apply feature importance to select the most significant parameters for optimizations.

Short description

In addition to classic correctness requirements, blockchains may need high performance regarding transactions per second, fast transaction confirmation, low block production time, and others. The requirements compete, for example, transaction confirmation time and performance. So the blockchain has a Pareto front of optimal operation regimes and needs to pick the proper regime through a trade-off between the current system needs.

Consensus mechanisms control the requirements with parameters. But how to meet qualitative requirements and optimize quantitative properties? We have the mainnet blockchain for a given system without access to change parameters for research purposes. The source code of the blockchain is publicly available. So we can launch a test system in our testnet environment and vary parameters as we wish. We observe parameters for both main and test network systems in time together with the resulting operation quantities. The mainnet provides historical data, while the testnet is an interactive black box.

The blockchain system runs in a distributed network. Both network and protocol parameters define the blockchain operation regime. Some of the network parameters are observable, for example, the current pool of unconfirmed transactions; some of the network parameters are not observable (unobservable, latent), for instance, blockchain network graph, connection latency and bandwidth. Unobservable parameters play an essential role in the performance; this can be how average propagation delay affects transaction latency, and it is helpful to estimate them with a certain accuracy. Latent parameters make it impossible to emulate within the mainnet but only simulate with the testnet. So we have a multi-objective optimization problem with multi-fidelity sources: high accuracy dataset and low accuracy interactive black box.

At the BIOTC21 conference, GBC.AI presented a machine learning view on a blockchain parameters adjustment problem. To the best of our knowledge, such a view is not discussed in the scientific literature yet. We also perform feature importance analysis for Solana (as an experiment) blockchain by SHAP algorithm to showcase how to apply machine learning to the blockchain parameters adjustment subtasks.

The balance of our paper follows this track:

Section 2 provides a few insights on blockchain parameters and their relation to the operating regime.

Section 3 lists machine learning problems for blockchain parameter adjustment.

Section 4 considers the Solana blockchain as a model example for an adjustment.

Section 5 sums up its parameters in SHAP algorithm, which provides feature importance analysis for Solana's parameters.

Section 6 presents our conclusions.

A complete version of the article is available for download and review at ACM digital library Machine Learning View on Blockchain Parameter Adjustment | 2021 3rd Blockchain and Internet of Things Conference.

The article is available free of charge for accessing the site from Internet access points inside institutes that have purchased subscriptions to relevant scientific publications. At the same time, the magazine itself is available for purchase at any time.

This article is the first article in a series of publications that the GBC.AI team is preparing in an ongoing study on applying AI technologies in blockchain technologies. Follow us for more information and updates.

Contact:

William De'Ath

Chief Communications Officer

will@gbc.ai

https://gbc.ai

GBC.AI incorporates artificial intelligence (AI) into all blockchains in a transformative way. We make blockchains smart.

Love podcasts or audiobooks? Learn on the go with our new app.