Velas Technologies: AIDPOS

The Velas team conducted fundamental research to understand how AI could be implemented within the blockchain architecture to optimize the performance of the entire network. Based on this research, we are designing and developing two innovative ecosystem components — AIDPoS consensus algorithm and Distributed Learning.

AIDPOS consensus algorithm

The main principle behind AIDPOS is to use AI to adapt the blockchain to current circumstances within the network, preserving the level of the performance and resilience primarily in the optimal ranges. This is achieved by embedding trained models within every full node, which is producing the optimal values of the key parameters for the blockchain, based on the data collected from the last epoch.

One of the most promising areas of integrating AI and blockchain technologies together like this is the development and training of a ‘recommender algorithm’ based on machine learning technologies, which would provide a dynamic change in the parameters of the network and the consensus smart-contracts from epoch to epoch. Such an algorithm should ensure that the Velas blockchain network stays secure, resilient, and productive for all of its participants. Velas proposes using the global network state data and local nodes state data of the previous epoch for predictions. The recommender algorithm will act as the objective “guardian” of the network.

There are two main concepts here that are important for proper network functioning — performance and resilience. Performance can be measured by two straightforward metrics — throughput, which is measured by transactions per second, and transaction confirmation time. Resilience is the ability of the blockchain to withstand any types of attacks and properly function during these events, should they ever happen. Two main characteristics that ensure Velas maintains resilience are security (the amount of resources that an attacker needs to spend to break the blockchain) and decentralization, which can be described as the absence of a single point of failure.

Distributed Learning

We propose utilizing the computing resources of Velas network participants for distributed computing and for the building of infrastructure for external developers within the fields of machine learning and deep learning. This advanced solution will allow network users to monetize their computing power, gaining a reputation for service provided and rewarded with Velas tokens. Additionally, this will enable external entities to use and utilize to the best of its ability the distributed power of the Velas community to execute personal tasks.

The best practices to work with neural network models and the most popular support frameworks are Tensorflow and PyTorch. It is assumed that a library will be implemented that resembles something similar to that of Horovod library, but at the same time supporting the gRPC protocol and fully decentralized computing. In general, a solution for distributed computing should be flexible, adaptable, and based on the popular, universally recognized open-source framework. Thus, users will not have to “learn” a new tool, and the bugs and problems of the framework will be eliminated by the community as quickly as possible. Also, there will be no barriers to entry for the writing offully custom training scripts.

The user will be able to independently choose a paradigm that meets his needs for the task of teaching his own model.

One of the use cases of developed distributed learning technology is to ensure decentralization when training a recommendation model of optimal Velas network parameters on new data (provided that the model training fits into the distributed learning paradigm and requires large computational resources). Unlike centralized training on a single server, with this approach, training will be carried out on the computing power of Velas network nodes, thereby making training honest and open. Also, any network member will be able to test the trained model to make sure the learning outcomes are correct.

Main Result

The main result is the first version of Velas blockchain with a recommendation model based on ML/DL technologies. During the last 6 months we have been finishing this scope of the work:

  • Formalization of optimized function, blockchain characteristics and parameters.
  • Development and launch simulation infrastructure.
  • Development of scenarios for simulating a dataset for model training.
  • Features formalization, Node’s logs data mining as like data for training ML/DL algorithms.
  • Development and training of v0.1 recommender model for Velas Blockchain.

Main deliverables:

  • Accumulating historical dataset
  • Developed data collection script;
  • First version of Velas Blockchain simulation model;
  • Data collected from simulation processes;
  • Proof of concept AIDPOS. Trained model for optimization objective functions.
The Velas AI module should have the possibility to adjust blockchain configurations to the appropriate data from historical datasets and from data from N-1 epoch.

Summary

After conducting this fundamental research, we are attempting to create a unique blockchain platform empowered by AI. This will lead us to the creation of a complex adaptive ecosystem that, as it’s core philosophy, embraces sustainability, high performance, and economic equability for its participants. To achieve these synergy effects we will introduce, at first, AIDPOS consensus and Distributed Machine Learning System. However, AI implementation within the Velas Ecosystem is not limited to these features. In the near future, we plan to develop valuable products which will also be enhanced with advanced AI technologies.


Share

By About

Leave a Reply

Your email address will not be published. Required fields are marked *