Explaining XAIN and Proof of Kernel Work (PoKW)
The core of XAIN to achieve its mission is the eXpandable AI Network, as a Blockchain systems that can be applied as an open or permissioned system with a focus on small IoT devices, such as ECUs in cars. Hereby, it leverages a new consensus mechanism, which has been developed together with researchers of the University of Oxford and Imperial College London and which enables two factors: First it leads to a pure democratization of the distributed network through randomization and the prevention of more influence through more computing power or more stakes in the system. Further, the consensus algorithm achieves full scalability and a reduction of the used computing power in the network and in the devices, which allows small devices like sensors, smartphones or ECUs in cars to participate in the consensus, rather than trusting large server pharms to do this work.
The eXpandable AI Network itself leverages the robust, proven Ethereum Blockchain — the industry-leading decentralized, global infrastructure platform for Smart Contracts — but deviates from it in some crucial ways. One significant difference from the public Ethereum Blockchain is that we have replaced the mechanism for electing a leader to decide on the contents of the next block in the Blockchain (the Block Producer) with a bespoke consensus mechanism.
The consensus mechanism that the eXpandable AI Network uses ( Proof of Kernel Work (PoKW)) relies on Ethereum’s Proof of Work, but restricts its use in determining a Block Producer first to a known, set of Ethereum addresses and secondly to a sub-set of these, that changes every block, and the members of which are selected by means of a secret lottery. The validity of any node’s claim to be part of the committee elected by means of the secret lottery can be verified any node on the system.
PoKW allows us to configure both the size of the super-set of Ethereum addresses that could possibly act as Block Producers (the whitelist) and the target size of the committee that are eligible to ‘mine’ a block through PoW for the given block. The whitelist of Ethereum addresses is maintained in a custom Smart Contract that is initialized in the eXpandable AI Network’s Genesis block. By using a Smart Contract to maintain the whitelist we achieve the following things:
· Every node in the network can easily verify which addresses are in the whitelist.
· Addresses can be added to or removed from the whitelist by means of Ethereum transactions — and we have complete control over which addresses can modify the whitelist and at when.
The mechanism for determining the committee of nodes to run the PoW algorithm (the secret lottery) is achieved by means of a “verifiable random function” which can not be manipulated by any of the nodes in the network. We initialize the random function such that is both verifiable and not possible to manipulate by embedding the initial seed in the Genesis block of the eXpandable AI Network.
The secret lottery’s random function is probabilistic; based on the target committee size maintained in a Smart Contract (we made this decision for similar reasons to those outlined above for the whitelist). The secret lottery is run privately on every node that is in the whitelist. All of the necessary components of the random function for the current block are taken from the previous block or Smart Contract — including the random seed and target committee size — meaning that
· Every node has all of the necessary components to execute the random function.
· Every node can verify the legitimacy of another node’s claim to be in the committee.
· The inputs to the random function cannot be modified or manipulated.
· The outcome of the random function is unique to each address.
· An attacker can’t determine ahead of time which nodes will be in the committee unless they have compromised their private keys.
We have also moved some of the existing parameters of Ethereum that we wish to be able to programmatically modify into a Smart Contract — for similar reasons of transparency to the network and control.
Furthermore, a second committee is elected to distribute the entire anomaly detection of the network. The reason for that is to avoid a single-point-of-failure, especially as the intelligence behind the control of machine learning could lead to a network deconstruction. As such, the decision of the anomaly detection is pursued in a consensus of PoKW. The anomaly detection itself decides about the participation in the whitelist as well as the participation in the entire network for reasons of stability.
The following graphics demonstrate the pursued processes in a UML diagram. Hereby, the first diagram described PoKW, while the second diagram is focused particularly on the machine learning (ML) processes for the anomaly detection in the network:
PoKW Consensus Protocol:
Machine Learning Consensus Mechanism:
Don’t forget to check out our last blog article in which we answered why we saw the necessity to build our eXpandable AI Network in first place and introduced our technical vision.
For more information on our business, feel free to visit our webseite: www.xain.io
Besides our technical blog section, we also have a rather business-oriented section, where we introduce various use-cases of blockchain and AI technology in the industry.