Half Jury: Blockchain Audited Boltzmann Reputation Protocol for Computational Verification
Additional Funding Sources
This project was supported by NSF CISE REU, Award No. 2051127.
Presentation Date
7-2022
Abstract
Distributed networks of independently operating computers often require protocols to prove procedural integrity, data provenance, and accuracy. A potential solution to this problem is a distributed peer-to-peer network of nodes who rely on each other for computational validation; however, a robust system for measuring reputation is necessary for a dependable system. Each pre-existing implementation has its own features it targets for optimized performance, but the main requirements and challenges remain universal. Primarily, a hard measure of “trust” must be produced by the system as an operational metric for quantifying reliability of an operator or element in that system. With these considerations in mind, this paper proposes a three-part protocol implemented over a decentralized network. First is a novel system that dynamically scores nodes on historical performance using geometrically expanding historical intervals, each assigned an Inverted Harmonic Weight (IHW) used to calculate reputation. Second is the Slow Boltzmann Estimator (SBE), which takes the reputations of a stochastically determined quorum and produces a log-normal likelihood of good-faith behavior by the participants, invariant of their performance in the current quorum. Applied iteratively over an entire network, this system is able to perform the desired functions of removing consistently under-performing nodes, dynamically adjusting parameterization based on changing network conditions and node behavior, and optimize the global average reputation of the entire network.
Half Jury: Blockchain Audited Boltzmann Reputation Protocol for Computational Verification
Distributed networks of independently operating computers often require protocols to prove procedural integrity, data provenance, and accuracy. A potential solution to this problem is a distributed peer-to-peer network of nodes who rely on each other for computational validation; however, a robust system for measuring reputation is necessary for a dependable system. Each pre-existing implementation has its own features it targets for optimized performance, but the main requirements and challenges remain universal. Primarily, a hard measure of “trust” must be produced by the system as an operational metric for quantifying reliability of an operator or element in that system. With these considerations in mind, this paper proposes a three-part protocol implemented over a decentralized network. First is a novel system that dynamically scores nodes on historical performance using geometrically expanding historical intervals, each assigned an Inverted Harmonic Weight (IHW) used to calculate reputation. Second is the Slow Boltzmann Estimator (SBE), which takes the reputations of a stochastically determined quorum and produces a log-normal likelihood of good-faith behavior by the participants, invariant of their performance in the current quorum. Applied iteratively over an entire network, this system is able to perform the desired functions of removing consistently under-performing nodes, dynamically adjusting parameterization based on changing network conditions and node behavior, and optimize the global average reputation of the entire network.