Core Concepts
In decentralized systems, the adversary benefits when the legitimate decoder recovers the data with high estimation error. This incentive motivates the adversary to act rationally, trying to maximize its gains. The authors propose a game-theoretic formulation, called the game of coding, that captures this unique dynamic where each of the adversary and the data collector have a utility function to optimize.
Abstract
The content discusses a new game-theoretic framework for coding, called the "game of coding", which is motivated by emerging applications in decentralized systems such as decentralized machine learning (DeML) and oracles.
The key insights are:
- In decentralized systems, the adversary benefits when the data is recoverable by the legitimate decoder, but with high estimation error. This incentivizes the adversary to act rationally to maximize its gains.
- The game of coding framework captures this dynamic, where the data collector (DC) and the adversary each have a utility function to optimize.
- The utility functions reflect the fact that both the DC and the adversary are interested in increasing the chance of data being recoverable, but the DC wants to estimate the input with lower error while the adversary wants the opposite.
- As a first step, the authors characterize the equilibrium of the game for the repetition code with a repetition factor of 2, for a wide class of utility functions with minimal assumptions.
- The game-theoretic view opens new avenues of cooperation between the adversary and the DC that enables the system to remain live even if the conventional trust assumptions (e.g., honest majority) are not satisfied.
- The authors determine the optimum strategy of the DC in accepting/rejecting the inputs, and the optimum noise distribution for the adversary that achieves the equilibrium.
Stats
The content does not contain any key metrics or important figures to support the author's key logics.
Quotes
The content does not contain any striking quotes supporting the author's key logics.