Game of Coding for Vector-Valued Computations

The game of coding is a new framework at the intersection of game theory and coding theory; designed to transcend the fundamental limitations of classical coding theory. While traditional coding theoretic schemes rely on a strict trust assumption, that honest nodes must outnumber adversarial ones to guarantee valid decoding, the game of coding leverages the economic rationality of actors to guarantee correctness and reliable decodability, even in the presence of an adversarial majority. This capability is paramount for emerging permissionless applications, particularly decentralized machine learning (DeML). However, prior investigations into the game of coding have been strictly confined to scalar computations, limiting their applicability to real world tasks where high dimensional data is the norm. In this paper, we bridge this gap by extending the framework to the general $N$-dimensional Euclidean space. We provide a rigorous problem formulation for vector valued computations and fully characterize the equilibrium strategies of the resulting high dimensional game. Our analysis demonstrates that the resilience properties established in the scalar setting are preserved in the vector regime, establishing a theoretical foundation for secure, large scale decentralized computing without honest majority assumptions.

💡 Research Summary

The paper extends the recently introduced “game of coding” framework—originally limited to scalar computations—to the general N‑dimensional Euclidean space, thereby addressing a critical gap for real‑world decentralized applications such as permissionless machine learning (DeML). In the model, a data collector (DC) outsources a computation to two external worker nodes: one honest node (H) and one adversarial node (A). The honest node reports a noisy version of the ground‑truth vector U∈ℝⁿ, where the noise N_h is uniformly distributed inside an N‑ball of radius Δ, i.e., N_h ∼ Unif(B_N(Δ)). The adversarial node knows the exact realization of U and may add any noise N_a drawn from a distribution g(·) of its own choosing; this distribution is private and unrestricted.

The DC first announces a scalar acceptance threshold η. It accepts the pair of reports (Y₁,Y₂) if the Euclidean distance between them satisfies ‖Y₁−Y₂‖₂ ≤ ηΔ. Upon acceptance, the DC outputs the average (Y₁+Y₂)/2 as its estimate 𝑈̂; otherwise it rejects the computation entirely. The probability of acceptance P_A(g,η) and the mean‑squared error MSE(g,η) are defined over the randomness of U, N_h, and N_a.

A game‑theoretic interaction is then formalized: the DC, acting as a leader, chooses η to maximize a utility that rewards high acceptance probability and low estimation error; the adversary, observing η, selects a noise distribution g to maximize a utility that balances the magnitude of injected error against the likelihood of passing the acceptance test (and thus receiving a reward). The authors assume only minimal, natural properties for the utility functions (monotonicity, continuity), making the analysis broadly applicable.

The central technical contribution is the reduction of an infinite‑dimensional optimization (over all possible g and η) to a tractable two‑dimensional problem. By introducing an intermediate optimization problem (equation 19), the authors show that the equilibrium strategies can be expressed solely in terms of η and a Lagrange multiplier λ that encapsulates the adversary’s optimal distribution. Theorem 1 proves that solving this 2‑D problem yields a Nash equilibrium for the original game, and Algorithm 1 provides a concrete grid‑search procedure to compute the equilibrium efficiently.

Theorem 2 supplies a closed‑form solution to the intermediate problem by exploiting the uniform‑ball noise model. The distance ‖Y₁−Y₂‖₂ follows a distribution that can be expressed using Beta and Gamma functions, allowing the authors to write P_A and MSE analytically as functions of η and Δ. Consequently, the DC can evaluate its expected utility for any η and select the optimal η* without enumerating all possible adversarial strategies.

Numerical experiments are conducted for dimensions N = 2, 5, 10. The results illustrate how the acceptance probability declines with increasing dimension for a fixed η, yet an appropriately tuned η* can keep the MSE within acceptable bounds. The simulations also compare two adversarial behaviors: “strategic denial” (choosing a distribution that just exceeds the threshold, causing rejection) versus “strategic corruption” (injecting large noise while still passing the test). The trade‑off between system liveliness and robustness to DoS attacks is quantified, showing that the reward design critically influences which adversarial mode dominates.

In conclusion, the paper demonstrates that the game‑of‑coding paradigm can guarantee correct and reliable decoding even when the majority of participants are adversarial, without relying on any honest‑majority assumption. By extending the framework to vector‑valued computations and providing a 2‑D equilibrium‑finding method, the authors lay a solid theoretical foundation for secure, large‑scale decentralized computing in permissionless settings. Potential future directions include extending the analysis to more than two workers, handling asynchronous communication, incorporating multi‑round interactions, and designing dynamic reward policies that adapt to observed adversarial behavior. These extensions would further bridge the gap between theory and practical deployment in blockchain‑based oracle networks, federated learning, and privacy‑preserving data markets.