Adaptive and Robust Cost-Aware Proof of Quality for Decentralized LLM Inference Networks

TL;DR

Decentralized LLM inference requires scalable quality verification, but evaluator heterogeneity and adversarial manipulation can erode incentive alignment. The authors propose an adversary-resilient cost-aware Proof of Quality (PoQ) by integrating robust consensus rules (mean, median, trimmed mean, adaptive weighted mean) with an adaptive, deviation-driven trust-weighting mechanism, while preserving cost-based rewards. Through Monte Carlo simulations on QA and summarization tasks with varying malicious ratios $ ho$ and evaluator set sizes $K$, they show that robust aggregation substantially improves alignment with ground truth and reduces sensitivity to noise and strategic attacks, though payout inflation can occur under boosting attacks. The work provides practical deployment guidance, including when to rely on robust aggregation, how to tune $K$, and how to balance robustness with reward variance in open, resource-constrained networks.

Abstract

Decentralized large language model inference networks require lightweight mechanisms to reward high quality outputs under heterogeneous latency and cost. Proof of Quality provides scalable verification by sampling evaluator nodes that score candidate outputs, then aggregating their scores into a consensus signal that determines rewards. However, evaluator heterogeneity and malicious score manipulation can distort consensus and inflate payouts, which weakens incentive alignment in open participation settings. This paper extends a cost-aware Proof of Quality mechanism by adding adversary-resilient consensus formation. We study robust aggregation rules, including median and trimmed mean, and an adaptive trust-weighted consensus that updates evaluator weights from deviation signals. Using question answering and summarization workloads with a ground truth proxy for offline analysis, we quantify evaluator reliability and show strong variance across evaluators, including task-dependent misalignment that can invert correlations. We then evaluate robustness under four adversarial strategies, including noise injection, boosting, sabotage, and intermittent manipulation, across a sweep of malicious ratios and evaluator sample sizes. Our results show that robust aggregation improves consensus alignment with the ground truth proxy and reduces sensitivity to noisy and strategic attacks compared with simple averaging. We further characterize the operational trade-off introduced by evaluator sampling, where larger evaluator sets reduce evaluator rewards and increase payoff variance while inference rewards remain relatively stable in our configuration. These findings motivate robust consensus as a default component for cost-aware Proof of Quality and provide practical guidance for selecting evaluator sampling parameters under adversarial risk and resource constraints.

Figures (14)

• Figure 1: System overview of adversary-resilient cost-aware PoQ. The figure highlights the main workflow from inference to evaluation, together with cost signals, reward assignment, robust aggregation, deviation signals, and trust weight updates.
• Figure 2: Threat model for malicious evaluators. Malicious evaluators manipulate submitted scores via noise injection, boosting, sabotage, and intermittent strategies. The consensus module uses robust aggregation and deviation signals to support adaptive trust weights.
• Figure 3: Baseline rewards for inference nodes and evaluator nodes under cost-aware PoQ.
• Figure 4: Evaluator reliability analysis, including task level trends and deviation statistics.
• Figure 5: Correlation heatmap between evaluator signals, consensus variants, and the ground truth proxy.