Abstract
We present an architecture to implement a decentralised data market, whereby agents are incentivised to collaborate to crowd-source their data. The architecture is designed to reward data that furthers the market's collective goal, and distributes reward fairly to all those that contribute with their data. This is achieved leveraging the concept of Shapley's value from Game Theory. Furthermore, we introduce trust assumptions based on provable honesty, as opposed to wealth, or computational power, and we aim to reward agents that actively enable the functioning of the market. In order to evaluate the resilience of the architecture, we characterise its breakdown points for various adversarial threat models and we validate our analysis through extensive Monte Carlo simulations.
Introduction
Traditional data-driven business models are predominantly centralized, which often leads to significant concerns regarding data ownership, privacy, and the lack of fair compensation for data generators. While decentralized alternatives based on blockchain technology have emerged, they frequently replicate "rich-get-richer" dynamics through Proof-of-Stake or waste immense energy through traditional Proof-of-Work.
Crowd-sourced spatial data, such as air quality measurements from connected vehicles, presents unique challenges for decentralized trading. Such data is easily replicated, its value is highly context-dependent (time and location), and it is susceptible to manipulation by malicious actors providing fake data to earn rewards or skew market outcomes. This paper proposes a hybrid architecture designed to be both fair—rewarding data based on its actual utility—and robust against adversarial manipulation.
System Architecture
The proposed architecture utilizes a hybrid approach, combining trusted infrastructure for identity and initial verification with decentralized protocols for core market functions like consensus and reward distribution.
As shown in Figure 1, the system is divided into Centralised and Decentralised Infrastructure:
- Verification Layer: A centralized verifier issues validity tokens (NFTs) and uses Proof-of-Position to ensure agents are physically present where they claim to be, mitigating "Wormhole" attacks.
- Consensus and Access Control: Decentralized committees handle data veracity through voting and consensus.
- Data Marketplace and Ledger: Final transactions and ownership records are stored on a distributed ledger, with rewards distributed based on contribution.
Technical Innovations
Combination Maximum Entropy Voting (C-MEV)
To manage the computational complexity of voting in a decentralized environment, the authors introduce C-MEV. Standard Maximum Entropy Voting (MEV) suffers from $N!$ complexity, making it impractical for large networks. C-MEV reduces this complexity by using combinations rather than permutations and employing an iterative election process to select committees.
Useful Proof-of-Work via Shapley Values
A core innovation is the "useful" Proof-of-Work (PoW). Instead of solving arbitrary cryptographic puzzles, agents perform work that benefits the market: calculating the Shapley Value of incoming data. The Shapley value measures the marginal contribution of a data point to the collective goal (e.g., improving an air quality prediction model).
Crucially, the difficulty of this PoW is inversely proportional to the value of the data provided. High-quality, unique data requires less computational work to "sell," while low-utility or redundant data acts as a barrier, requiring more work. This aligns incentives and naturally filters out spam and low-quality adversarial inputs.
Robust Consensus
To aggregate measurements while resisting outliers and malicious actors, the system employs a decentralized mean-median algorithm. This provides a higher "breakdown point" (the threshold of malicious actors needed to subvert the system) compared to a simple mean calculation.
Evaluation and Results
The architecture was evaluated using real-world pollution data from the Central Pollution Control Board in India and simulated adversarial scenarios.
- Attack Resilience: Simulations demonstrated that the Mean-Median algorithm remains robust until the percentage of adversaries reaches approximately 50%. When combined with C-MEV and a reputation system, a malicious coalition would need to control 40% to 60% of the network to successfully execute a Data Poisoning attack.
- Efficiency: The iterative C-MEV approach showed a "drastic decrease" in computational complexity, allowing the system to scale effectively.
- Utility Mapping: Using SHAP (SHapley Additive exPlanations), the researchers identified that specific features, like PM2.5 and PM10 concentrations, held the highest value in their automotive use case. Providers of this data would receive the highest rewards and face the lowest PoW requirements.
Conclusion
This research demonstrates that a hybrid decentralized architecture can effectively address the fairness and security challenges of crowd-sourced data markets. By linking computational work to data utility through Shapley values and optimizing voting mechanisms via C-MEV, the system creates a resilient environment for data exchange. Future work will focus on integrating decentralized identity (DID) solutions to further reduce reliance on central authorities and exploring "Shapley Robust" algorithms to better handle data duplication.
Summary updated