.png){kind=icon}
.png){kind=icon}
Return
Shapley value-based data valuation for machine learning data markets
.png){kind=icon}
OA
AI
DOI：10.1007/s42452-025-07328-z.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Data valuation, the process of assigning monetary worth to data based on its utility, impact, and scarcity, has become increasingly critical in today’s digital landscape. As data becomes a valuable asset for organizations, governments, and individuals, effective data valuation supports informed decision-making related to the acquisition, sharing, and monetization of data. The Shapley Value, a game-theoretic concept, has emerged as a prominent approach to data valuation due to its equitable distribution of value among contributors. In this paper, we investigate the use of Shapley Value-based data valuation in Machine Learning Data Markets (MLDM). MLDM is a distributed ML platform where data is traded among agents developing individual models for a given problem. Accurate valuation is, thus, crucial for ensuring fair and efficient exchanges. We introduce the Gain Data Shapley Value (GDSV), a new data valuation method for MLDM, and conduct a comprehensive empirical study to compare its effectiveness against performance-based data valuation. Our findings demonstrate that considering the contribution of data sets to performance scores can lead to systematic improvements in learning performance in MLDM.
Keywords:
Multi-agent systems
Data valuation
Machine learning
Data markets
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
D
IF:
2.4
Papers:
1.6K
Citations:
492