478 / 2024-04-25 16:13:59

Ex Ante Data Valuation

Data value,Transfer learning

In the increasingly data-driven decision-making landscape, the emergence of various data markets has facilitated transactions among a wide array of participants. A burgeoning challenge these stakeholders face is the accurate ex-ante assessment of data value, especially when access is limited to a small trial data sample. To address this, we introduce a novel approach that amalgamates transfer learning and a widely adopted post hoc data valuation method to evaluate data value prior to transaction. This approach synthesizes potential buyers' prior knowledge of data distribution with new insights gleaned from the trial data sample. We provide a theoretical analysis of the error bound of the proposed approach and evaluate its performance using both real-world and simulated data. The theoretical and empirical results underscore the approach's potential to assist data buyers in making informed acquisition decisions across a range of scenarios.