Data Provenance
Overview
Data provenance studies the origins, transformations, and relationships of data and the outputs built from it, helping people reproduce analyses, assign credit, audit decisions, and manage privacy or reuse rights. Common approaches include workflow metadata and lineage graphs, structured claim-to-evidence records, cryptographic signatures and append-only logs, and watermarks or statistical fingerprints for identifying sources. Work increasingly applies these methods to AI training data, generated media, retrieval systems, and tool-using agents, where records can connect outputs to sources, actions, and versions. A central challenge is that tamper-evident records or plausible citations do not by themselves prove that an event occurred or that evidence supports a claim; provenance must balance completeness and verifiability with privacy, cost, and robustness to transformations.
Latest papers
191 papers35 papers in the last four weeks, up 59% on the four weeks before. 0.2% of all new papers.