Reproduce “Data Valuation by Fusing Global and Local Statistical Information” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xiaoling Zhou, Ou Wu, Michael K. Ng, Hao Jiang
Zhou et al. introduced GLOC, a data valuation method that integrates global (Gaussian) and local statistical properties of data values to refine Shapley estimations, achieving up to 206-fold lower MSE. Their IncGLOC and DecGLOC methods further enable efficient dynamic data valuation, inferring updated values up to 506 times faster than existing baselines.
Key insights
- Empirical analysis revealed that data value distributions more closely follow a Gaussian distribution than a Laplace distribution across various datasets.
- Data values exhibit strong local consistency, where samples within the same category have similar values and those from different categories are distinct.
- Changes in data values due to dynamic data modifications (addition or removal of samples) are relatively minor, suggesting that updated values can be efficiently inferred.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.