Reproduce “Data Pricing in Machine Learning Pipelines” on OpenResearch

About this paper

Zicun Cong, Xuan Luo, Pei Jian, Feida Zhu, Yong Zhang

This survey provides a comprehensive overview of data pricing principles and developments across the entire machine learning pipeline, from data collection and labeling to collaborative model training and model deployment. The work categorizes existing methods for pricing raw data, data labels, and machine learning models, and highlights the introduction of 'effort elicitation' as a key desideratum for data pricing.

Key insights

• The paper introduces 'effort elicitation' as a new, crucial desideratum for data pricing models, particularly relevant for crowdsourced data labeling.
• It structures the complex field of data pricing by segmenting the machine learning pipeline into distinct stages: training data collection, collaborative model training, and ML model deployment.
• The survey highlights the interdisciplinary nature of data pricing, integrating concepts from economics, game theory, privacy, and machine learning to address specific challenges in each pipeline stage.

Summary by alphaXiv.