Random sampling outperforms complex methods for AI data quality profiling

A new research paper introduces a benchmark for data quality profiling in large-scale AI pipelines, evaluating nine different sampling strategies. The study found that simple, schema-free random uniform sampling performed best, achieving the lowest mean relative error on various real-world and synthetic datasets. In contrast, more complex proxy-guided methods, such as those using Metropolis-Hastings or directed acyclic graphs, were significantly less effective and computationally more expensive, especially at scale. AI

IMPACT Highlights the importance of efficient data quality profiling for AI pipelines, suggesting simpler methods may be more effective at scale.

RANK_REASON Academic paper detailing a new benchmark and findings on sampling strategies for data quality profiling. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Data-Centric AI Pipelines
• Data Quality Profiling
• directed acyclic graph
• Laure Berti-Equille
• NYC 311
• NYPD arrests
• Progressive sampling-based Bayesian optimization for efficient and automatic machine learning model selection
• UCI Adult
• Ultra-Marathon Running

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →