New Gram determinant score assesses dataset reliability without ground truth

Researchers have introduced a novel method called the Gram determinant score to assess dataset reliability without needing ground truth data. This score measures the volume spanned by vectors representing the empirical distribution of observed data and experiment outcomes. Experiments on synthetic data, CIFAR-10 embeddings, and real employment data indicate that the Gram determinant score effectively captures data quality across various observation processes, maintaining reliability rankings regardless of the specific experiment. AI

IMPACT Provides a new metric for evaluating data quality in AI model training without ground truth, potentially improving model robustness.

RANK_REASON Academic paper introducing a new scoring method for datasets. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• CatalyzeX Code Finder for Papers
• CIFAR-10
• Connected Papers
• CORE Recommender
• DagsHub
• Gotit.pub
• Gram determinant score
• Hugging Face
• IArxiv Recommender
• Litmaps
• ScienceCast
• scite Smart Citations
• Shi Feng

• paper
• other

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