New 'Local Shapley' method drastically cuts data valuation computation

Researchers have introduced "Local Shapley," a novel method for data valuation that significantly reduces computational complexity. Unlike traditional approaches that consider all possible training data combinations, Local Shapley leverages the inherent locality of modern predictive models, focusing only on the subsets of data that directly influence a prediction. This approach reframes the valuation problem as a structured data processing task, leading to theoretical lower bounds on retraining operations and enabling efficient algorithms like LSMR and LSMR-A. Experimental results across various model families show substantial reductions in retraining time and computational resources while maintaining high accuracy in data valuation. AI

IMPACT Reduces computational cost for data valuation, potentially accelerating model development and deployment.

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

Read on arXiv cs.LG →

• arXiv
• graph neural networks
• k-nearest neighbors algorithm
• Lester Smith Medical Research Institute
• Local Shapley
• LSMR-A
• Shapley value
• Xuan Yang

• paper
• infra

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