Achieving 10,000x training data reduction with high-fidelity labels

Google Research Blog ·

Google reports a labeling method that can reduce required training data by 10,000x while preserving model performance.

Categories: Research

Excerpt

Human-Computer Interaction and Visualization