Dash: Semi-Supervised Learning with Dynamic Thresholding

Benchmark Model Rank Results
semi-supervised-image-classification-on-cifar-10-250-labelsDash (RA)–Percentage error: 4.56±0.13
semi-supervised-image-classification-on-cifar-10-4000-labelsDash (RA, ours)–Percentage error: 4.08±0.06
semi-supervised-image-classification-on-cifar-100-10000-labelsDash (RA, WRN-28-8)–Percentage error: 21.97±0.14
semi-supervised-image-classification-on-cifar-100-2500-labelsDash (RA, WRN-28-8)–Percentage error: 27.18±0.21
semi-supervised-image-classification-on-cifar-100-400-labelsDash (CTA, WRN-28-8)–Percentage error: 44.83±1.36
semi-supervised-image-classification-on-cifar-100-400-labelsDash (RA, WRN-28-8)–Percentage error: 44.76±0.96