Skip to content

On Quality Metrics

Mahdi S. Hosseini edited this page Aug 27, 2020 · 3 revisions

This page includes the metrics developed in AdaS paper to monitor the quality of CNN training using the knowledge gain and mapping condition.

Knowledge Gain Vs. Mapping Condition - CNN Quality Metrics

Evolution of knowledge gain vs. mapping condition across different training epochs using ResNet34 on CIFAR10. Different colours represnets different conv blocks, and transparency is correlated to training epoch--lower transparency means higher epoch

OneCycle gif StepLR gif Adam gif

AdaBound AdaS 0.800 gif AdaS 0.975 gif

Clone this wiki locally