I'm curious to see if they can apply this to industry-standard tests like ImageNet classification. The workloads that they test on make it hard to quantify the

I'm curious to see if they can apply this to industry-standard tests like ImageNet classification. The workloads that they test on make it hard to quantify the broad applicability of their work.