Why do I get better performance for the full dataset compared to the subset?
Even if I do not consider the time for subset selection, the 1% line becomes close to the full dataset line but still cannot beat it. I am using CIFAR10, 128 batch size, ResNet20, NVIDIA RTX A6000 GPU, and the same hyperparameters for the full set and subset.
Please help.
Why do I get better performance for the full dataset compared to the subset?
Even if I do not consider the time for subset selection, the 1% line becomes close to the full dataset line but still cannot beat it. I am using CIFAR10, 128 batch size, ResNet20, NVIDIA RTX A6000 GPU, and the same hyperparameters for the full set and subset.
Please help.