Are 1000 Seconds Enough to Train?

Designing a single strategy with 3 datasets—MNIST, Fashion MNIST, and CIFAR 10—to achieve near SOTA accuracy within 1000 seconds

This is an experiment where several optimization techniques for faster convergence have been tried for MNIST, Fashion MNIST, and CIFAR 10 dataset, with the only restriction of 1000 seconds on the Google colab provided GPUs. This is helpful when building in-house models or experimenting with a dataset, this technique can be used as base case …

data-augmentation deep learning machine learning resnet