Paper: Speeding up the Hyperparameter Optimization of Deep Convolutional Neural Networks


#1

The paper (https://arxiv.org/abs/1807.07362) mentioned that to tune the hyperparams, they can use lower resolution images with little loss in accuracy for the final model-- I just thought that it aligns with the trick that Jeremy said about training networks initially with smaller images. I think he said there hasn’t been any paper that talks about this, so maybe this one can yield some extra insights…