I have a really large network that I need to train but I don't have enough computational power. I tried using Colab but colab doesn't allow multiple parallel runtimes. I'd like to use the caiver approach for finetuning my learning rate and other hyperparameters but I can't afford to babysit as it'll take me years. Could you suggest any online resources like Colab, Kaggle and such?

  • 1
    $\begingroup$ That is a really good question. I have thought the same thing. I have not done an exhaustive search but I doubt there are a lot of options out there. My own solution is to cut the problem down to a reasonable size to run on my own hardware. I take the divide and conquer approach as much as I can. Sorry, I can't be more helpful. $\endgroup$
    – mccurcio
    Commented May 2, 2020 at 20:54
  • 1
    $\begingroup$ What about datalore.io ? They give you up to 3 parallel computations. Not sure if there other community-level limitations impact you/make your problem worse. $\endgroup$
    – mwilson
    Commented May 3, 2020 at 1:41


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.