Code & Data

Cloud Computing Resources:

Free TPU time (~ a Google GPU) is available from Google Colab. Colab is basically a Python notebook in your browser that executes the code on a remote machine in the Google Cloud. This tutorial is a reasonably good one to get started training an ML model on Colab.

To get data off the machine where the code is executed (for example model weights), Google want you to create a "data bucket" in their cloud storage service. They give you $300 free credit for this. This is all covered in the tutorial linked above.

• Colab: free cloud computing resource by Google, for educational purposes
• Colab TPU tutorials
• Google Cloud Storage: storage buckets that can be linked to Colab instances

Deep learning software packages

These packages offer automatic differentiation and numerics on GPU.

• Pytorch: Facebook's package
• Tensorflow: Google's package

Probabilistic programming software packages:

These packages allow you to specify an arbitrary probabilistic model then performance inference on it.

• Edward: Tensorflow backend
• Pyro: Pytorch backend

Repositories:

• Aggregated lists:
  • https://paperswithcode.com/
  • https://github.com/zziz/pwc
  • https://github.com/nashory/gans-awesome-applications
• Progressive GAN: (includes pre-trained models)
• Generative Models @ ModelZoo: https://modelzoo.co/category/generative-models