4 ms·

Can someone please ELI5 what this does and why/where it is/can be useful?

• tempodox 8y ago
  Don't trust any machine learning algorithm that you haven't faked yourself. You can make random noise mean anything you want.
• cgearhart 8y ago
  His stated purpose was image compression (although I didn’t see evidence that it worked). If the distribution encoded by the is smaller than your image, then you can send the small set of model parameters (instead of the image) and then use the model to reconstruct the target image.