Julia Evans

Day 18: an answer to an autoencoder question

I’m going to keep this one short because I want to get back to coding!

One of the questions I had in the last post was:

The Encoder class in the translation tutorial outputs 2 vectors, an output and a hidden vector. Which one is the encoding, the output or the hidden vector? (or both???)

I’ve been struggling a bit to find out the answers to questions like this – there are 10 million blogs about deep learning, but somehow I feel like they often don’t answer my questions. My best strategy so far has to been to search the PyTorch forums, and that’s how I found the answer to this one! I searched for something like “autoencoder” and found this answer

The answer to that question is that the hidden vector is the encoding, and you just throw out the output.

That answer helped me a LOT and I managed to get a simple autoencoder to work once I had the answer! It’s always surprising to me every single time how helpful it is to focus and articulate the questions I have.

I still don’t know what the relu is for, but it’s nice to have the answer to at least one question.

Day 17: trying to wrap my head around autoencoders Day 20: trying to figure out how Google Cloud IAM works