fastai-fall2018

Notes from the USCF Deep Learning course (fast.ai v3)

Resources

• Course Guide
• Forum

Blog posts ideas

• Week One: Building a Facial Recognition App to Make it Through Your Significant Other's Thanksgiving Dinner
• Week Two: Training a Linear Regression Model with Fast.ai and PyTorch
• Week Three: Pizza Multi-Label Classifier
• blog post about untarring and tarring data
• Regular expressions guide
• Flashcards for powers of 2
• Flashcards for binary numbers
• Create Anki deck for Fast.ai
• Research / create easy way to create anki deck
• research how to connect anki deck to github
• learn scientific notation
• salad bowl DL game

Week one

• Bring water bottle to class!
• Apply Google discount to GCP
• Apply AWS discount to AWS
• Set up Slack for Group 23
• Set up FloydHub
• Create a "Platform: FloydHub" page in fast.ai forum
• Submit PR to the v3 courseware repo for FloydHub
• Complete notebook tutorial
• Complete lesson 1
• Complete download images notebook
• Read Class One resources
• train your own dataset
• create an app
• Try out Gist.it
• Review Python resources
• Do numpy tutorial
• Set up GCP
• Set up AWS
• Set up Paperspace
• Set up Salamander
• Set up SageMaker

Week two

• Kaggle setup and integration for FloydHub
• IDEA: children's artwork GAN
• IDEA: YC company logo - will they fail or not based on their logo
• set up Starlette/responder on serving jobs
• spit things to tensorboard with fastai
• np.argmax (look this up)
• BLOG: Adam Geitgey (blogger) https://medium.com/@ageitgey
• animated through the images like a slot machine, or pull from google images (frog project)
• make gif from a jupyter notebook (visaliing the animation)

Week Three

• Add FloydHub to Production section
• JOSHFP blog post about SGD. this is insanely useful.

Week Four

• finish blog post about cousins
• pizza categorization
• move week 3 notes (and these) to github
• language model that understands code (write lex, tokenizer) (sloppy code, good code). downloda this from leetcode?. this could grade your code as you type. will it compile before you run it. will it do what you want?
• html / ast (tree models)
• something with computer chips like 8-bit guy