Learning was more efficient today, though I didn't work as long as I want. Last night I went to bed at midnight after watching a good movie, today I got up late and started work at 11.

• Reviewed information theory and made Anki cards. I understand KL divergence well now.
• Lots of ML videos, learned about Resnets, BatchNorm, Adam and RMSProp,
• Improved my understanding of ConvNets, played with LeNet
• Played with adversarial stuff a little. Implementing interpretability stuff is a good idea, could make cool demos too.