Day #136 - RNN Session

Notes from session - Andrej Karpathy, Research Scientist, OpenAI - RE•WORK Deep Learning Summit 2016 

RNN

• Operate on Sequences
• Image Captioning (Produce Sequence of words)
• Sentiment Classification / Machine Translation
• Video Classification
• RNN has a state which update everytime when vector comes in x
• RNN can process vectors so it is encoded
• Use RNN to generate sequences after learning from dataset
• Interesting examples of RNN code, function, bogus comments, syntactically very few mistakes
• min-char-rnn.py, char-rnn projects to look out for (https://gist.github.com/karpathy/d4dee566867f8291f086)
• Detect and describe objects in image in single pass

Happy Learning!!!