,
8 tweets,
5 min read
Read on Twitter
I believe a dev can get up to speed on neural networks in 3h and then learn by himself. Ready for a crash course? /1 
Got 3 more hours ? The "Tensorflow without a PhD" series continues. First a deep dive into modern convolutional architectures: .
This session walks you through the construction of a neural network that can spot airplanes in aerial imagery. A good place to start for software devs who know some basics (relu, softmax, ...) and want to see a real model built from scratch.
The model uses the latest Tensorflow high-level APIs like tf.layers, Estimator or the Dataset API (👍) and I ran 222 trainings on Google Cloud ML Engine to get it right -which shows that I am not yet very good at this 😅. I gave up my local GPU for ML Engine. It's a better tool.
Google Translate in 10 lines of Tensorflow code ? Now that I’ve got your attention (pun intended 😁), come and see what you can do with recurrent neural networks, attention mechanisms and Tensorflow’s seq2seq API.
The four steps of RNN architecture were laid out by @honnibal in this post: explosion.ai/blog/deep-lear…. We use them in the "modern RNN architectures" video () to build a translation model and a toxic comment detector.
@honnibal Toxic comment detection model designed by co-speaker @nithum from Google's @JigsawTeam. And yes, this session contains profanity 🤬 and ☠️ strong 🐷 language 💩, for educational and research purposes of course 😇
A short primer on reinforcement learning and policy gradients: . The “Tensorflow and deep learning without a PhD” cycle is now complete. Thank you for watching.
