It shows that you can actually encode procedural algorithms in DL models, given lots of training data and the appropriate architecture (in this case, it's hard to do without a RNN)
Show this thread
I updated the addition RNN example script (originally written by @Smerity) to tf.keras. It's a LSTM that learns to add numbers of provided as strings, e.g. take the string "345+67" and returns the string "412".https://colab.research.google.com/drive/1yabxTmMA_-iZ8Dyifta28ux5FxDL90Ak …
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Niceee
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
This was one of the first RNNs I ever trained! Did it during my spare time at
@CommonCrawl. Keras made it joyously simple! I'd written maximum entropy classifiers for NLP but never played with anything beyond a trivial MLP for neural networks. Glad this example lived on :)Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Very cool. What happens if you try to make it add larger numbers? Can you make it do the same thing for multiplication?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.