neural_turing_machine

Torch implementation of Neural turing Machine, as presented by Alex Graves, Greg Wayne & Ivo Danihelka.

Original article : https://arxiv.org/abs/1410.5401

Test

To test a pre-trained model on a specific task, just run the associated file :

• Copy task : load_copy.lua
• Repeat copy task : load_rep_copy.lua
• Association recall task : load_assoc.lua

Three plots will pop :

1. The randomly sampled inputs.
2. The outputs generated by the Neural Turing Machine.
3. The difference between the expected outputs and the actual outputs.

#TO DO :

• Enhance Shifter module performances (caching shiftMat allows to reduce execution time by half).
• Implement a the GRU controller.
• Make training parameters configuration easier.