0.25
0.5
0.75
1.25
1.5
1.75
2
Neural Programmer-Interpreters
Published on May 27, 20165874 Views
We propose the neural programmer-interpreter (NPI): a recurrent and compositional neural network that learns to represent and execute programs. NPI has three learnable components: a task-agnostic recu
Related categories
Chapter list
Neural Programmer-Interpreters00:00
Neural Programmer Interpreter (NPI) goals00:03
Model00:38
NPI training data00:41
Time step01:12
Demos02:44
Adding numbers together - environment02:47
Adding numbers together - learned programs03:23
Adding numbers together03:37
Bubble sort - environment04:35
Bubble sort - learned programs04:57
Bubble sort05:04
3D car models - environment05:35
3D car models - learned programs06:09
Canonicalizing the view of 3D car models06:17
Detailed example06:40
Experiments08:34
Data Efficiency - Sorting08:36
Generalization - Sorting09:37
Generalization - Addition10:18
Generalization - Addition problems - 110:39
Generalization - Addition problems - 211:05
Generalization - Addition problems - 311:25
Generalization - Addition problems - 411:31
Generalization - Addition problems - 511:39
Multi-task NPI - Core is shared across all programs11:44
Learning new programs with a fixed NPI core - 112:38
Learning new programs with a fixed NPI core - 213:17
Quantitative Results13:37
Conclusions & Next Steps14:02
Related work14:32
Thanks!15:25