Emergent neural turing machine and its visual navigation
Zejia Zheng 1,Xiang Wu 2,Juyang Weng 1“Emergent Neural Turing Machine and Its Visual Navigation.” Neural Networks, vol. 110, 2019, pp. 116–130.
ideas:
1.we improve DN-1 with several new mechanisms and presenta new emergent neural Turing Machine — Developmental Network 2 (DN-2). DN-2 grows complex hierarchies by dynamically allowing initialization of neurons with different ranges of connection.
2.We present a complex task — vision guided navigation in simulated and natural worlds using DN-2.
Challenge:
1.It is harder to learn clean logic by using emergent representation.In Weng (2011), Weng argued that DN-1, a connectionist model, learns a Turing Machines(TM) error-free with emergent representation, thus the DN family resolves the scruffy abstraction problem and has clean logic inside.
2.But DN-1 lacks hierarchy in its internal representations, and is hard to handle the complex visual navigation tasks.
Overview:
Method:
Experiments: