Main Invention:
"Key to our approach is a novel differentiable approximation of the value-iteration algorithm, which can be represented as a convolutional neural network, and trained end-to-end using standard backpropagation."
Comment:
Although the author used a very clever way to embed a RL problem in CNN like network, it's rarely used in public papers or projects somehow.
And it can only be applied to low dimension domain due to it's nature. There are some papers extend this capability though.
Not sure if it could be combined with HRL or maybe HRL itself will do the job without any VIN.
Best Paper Award NIPS 2016.
Video: MSDN Channel 9
Main Invention: "Key to our approach is a novel differentiable approximation of the value-iteration algorithm, which can be represented as a convolutional neural network, and trained end-to-end using standard backpropagation."
Comment: Although the author used a very clever way to embed a RL problem in CNN like network, it's rarely used in public papers or projects somehow. And it can only be applied to low dimension domain due to it's nature. There are some papers extend this capability though. Not sure if it could be combined with HRL or maybe HRL itself will do the job without any VIN.