|
I was wondering if anybody had some good references on dynamic neural networks. The goal is to have a neural network that responds in real-time (and is governed by say a system of DEs) instead of at discrete time-steps. The prototypical application is a robot: we have a series of sensors that are constantly feeding data into the input neurons, and we have output neurons that are constantly telling the motors what to do. We want to optimize this robot to do some tasks. Is there a nice recent survey on dynamic neural nets? In particular one that discusses recurrent DNN, feedforward DNN, and DNN with and without learning. Alternatively, if continuous dynamic neural networks are not the right approach for this sort of task, then suggestions of other types of neural nets are also welcome. This question is cross posted from Cross Validated. |
|
It sounds like you're referring to Spiking Neural Networks(SNN), which generally employ much more complex neuron models than traditional ANNs, when you say dynamic neural networks. This isn't an area I'm very familiar with, so I can't offer much information besides the terminology which might yield useful results from google. As far as control, recently I've seen some papers about using chaotic neural networks to do things like control gait in robots. For example this one. Perhaps this could be another interesting direction to look? Also, in general it sounds like you're more interested in reinforcement learning. RL can be done with neural nets (for example Tesauro's famous td-gammon network), but it is by no means the only way to approach the problem. There is a online version of Sutton and Barto's book here, which is a great place to get started with reinforcement learning. |
I would look into ALVIN, a self-driving car controlled by neural networks from 1990 (!). What task do you want to do? If you're controlling a robot arm or deciding between a few simple actions, you can probably use an NN (and a feedforward or recurrent net will be easier). If you're looking for more complex control, such as coordination of many parts, or planning, you're out of luck.