Convolutional neural network for time-dependent functions

Question

I need to do a dimension reduction from a series of images. More specifically, each image is a snapshot of the ball, and its position and speed are optimal characteristics. As far as I know, CNN is a modern tool to reduce the ability to classify images, but in this case only one frame is provided. Is it possible to extract also time-dependent functions, given the many images at different time steps? Otherwise, are these the most modern methods for this?

This is the first time I use CNN, and I would also appreciate any link or any other suggestion.

Answer 1

If you want the network to somehow recognize a time-dependent progression, you should probably study recurrent neural networks (RNNs). Since you will be working on video, you should study repetitive convolutional neural networks (RCNNs) such as: http://jmlr.org/proceedings/papers/v32/pinheiro14.pdf

Repeatability adds some memory about the previous state of the input. See this nice Carpathian explanation: http://karpathy.imtqy.com/2015/05/21/rnn-effectiveness/

, . , , , ( ), . ( , , , ?)

: , ? "" , . . , , .