I have developed a MultiLayer Backprobagating Net, and I''am trying to put it to some sort of use, particulary for Recognising for example Tricks performed by a Player in a Snowboard game. My question, what is the best way to train it? How to you deal with time, as certain actions happen over time, how do you feed the data into the net(Or best method for sampling), is Time one of the Inputs to the net, or is it delta time. How do you teach a net to disregard normal elements, e.g. If you utilize a NN for mouse movement detection, and say you wish to teach the net how to recognise a circular action a) How do I teach the net NOT to recognise normal mouse usage and only recognise a circular motion. (Do I have feed normal mouse movement noise to the net first) Thanks.