--SSO TD (Temporal difference) is a reinforcement learning concept, yes. The TD in TDNN is however not related to reinforcement learning but stands for Time Delay or Tap Delay. I.e. you just keep a backlog of previous inputs and feed them into the net each time. Your net will be lost without the capability to handle temporal information an TDNNs are the simplest form of nets that can handle it. -- Anon |
--SSO |