Application Deployment

  • Extract significant features from sensory and data inputs and convert them into patterns of 256 bytes before broadcast to the neurons
  • Multiple feature extractions can occur in parallel on the same input data stream or on different input streams
  • Each type of feature is used to train one NeuroMem neural network (1 feature per wavelength; 1 feature per focal length, etc)
  • Control the external and programmable events which can trigger the learning of a new or existing category
  • In the case of unsupervised learning, define the category to learn
  • Control which response(s) to retrieve from the neurons
    • A simple recognition status (identified, uncertain or unknown), a best match, or the list of the K closest matches
    • Consolidate the responses per neural network and accross multiple neural networks if applicable
  • Format the responses when positive recognition occurs
    • Time stamps, locations in a video frame, etc.
  • Based of the recognized content, actuate, store, transmit, etc.