Data formats processed by the platform

The basic Kaspersky Neuromorphic Platform data format (hereafter KNP format) is used to store input and output data as a time series. The base format file contains the following information:

• File header containing the following information:
  • File format version
  • Array of data senders' UIDs for each step of the neural network execution (optional)
  • Array of tags
• The input or output data is a time series consisting of a list of structures with the following information:
  • Neural network execution step when the spikes were generated
  • Indices of the neurons that generated spikes at this step

Using the Kaspersky Neuromorphic Platform functions, you can convert files in the basic format to SpikeMessages.

KNP messages are serialized in the following notations:

• HDF5 is the main notation for storing data in a binary form
• JSON is a JavaScript object notation that can be used to debug a neural network

In addition to the basic format, the input data can also be stored and transmitted as CSV files, which contain the indices of the neurons that generated the spikes. CSV files with the input data can be passed to the objects of the IndexConverter class for further processing in the input channels.

Page top