| CPC G06N 3/044 (2023.01) [G06N 3/08 (2013.01)] | 18 Claims |
|
1. A computer-implemented method comprising:
training a functional state of a recurrent artificial neural network that includes two thousand nodes or more to propagate an input according to a learned signal flow within the recurrent artificial neural network until processing of the input reaches a predefined end and the recurrent artificial neural network returns to a state in which only background or no signal transmission activity occurs, the training comprising:
structuring the recurrent artificial neural network to obtain a trained recurrent artificial neural network, the structuring comprising:
reading a plurality of digits output from the recurrent artificial neural network, wherein each digit represents whether or not activity within a particular group of nodes in the recurrent artificial neural network comports with a respective pattern of activity; and
evolving a structure of the recurrent artificial neural network, wherein evolving the structure of the recurrent artificial neural network comprises:
characterizing a complexity of each pattern of signal transmission activity in the structure, wherein the complexity is measured based on simplex counts or Betti numbers of the pattern,
using the characterization as an indication of whether the structure has optimized a reward, and
iteratively evolving the structure if the reward has not been optimized;
processing the input by the trained recurrent artificial neural network, the input comprising a stream of video or audio data;
monitoring complexities of patterns of signal transmission activity that occur between three or more nodes in the trained recurrent artificial neural network in response to the processing of the input, the monitoring comprising, during the processing of the input:
outputting, by the trained recurrent artificial neural network, a second plurality of digits, wherein a value of each digit in the second plurality of digits represents whether or not signal transmission activity that occurs between the three or more nodes in the trained recurrent artificial neural network in response to the processing of the input comports with a respective pattern of signal transmission activity, regardless of a location at which the respective pattern of signal transmission activity occurs in the trained recurrent artificial neural network, wherein the second plurality of digits enables an isomorphic topological reconstruction of a graph of the trained recurrent artificial neural network;
determining the complexity of each pattern of signal transmission activity;
identifying a time at which an object or sound is recognized in the stream of video or audio data, comprising identifying, based on the complexity of each pattern of signal transmission activity, a distinguishable level of complexity of the signal transmission activity, wherein the distinguishable level of complexity comprises an upward deviation, downward deviation, or change of complexity from a predetermined threshold level of complexity;
outputting, at the identified time and before the trained recurrent artificial neural network completes the processing of the input to reach the predefined end and return to the state in which only background or no signal transmission activity occurs, the patterns of signal transmission activity that have the distinguishable level of complexity; and
transmitting, encrypting, or storing the output patterns of signal transmission activity that have the distinguishable level of complexity; and
allowing the trained recurrent artificial neural network to return to the state in which only background or no signal transmission activity occurs.
|