Systems and methods for adaptive training neural networks

Abstract

The present disclosure relates to systems and methods for creating and training neural networks. The method includes collecting a set of signals from a database; applying a transform to each signal to create a modified set of signals, wherein signals of the modified set of signals are wavelets; iteratively, for each of a subset of the modified signals: training the neural network using a modified signal of the subset by adding at least one node to the neural network in response to an error function of an analysis of the modified signal exceeding a threshold; removing nodes from the neural network with activation rates below an activation rate threshold; and grouping each node into a lobe among a plurality of lobes, wherein nodes belonging to a lobe have a common characteristic.

1 - 20 . (canceled) 
     
     
         21 . A system for training a neural network, comprising:
 a memory storing instructions; and   at least one processor configured to execute the instructions to perform operations comprising:
 receiving a neural network comprising one or more nodes; 
 adding at least one first node to the neural network in response to an error function of an analysis of a wavelet signal exceeding a threshold; 
 removing at least one second node from the neural network with an activation rate below an activation rate threshold; and 
 grouping at least one third node into a lobe based on a common characteristic. 
   
     
     
         22 . The system of  claim 21 , wherein adding the at least one first node comprises:
 determining an output vector of the at least one first node by:
 determining a plurality of products of input unit vectors and corresponding magnitudes; 
 determining a first sum of the plurality of products of input unit vectors; and 
 setting the output vector based on a product of the first sum and a node matrix of the at least one first node, 
 wherein the node matrix represents the wavelet signal. 
   
     
     
         23 . The system of  claim 22 , wherein the error function represents a distance between the output vector and a desired output vector. 
     
     
         24 . The system of  claim 22 , wherein the node matrix stores frequencies and magnitudes that make up the wavelet signal. 
     
     
         25 . The system of  claim 21 , wherein the activation rate represents an impact of the at least one second node on an outcome of the neural network. 
     
     
         26 . The system of  claim 21 , wherein the wavelet signal corresponds to a permutation of events represented as an oscillation. 
     
     
         27 . The system of  claim 21 , wherein the at least one processor is further configured to execute the instructions iteratively for a plurality of data signals, wherein the plurality of data signals are transformed into a plurality of individual wavelet signals. 
     
     
         28 . The system of  claim 27 , wherein the operations further comprise:
 trimming the plurality of data signals based on a subset of the plurality of data signals with a low frequency.   
     
     
         29 . The system of  claim 27 , wherein the operations further comprise:
 identifying a dynamic set of the plurality of individual wavelet signals,
 wherein a size of the dynamic set increases until the neural network reaches a threshold accuracy. 
   
     
     
         30 . The system of  claim 29 , wherein the dynamic set of the plurality of individual wavelet signals is identified based on one or more characteristics of the individual wavelet signals. 
     
     
         31 . A method for training a neural network, comprising:
 receiving a neural network comprising one or more nodes;   adding at least one first node to the neural network in response to an error function of an analysis of a wavelet signal exceeding a threshold;   removing at least one second node from the neural network with an activation rate below an activation rate threshold; and   grouping at least one third node into a lobe based on a common characteristic.   
     
     
         32 . The method of  claim 31 , wherein adding the at least one first node comprises:
 determining an output vector of the at least one first node by:
 determining a plurality of products of input unit vectors and corresponding magnitudes; 
 determining a first sum of the plurality of products of input unit vectors; and 
 setting the output vector based on a product of the first sum and a node matrix of the at least one first node, 
 wherein the node matrix represents the wavelet signal. 
   
     
     
         33 . The method of  claim 32 , wherein the error function represents a distance between the output vector and a desired output vector. 
     
     
         34 . The method of  claim 32 , wherein the node matrix stores frequencies and magnitudes that make up the wavelet signal. 
     
     
         35 . The method of  claim 31 , wherein the activation rate represents an impact of the at least one second node on an outcome of the neural network. 
     
     
         36 . The method of  claim 31 , wherein the wavelet signal corresponds to a permutation of events represented as an oscillation. 
     
     
         37 . The method of  claim 31 , further comprising: repeating the method iteratively for a plurality of data signals, wherein the plurality of data signals are transformed into a plurality of individual wavelet signals. 
     
     
         38 . The method of  claim 37 , further comprising:
 trimming the plurality of data signals based on a subset of the plurality of data signals with a low frequency.   
     
     
         39 . The method of  claim 37 , further comprising:
 identifying a dynamic set of the plurality of individual wavelet signals,   wherein a size of the dynamic set increases until the neural network reaches a threshold accuracy.   
     
     
         40 . The method of  claim 39 , wherein the dynamic set of the plurality of individual wavelet signals is identified based on one or more characteristics of the individual wavelet signals.