Matrix representation of neural networks

Abstract

A method includes determining a plurality of rows and a plurality of columns of a matrix. A count of rows of the plurality of rows is determined based on an architecture of a neural network, and a count of columns of the plurality of columns is determined based on a grammar. The method also includes assigning a value to each row/column pair. The value assigned to a particular row/column pair indicates a hyperparameter descriptive of the neural network. The method further includes storing the matrix as a representation of the neural network.

What is claimed is: 
     
         1 . A method comprising:
 determining a plurality of rows and a plurality of columns of a matrix, a count of rows of the plurality of rows determined based on an architecture of a neural network, and a count of columns of the plurality of columns determined based on a grammar;   assigning a value to each row/column pair, the value assigned to a particular row/column pair indicating a hyperparameter descriptive of the neural network; and   storing the matrix as a representation of the neural network.   
     
     
         2 . The method of  claim 1 , wherein the values assigned to each row/column pair of a first row of the matrix represent hyperparameters of a first set of hidden layers of the neural network, and wherein the values assigned to each row/column pair of a second row of the matrix represent hyperparameters of a second set of hidden layers of the neural network, the second set of hidden layers distinct from the first set of hidden layers. 
     
     
         3 . The method of  claim 1 , wherein an order of the rows of the matrix indicates an arrangement of corresponding hidden layers of the neural network. 
     
     
         4 . The method of  claim 1 , wherein a first column of the matrix includes logical values and a second column of the matrix include integer values. 
     
     
         5 . The method of  claim 1 , wherein a particular column of the matrix includes a first row/column pair including a first value indicting an interconnection scheme associated with a first set of hidden layers of the neural network and includes a second row/column pair including a second value indicting an interconnection scheme associated with a second set of hidden layers of the neural network. 
     
     
         6 . The method of  claim 1 , wherein a particular column of the matrix includes a first row/column pair including a first value indicting a first activation function associated with one or more nodes of a first set of hidden layers of the neural network and includes a second row/column pair including a second value indicting a second activation function associated with one or more nodes of a second set of hidden layers of the neural network. 
     
     
         7 . The method of  claim 1 , wherein a particular column of the matrix includes a first row/column pair including a first value indicting a first count of nodes of a first set of hidden layers of the neural network and includes a second row/column pair including a second value indicting a second count of nodes of a second set of hidden layers of the neural network. 
     
     
         8 . The method of  claim 1 , wherein a particular column of the matrix includes, in a first row, a first row/column pair including a first value indicting a first count of adjacent hidden layers of the neural network represented by the first row and includes, in a second row, a second row/column pair including a second value indicting a second count of adjacent hidden layers of the neural network represented by the second row. 
     
     
         9 . The method of  claim 1 , wherein a particular row of the matrix includes multiple row/column pairs that represent mutually exclusive layer structures or hyperparameters. 
     
     
         10 . The method of  claim 1 , further comprising, before determining the plurality of rows and the plurality of columns of the matrix:
 accessing a data set that is to be input to the neural network;   performing a comparison of characteristics of the data set to one or more rules that map the characteristics of the data set to one or more grammars; and   determining the grammar based on the comparison.   
     
     
         11 . The method of  claim 1 , wherein, after storing a population of matrices representing a population of neural networks, the population of matrices including the matrix and one or more additional matrices representing other neural networks of the population of neural networks, the method further comprises:
 providing a first matrix of the population of matrices as input to a machine learning model, the first matrix representing a first neural network of the population of neural networks; and   receiving, as output from the machine learning model, relative fitness data indicating predicted fitness of the first neural network for processing a data set.   
     
     
         12 . The method of  claim 11 , wherein the first matrix is input to the machine learning model row-by-row in an order based on an order of hidden layers of the first neural network. 
     
     
         13 . A computer-readable storage device storing instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:
 determining a plurality of rows and a plurality of columns of a matrix, a count of rows of the plurality of rows determined based on an architecture of a neural network, and a count of columns of the plurality of columns determined based on a grammar;   assigning a value to each row/column pair, the value assigned to a particular row/column pair indicating a hyperparameter descriptive of the neural network; and   storing the matrix as a representation of the neural network.   
     
     
         14 . The computer-readable storage device of  claim 13 , wherein the values assigned to each row/column pair of a first row of the matrix represent hyperparameters of a first set of hidden layers of the neural network, and wherein the values assigned to each row/column pair of a second row of the matrix represent hyperparameters of a second set of hidden layers of the neural network, the second set of hidden layers distinct from the first set of hidden layers. 
     
     
         15 . The computer-readable storage device of  claim 13 , wherein an order of the rows of the matrix corresponds to an order of hidden layers of the neural network. 
     
     
         16 . The computer-readable storage device of  claim 13 , wherein the values assigned to the row/column pairs include two or more of a logical value, an integer, a floating point value, or a string. 
     
     
         17 . The computer-readable storage device of  claim 13 , wherein a particular column of the matrix includes a first row/column pair including a first value indicting an interconnection scheme associated with a first set of hidden layers of the neural network and includes a second row/column pair including a second value indicting an interconnection scheme associated with a second set of hidden layers of the neural network. 
     
     
         18 . The computer-readable storage device of  claim 13 , wherein a particular column of the matrix includes a first row/column pair including a first value indicting a first activation function associated with one or more nodes of a first set of hidden layers of the neural network and includes a second row/column pair including a second value indicting a second activation function associated with one or more nodes of a second set of hidden layers of the neural network. 
     
     
         19 . The computer-readable storage device of  claim 13 , wherein a particular column of the matrix includes a first row/column pair including a first value indicting a first count of nodes of a first set of hidden layers of the neural network and includes a second row/column pair including a second value indicting a second count of nodes of a second set of hidden layers of the neural network. 
     
     
         20 . The computer-readable storage device of  claim 13 , wherein a particular column of the matrix includes, in a first row, a first row/column pair including a first value indicting a first count of adjacent hidden layers of the neural network represented by the first row and includes, in a second row, a second row/column pair including a second value indicting a second count of adjacent hidden layers of the neural network represented by the second row. 
     
     
         21 . The computer-readable storage device of  claim 13 , wherein a particular row of the matrix includes multiple row/column pairs that represent mutually exclusive layer structures or hyperparameters. 
     
     
         22 . A computing device comprising:
 one or more processors; and   a memory storing instructions that, when executed by the one or more processors, cause the one or more processors to:
 determine a plurality of rows and a plurality of columns of a matrix, a count of rows of the plurality of rows determined based on an architecture of a neural network, and a count of columns of the plurality of columns determined based on a grammar; 
 assign a value to each row/column pair, the value assigned to a particular row/column pair indicating a hyperparameter descriptive of the neural network; and 
 store the matrix as a representation of the neural network. 
   
     
     
         23 . The computing device of  claim 22 , wherein the values assigned to each row/column pair of a first row of the matrix represent hyperparameters of a first set of hidden layers of the neural network, and wherein the values assigned to each row/column pair of a second row of the matrix represent hyperparameters of a second set of hidden layers of the neural network, the second set of hidden layers distinct from the first set of hidden layers. 
     
     
         24 . The computing device of  claim 22 , wherein an order of the rows of the matrix indicates an arrangement of corresponding hidden layers of the neural network. 
     
     
         25 . The computing device of  claim 22 , wherein a first column of the matrix includes logical values and a second column of the matrix include integer values. 
     
     
         26 . The computing device of  claim 22 , wherein a particular column of the matrix includes a first row/column pair including a first value indicting an interconnection scheme associated with a first set of hidden layers of the neural network and includes a second row/column pair including a second value indicting an interconnection scheme associated with a second set of hidden layers of the neural network. 
     
     
         27 . The computing device of  claim 22 , wherein a particular column of the matrix includes a first row/column pair including a first value indicting a first activation function associated with one or more nodes of a first set of hidden layers of the neural network and includes a second row/column pair including a second value indicting a second activation function associated with one or more nodes of a second set of hidden layers of the neural network. 
     
     
         28 . The computing device of  claim 22 , wherein a particular column of the matrix includes a first row/column pair including a first value indicting a first count of nodes of a first set of hidden layers of the neural network and includes a second row/column pair including a second value indicting a second count of nodes of a second set of hidden layers of the neural network. 
     
     
         29 . The computing device of  claim 22 , wherein a particular column of the matrix includes, in a first row, a first row/column pair including a first value indicting a first count of adjacent hidden layers of the neural network represented by the first row and includes, in a second row, a second row/column pair including a second value indicting a second count of adjacent hidden layers of the neural network represented by the second row. 
     
     
         30 . The computing device of  claim 22 , wherein a particular row of the matrix includes multiple row/column pairs that represent mutually exclusive layer structures or hyperparameters. 
     
     
         31 . A method comprising:
 determining a first set of values representing hyperparameters of a first set of hidden layers of a neural network, an order of values of the first set of values determined based on a grammar;   determining a second set of values representing hyperparameters of a second set of hidden layers of the neural network, an order of values of the second set of values determined based on the grammar; and   storing the first set of values and the second set of values in a data structure in a manner that indicates an arrangement of the first set of hidden layers and the second set of hidden layers in the neural network.   
     
     
         32 . The method of  claim 31 , wherein the first set of values includes a first value indicating an interconnection scheme associated with the first set of hidden layers or an activation function associated with a node of the first set of hidden layers, and the first set of values includes a second value indicating a count of nodes of the first set of hidden layers. 
     
     
         33 . A computing device comprising:
 one or more processors; and   a memory storing instructions that, when executed by the one or more processors, cause the one or more processors to:
 determine a first set of values based on a first set of hidden layers of a neural network, wherein the first set of values represent hyperparameters of the first set of hidden layers and are arranged in an order based on a grammar; 
 determine a second set of values based on a second set of hidden layers of a neural network, wherein the second set of values represent hyperparameters of the second set of hidden layers and are arranged in the order based on the grammar; and 
 store the first set of values and the second set of values in a data structure in a manner that indicates an arrangement of the first set of hidden layers and the second set of hidden layers in the neural network. 
   
     
     
         34 . The computing device of  claim 33 , wherein the first set of values includes two or more of a logical value, an integer, a floating point value, or a string.