CSE 4311 - Assignments - Assignment 2
Neural Network File Format

In the following specifications, red color indicates placeholders that are replaced by actual numbers in the file:

• The first line follows this format:

  layers: number_of_layers

  Here, number_of_layers is an integer specifying the number of layers in the network (note that the input layer is counted as layer #1).
• The second line follows this format:

  units: J1 J2 ... Jnumber_of_layers

  Here, the string units: is followed by number_of_layers integers. The L-th integer is denoted as J_L, and it specifies the number of units in the L-th layer.
• After the second line, the rest of the file consists of a sequence of chunks, where each chunk provides information about the bias and weight values of the corresponding layer. We assume that all layers are fully connected (meaning that all units of a layer receive inputs from all units in the previous layer). Note that the first chunk describes layer 2, since layer 1 (the input layer) does not contain any bias or weight values. The chunk describing layer L is structured as follows:
  • The first four lines of the chunk follow this format:

    start layer
    start bias
    bL,1 bL,2 ... bL,JL
    end bias
    

    Here, we denote by b_L,i the bias value for the i-th unit of layer L, and we denote by J_L the number of units in layer L.
  • The next lines of the chunk follow this format:

    start w
    wL,1,1 wL,1,2 ... wL,1,JL-1
    wL,2,1 wL,2,2 ... wL,2,JL-1
    ...
    wL,JL-1,1 wL,JL-1,2 ... wL,JL-1,JL-1
    end w
    

    Here, we denote by w_L,i,j the weight value for the edge connecting the output of the j-th unit of layer L-1 to the input of the i-th unit of layer L. As before, we denote by J the number of units in layer L, and therefore J_L-1 is the number of units in the preceding layer L-1. Essentially, the weights for layer L are represented as a matrix of J_L rows and J_L-1 columns.