Perl AI Deep Learning Tutorial

What is a weight?

Weights in deep learning are the "terms multiplied by inputs" of a function that transforms m inputs into n outputs.

Below is a function that converts two inputs into three outputs. The weights are expressed as column-major matrix.

use strict;
use warnings;

sub m_to_n_convert {
  my ($inputs) = @_;
  
  my $outputs = [];
  
  #Bias
  my $biases = [
    0.12,
    0.43,
    0.16,
  ];;
  
  # weight
  my $weights_mat = {
    rows_length => 3,
    columns_length => 2,
    values ​​=> [
      0.25,
      0.58,
      0.13,
      0.43,
      0.98,
      0.47,
    ]
  };;
  my $weights_mat_values ​​= $weights_mat->{values};
  
  #Weight is a term that is multiplied by the input of a function that transforms m inputs into n outputs.
  $outputs->[0] = $weights_mat_values->[0] * $inputs->[0] + $weights_mat_values->[3] * $inputs->[1] + $biases->[0];
  $outputs->[1] = $weights_mat_values->[1] * $inputs->[0] + $weights_mat_values->[4] * $inputs->[1] + $biases->[1];
  $outputs->[2] = $weights_mat_values->[2] * $inputs->[0] + $weights_mat_values->[5] * $inputs->[1] + $biases->[2];
  
  return $outputs;
}

The number of rows of weights is the same as the number of outputs. The number of columns of weights is the same as the number of inputs.

Weight is a parameter that is automatically adjusted

Weights are parameters that are automatically adjusted using a learning algorithm. The value of loss function, which is an index of error, is updated to be smaller. Parameter update algorithms such as gradient descent are used for the update.