In this post we are going to learn about the different components, types and
architecture of ANN
To understand how
ANNs work, we need to know some of the main concepts and terminologies that are
used to describe them. Here are some of the most important ones:
· Weights: These are the
numerical values that are associated with the links between the neurons. The
weights represent the strength and direction of the connection between the
neurons. The weights can be positive or negative and can be represented by a
matrix or a vector. The weights are the main parameters that are learned and
adjusted by the network during the learning process.
·
Bias: This is a constant
value that is added to the weighted sum of the inputs of a neuron. The bias
represents the offset or shift of the output of the neuron. The bias can be
positive or negative and can be represented by a scalar or a vector. The bias
is another parameter that is learned and adjusted by the network during the
learning process.
·
Threshold: This is a constant
value that is compared to the net input of a neuron to determine the output of
the neuron. The threshold represents the minimum or maximum value that the net
input must reach or exceed to activate the neuron. The threshold can be
positive or negative and can be represented by a scalar or a vector. The
threshold is sometimes used as a simple activation function for the neurons.
·
Learning rate: This is a constant
value that is multiplied by the error term to update the weights and bias of
the neurons. The learning rate represents the speed and magnitude of the
learning process. The learning rate can be positive or negative and can be
represented by a scalar or a vector. The learning rate is usually chosen by the
user or the algorithm and can be fixed or variable.
·
Target value: This is the
desired or expected output of the network for a given input. The target value
represents the goal or objective of the learning process. The target value can
be continuous or discrete and can be represented by a scalar or a vector. The
target value is usually provided by the user or the data and can be known or
unknown.
·
Error: This is the
difference between the actual output and the target output of the network for a
given input. The error represents the accuracy or performance of the network.
The error can be positive or negative and can be represented by a scalar or a
vector. The error is usually calculated by a loss function or a cost function and can be absolute or relative.
· Activation
function: This is a mathematical function that is applied to the net input of a
neuron to produce the output of the neuron. The activation function represents
the non-linearity or complexity of the network. The activation function can be
linear or non-linear and can be represented by an equation or a graph. The
activation function is usually chosen by the user or the algorithm and can be
fixed or variable.