Tuesday, April 16, 2019

Convolution Neural Network :

Convolution Neural Network

Nowadys, Convolution Neural Network (CNN) is a hot topic for a research over unstructured data such as images, speech, time series and many more.
Two main reason for understanding CNN architeture is:

  1. It can produce an modified signal which is better than original one.
  2. It can learn through various filters with their respective weights and bias.

Actually what is convolution operation ?

To do convolution operation, both input and kernel signal should satisfy following properties:

  • A signal is linear if it have: a) scaling property and b) super-position property: Let the input and output signal can be written as:
    y(t) = f(x(t)) .......... (1)
    Equation (1) should follow scaling and super-position properties.
  • And a that signal should be Time Invarient : A time-independent signal can be expressed as
    f(x(t-k)) = y(t-k) .......... (2)

    Basic architecture of CNN:

    The basic architecture of CNN include Conv-Relu_maxpool layer along with end attaching fully connected layer as shown below.

    Each component of CNN is described as below:

    Posted by at No comments:
    • Subscribe to: Posts (Atom)