Understanding Convolutional Neural Networks

Convolution is the process of merging two datasets into one.

Convolution refers to the act of normalizing input data in CNNs.

Convolution is a technique used to increase the size of the input data in CNNs.

Convolution is the process of applying a filter to input data to extract features in CNNs.

Strided convolution with stride s=2 reduces the output dimensions by half, moving the filter two pixels at a time.

Strided convolution with stride s=2 increases the output dimensions by doubling the filter size.

Strided convolution with stride s=2 moves the filter one pixel at a time, maintaining the original dimensions.

Strided convolution with stride s=2 applies multiple filters simultaneously without changing the output size.

11 filters

3 filters

96 filters

55 filters

The ReLU activation function outputs a constant value of one for all inputs.

The ReLU activation function outputs the square of the input if positive, otherwise outputs negative input.

The ReLU activation function outputs the input if positive, otherwise outputs zero.

The ReLU activation function outputs the input as is, regardless of its value.

270 parameters

280 parameters

• 300 parameters

100 parameters

To increase the number of layers without changing the architecture.

To bypass complex computations and reduce training time.

To help gradients flow through the network and avoid the vanishing gradient problem.

• To eliminate the need for pooling layers.