Eroxl's NotesGraph
Convolutional Layer

A convolutional layer is a layer that unlike dense layers operates on a small section of its inputs at a time. It does this through convolving the kernel.md) over the top of its inputs, this is where it gets its name of the convolutional layer.

Hyper Parameters

Kernel Size

The kernel size controls how large the square matrix that the kernel uses.

101010101010101010101010101010101010000000000000000000121000-1-2-1InputKernelOutput000040404040404040400000

Example of the final iteration of the convolution performed with a kernel size of

101010101010101010101010101010101010000000000000000000InputKernel122100000000-1-2-2-1Output606060606060606060

Example of the final iteration of the convolution performed with a kernel size of

Stride

The stride controls how the offset of the kernel.md) changes after each convolution iteration.

101010101010101010101010101010101010000000000000000000121000-1-2-1InputKernelOutput0101010101010101010101010101010101010000000000000000000121000-1-2-1InputKernelOutput00Step 1Step 2

Example of two iterations of the convolution performed with a step size of

101010101010101010101010101010101010000000000000000000121000-1-2-1InputKernelOutput0101010101010101010101010101010101010000000000000000000121000-1-2-1InputKernelOutput00Step 1Step 2

Example of two iterations of the convolution performed with a step size of

Definition

Forwards Pass

A single convolutional layer can be written as a convolution between two matricies