Kernel size

As explained in Inside Deep Learning: Math, Algorithms, Models

The spatial extent of a convolutional filter.

Kernel size is the spatial extent of a convolutional or pooling window. In PyTorch, it is the numeric argument that determines how many neighboring pixels each operation considers. section 3.4.2 section 3.5

Use kernel size to control the local operation and its interaction with stride and padding:

• Convolutions commonly use stride 1, evaluating every possible location; padding can preserve height and width. section 3.4.2 section 3.5
• Max pooling uses the kernel as its selection window and commonly sets stride equal to kernel size, shrinking each dimension by that factor. section 3.5
• GAN architectures may choose kernel size 5 for ordinary convolutions and 4 for transposed convolutions to produce smoother, evenly contributed expansions. section 9.4.1

PyTorch choices

A same-sized convolution can use nn.Conv2d(..., 3, padding=3//2). A GAN helper instead passes k_size=5 to nn.Conv2d and uses k_size_t=4 with nn.ConvTranspose2d(..., stride=2, padding=1). section 3.4.2 section 9.4.1

Figure 3.14 Example of max pooling for K = 2 (top) and K = 3 (bottom). Each region of the input (left) has a color indicating the group of pixels participating in the pooling, and the output (right) shows which value was selected from the input region. Note that the output is smaller than the input by a factor of K.

Compare K=2 and K=3 max-pooling windows and see how kernel size reduces the output. section 3.5

How Inside Deep Learning: Math, Algorithms, Models covers it

How deep it goes

Defines window size and relates it to stride and padding, but does not derive convolution output-size formulas.

Examples

Uses K=3 convolutions, K=2 and K=3 pooling windows, and GAN settings k_size=5 and k_size_t=4.

What you can do after

Apply it via PyTorch by selecting kernel size alongside padding and stride in convolutional and pooling layers.

How it connects

Connects convolution, pooling, stride, padding, and transposed convolution through shared spatial-window choices.

liveBookmanning.com

Read Inside Deep Learning: Math, Algorithms, Models on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Evolutionary Deep Learning: Genetic algorithms and neural networksKernel size is the spatial extent of a convolutional filter applied by a convolutional layer.
• 

  liveBookmanning.com
  Learn Generative AI with PyTorchThe spatial size of a convolutional kernel or receptive window applied to an input.
• 

  liveBookmanning.com
  Math and Architectures of Deep LearningThe spatial extent of the sliding window used by a pooling layer, determining the patch size summarized at each step.

Compare how 4 books explain kernel size →

Related concepts

• convolutional layeruses · A convolutional layer is a neural-network layer that applies shared learned filters to local regions of structured input to produce feature representations.
• max poolingused by · Max pooling is an aggregation operation that replaces each local or grouped region with its maximum value, often to downsample a representation.
• poolingused by · Pooling is an operation that aggregates values or representations across regions or axes to produce a reduced or summarized representation.
• hyperparameterkind of · A hyperparameter is a setting chosen outside ordinary parameter learning that controls a model, algorithm, or learning procedure.
• paddingcontrasts with · Padding is the addition of extra space, values, or data around or within a structure to satisfy size, alignment, layout, or processing requirements.
• expressive powercontrasts with · Expressive power is the capacity of a model, language, or system to represent or convey a range and complexity of functions, patterns, distinctions, or ideas.

This explainer was generated by Manning's knowledge graph from the text of Inside Deep Learning: Math, Algorithms, Models. The section links point to the parts of the book it draws on.