Convolution kernel

As explained in Grokking AI Algorithms, Second Edition

A convolutional filter is a small learned grid of weights that detects a particular local pattern in image data.

A convolution kernel, also called a filter, is a small grid of learned weights that looks for a local pattern in an image. Think of it as a tiny magnifying glass: one filter may become sensitive to vertical edges, while another responds to textures. A common size is 3×3. The network starts with random weights and adjusts them during training rather than being told which patterns to find. section 12.14

To use a filter, place it over a patch of pixels. Multiply each weight by the pixel beneath it, then add the results. That sum becomes one value in a feature map—a grid showing where the filter responds. Slide the same filter across the image and repeat. section 12.14

One small calculation

For one row, suppose the filter weights are [2, 7, 2] and the pixels beneath them are [11, 2, 1]. Their contribution is (2×11) + (7×2) + (2×1) = 38. A full 3×3 calculation also includes the other two rows. The filter then moves to the next position and performs the calculation again. section 12.14

Figure 12.24 How a CNN filter scans an image

See how a small filter moves across an image, producing a calculation at each position. section 12.14

Different filters produce different feature maps. In a U-Net encoder example, eight distinct 3×3 filters scan an 8×8 noisy image, producing eight 8×8 feature maps. Each filter also has a learnable bias added to its calculated values. Padding around the image lets the filters calculate values at its edges while keeping these feature maps 8×8. section 12.17 section 12.17

Filter size and movement also affect resolution. The first encoder convolution uses a 3×3 filter with a one-pixel stride; a later downsampling convolution uses 4×4 filters with a two-pixel stride. In the decoder, a transposed convolution uses trainable filters differently: each input value contributes a larger patch to the output, and overlapping contributions are added. section 12.17 section 12.19

The practical takeaway: a kernel is not a fixed edge-finding rule. It is a reusable set of weights whose responses help a convolutional network extract useful image features—or, in a transposed convolution, help enlarge a feature map. section 12.14 section 12.19

How Grokking AI Algorithms, Second Edition covers it

How deep it goes

Works through sliding dot products and feature-map shapes, then sketches NumPy implementations; does not derive weight updates.

Examples

Uses a numeric filter-row calculation and eight 3×3 filters scanning an 8×8 noisy image.

What it assumes

Assumes basic familiarity with images and pixels while defining filters and explaining the dot-product calculation.

What you can do after

Build it from scratch using NumPy-style loops to scan image patches and calculate feature maps.

How it connects

Connects convolutional layers, feature maps, backpropagation, padding, downsampling, and transposed convolution.

liveBookmanning.com

Read Grokking AI Algorithms, Second Edition on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Natural Language Processing in Action, Second EditionA convolutional kernel is a parameter matrix or tensor whose sliding weighted operation detects a local pattern in an input sequence.
• 

  liveBookmanning.com
  Math and Architectures of Deep LearningA learned kernel applied in a convolution layer to detect local spatial patterns in images.
• 

  liveBookmanning.com
  Inside Deep Learning: Math, Algorithms, ModelsA learned array of weights applied over local input regions to compute convolution outputs.
• 

  liveBookmanning.com
  Deep Learning with PyTorch, Second EditionA small learned filter window applied over local regions of an input to compute weighted sums.
• 

  liveBookmanning.com
  Deep Learning with Python, Third EditionThe learned weight tensor used by a convolution to transform local input patches.
• 

  liveBookmanning.com
  Deep Learning with R, Third EditionA convolution kernel is a learned weight matrix that transforms each extracted input patch into an output feature vector.
• 

  liveBookmanning.com
  Graph Neural Networks in ActionA learnable weighted operation that transforms input data to emphasize particular patterns or features.
• 

  liveBookmanning.com
  Learn Generative AI with PyTorchA learned convolutional kernel whose weights slide across local input regions to produce feature responses.
• 

  liveBookmanning.com
  Sutskever's ListA convolutional kernel is a grid of learned weights applied across local regions of an input to produce spatial responses.
• 

  liveBookmanning.com
  Evolutionary Deep Learning: Genetic algorithms and neural networksA convolutional filter is a learned array of weights that is slid across local input regions to produce feature responses.

Compare how 11 books explain convolution kernel →

Related concepts

• convolutional neural networkused by · A convolutional neural network is a neural-network architecture that applies learned convolutional operations to extract patterns from structured data.
• convolutional layerused by · A convolutional layer is a neural-network layer that applies shared learned filters to local regions of structured input to produce feature representations.
• feature mapproduces · A feature map is a spatially organized array of neural-network activations representing the presence or strength of learned features at different locations.
• image processinguses · Image processing is the computational manipulation and analysis of digital image data to transform images or extract information.

This explainer was generated by Manning's knowledge graph from the text of Grokking AI Algorithms, Second Edition. The section links point to the parts of the book it draws on.