Downsampling

As explained in Grokking AI Algorithms, Second Edition

Downsampling is the process of reducing the spatial resolution of a representation while retaining information about larger-scale structure.

Downsampling reduces spatial resolution so a network can move beyond individual pixels and capture larger-scale meaning. In a U-Net, the encoder progressively shrinks a noisy image while acting as a visual summarizer: “What am I looking at?” section 12.17

A downsampling layer combines several operations:

• A convolution scans feature maps with learned filters, combining local regions into new features. section 12.17
• Padding preserves border coverage, while a stride of 2 makes the output spatially smaller. section 12.17
• Additional channels let the smaller representation carry richer detected patterns; normalization and activation prepare values for later layers. section 12.17

From an 8×8 image to a smaller representation

An 8×8 noisy image first passes through eight 3×3 filters, producing an 8×8×8 set of feature maps. A later convolution uses 4×4 filters, padding of 1, and stride 2, shrinking the spatial dimensions while learning increasingly abstract features. section 12.17

Figure 12.34 The downsampling convolutional layer in our U-Net

Visualizes where the downsampling convolution sits within the U-Net. section 12.17

How Grokking AI Algorithms, Second Edition covers it

How deep it goes

Explains convolution, padding, stride, feature maps, normalization, and activation through a small worked implementation.

Examples

Uses an 8×8 noisy image, eight 3×3 filters, and a 4×4 stride-two convolution as concrete examples.

What you can do after

Build it from scratch with padding, convolution loops, filters, biases, and output-dimension calculations.

How it connects

Connects downsampling with U-Net, encoders, convolution, feature maps, group normalization, and activation functions.

liveBookmanning.com

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

Also explained in

• 

  liveBookmanning.com
  Machine Learning System DesignThe process of reducing the number of data examples or observations retained for analysis or training.
• 

  liveBookmanning.com
  Deep Learning with R, Third EditionDownsampling is the reduction of a representation's spatial resolution by selecting, aggregating, or skipping input locations.
• 

  liveBookmanning.com
  Build a Text-to-Image Generator (from Scratch)Downsampling is the reduction of the spatial resolution of a representation.
• 

  liveBookmanning.com
  The AI Pocket BookA spatial reduction step that shrinks an image or feature map to a smaller resolution, usually discarding fine-grained location detail.

Compare how 5 books explain downsampling →

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 poolingincludes · Max pooling is an aggregation operation that replaces each local or grouped region with its maximum value, often to downsample a representation.
• poolingequivalent to · Pooling is an operation that aggregates values or representations across regions or axes to produce a reduced or summarized representation.
• overfittingaddresses · Overfitting is a modeling condition in which a model fits patterns or idiosyncrasies in its training data too closely and consequently generalizes poorly to new data.
• negative sampleuses · A labeled or designated example representing a non-target, absent, incorrect, or otherwise undesired outcome in a task.
• feature mapuses · A feature map is a spatially organized array of neural-network activations representing the presence or strength of learned features at different locations.

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.