Data augmentation

As explained in Delivering Machine Learning Projects: From design to deployment

Data augmentation is the creation of additional training examples by applying transformations or alterations to existing data.

Data augmentation creates extra training examples by applying transformations or alterations to existing data. It broadens narrow datasets so models are less likely to learn coincidental features and can become more robust and general. section 7.2.1

Use augmentation processes to widen the signals available during training:

• Images: rotate, mirror, scale, reposition, invert, alter contrast or brightness, add objects, and introduce noise or distortions. section 7.2.1
• Text: add spelling errors, replace non-stop words with synonyms, or use auto-translation to change phrasing. section 7.2.1
• Pipelines: preserve maintainable infrastructure so new manipulations can be added while models are tested and compared. section 5.4

Augmenting a cat image

Start with one cat image, then generate additional examples by rotating, adding objects, introducing noise, inverting, scaling, repositioning, and mirroring it. section 7.2.1

Figure 7.3 Augmenting an image of a cat with simple manipulations (Wikipedia commons, https://commons.wikimedia.org/wiki/File:Black_and_White_Cat_Sketch.svg).

Inspect the visual sequence of simple image manipulations used to create augmented examples. section 7.2.1

How Delivering Machine Learning Projects: From design to deployment covers it

How deep it goes

Explains the rationale and lists image and text transformations, while leaving implementation details of each manipulation unspecified.

Examples

Uses rotated, mirrored, noisy, scaled, repositioned, and otherwise altered cat images, plus spelling, synonym, and translation changes for text.

What it assumes

Assumes existing training data and introduces the technique without requiring advanced mathematical treatment.

What you can do after

Build additional training data that can help models become more robust when original examples are narrow.

How it connects

Connects data augmentation with training data, data pipelines, unstructured data, model robustness, and model evaluation.

liveBookmanning.com

Read Delivering Machine Learning Projects: From design to deployment on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Deep Learning with R, Third EditionData augmentation is a technique that generates additional training examples by applying label-preserving transformations to existing data.
• 

  liveBookmanning.com
  Deep Learning with PyTorch, Second EditionA training technique that creates modified versions of examples to increase effective dataset diversity while preserving task-relevant meaning.
• 

  liveBookmanning.com
  Deep Learning with Python, Third EditionA regularization approach that creates additional training examples by applying label-preserving random transformations to existing samples.
• 

  liveBookmanning.com
  Machine Learning for Tabular Data: XGBoost, Deep Learning, and AIA technique that expands or diversifies training data by applying label-preserving transformations or generating additional examples.
• 

  liveBookmanning.com
  Deep Learning with JAXThe practice of creating modified training examples to increase diversity and improve model robustness.
• 

  liveBookmanning.com
  Inside Deep Learning: Math, Algorithms, ModelsThe practice of creating additional training examples by applying label-preserving transformations to existing data.
• 

  liveBookmanning.com
  Generative AI in ActionData augmentation is the creation or transformation of training examples to increase dataset variety while preserving task-relevant information.
• 

  liveBookmanning.com
  Sutskever's ListData augmentation is the creation of varied training examples by applying label-preserving transformations to existing data.

Compare how 9 books explain data augmentation →

Related concepts

• AlexNetused by · AlexNet is a convolutional neural network architecture designed for image classification and recognized for its landmark performance in large-scale image recognition.
• 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.
• training datasetproduces · A collection of examples or observations supplied to a learning procedure to fit or adapt a model or learn its patterns or parameters.
• demonstrationuses · A demonstration is an example pairing an input, request, or prompt with a corresponding output, answer, or label to illustrate a task or expected behavior.
• lax.switchuses · A JAX control-flow primitive that selects one branch from multiple alternatives based on an integer index.
• regularizationkind of · Regularization is a modeling technique that constrains or penalizes solutions during fitting or optimization to discourage undesirable complexity or behavior and improve generalization or estimation.

This explainer was generated by Manning's knowledge graph from the text of Delivering Machine Learning Projects: From design to deployment. The section links point to the parts of the book it draws on.