Data augmentation
Data augmentation is the process of adding or deriving examples, variables, or associated information from existing data to increase its variety or support subsequent analysis or computation.
Deep Learning with R, Third Edition
Data augmentation is a technique that generates additional training examples by applying label-preserving transformations to existing data.
- How it is used
- Mitigates overfitting in small-data image classification by varying training images before fitting a convnet.
- How deep it goes
- Details random image transformations, Dataset mapping, placement trade-offs, and testing, without isolating augmentation’s effect from added dropout.
- Examples
- Uses transformed cat pictures, a cat–dog convnet, R/Keras pipeline code, and before-and-after test results.
- What you can do after
- Apply it via Keras augmentation layers and dataset_map() to vary training batches while preserving their labels.
Read how Deep Learning with R, Third Edition explains data augmentation →
Deep Learning with PyTorch, Second Edition
A training technique that creates modified versions of examples to increase effective dataset diversity while preserving task-relevant meaning.
- How it is used
- Expands effective training diversity to discourage memorization and improve generalization, especially with limited data.
- How deep it goes
- Works through five CT augmentation types, affine resampling, caching order, and comparative training runs; leaves transformation strength open to experimentation.
- Examples
- Uses mirrored face images, translated airplanes, transformed CT nodule candidates, and TensorBoard comparisons of augmentation runs.
- What you can do after
- Build it from scratch for CT candidates with PyTorch transforms, then evaluate individual and combined changes against validation metrics.
Read how Deep Learning with PyTorch, Second Edition explains data augmentation →
Deep Learning with Python, Third Edition
A regularization approach that creates additional training examples by applying label-preserving random transformations to existing samples.
- How it is used
- Mitigates overfitting when training image classifiers on a small dataset.
- How deep it goes
- Works through random transformations, pipeline placement, and training code; cautions that transformed samples remain correlated.
- Examples
- Shows flipped, rotated, and zoomed image variations, then trains a ConvNet on augmented batches.
- What you can do after
- Apply it via Keras augmentation layers and a mapped training Dataset while checking the resulting images.
Read how Deep Learning with Python, Third Edition explains data augmentation →
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.
- How it is used
- Uses transformations and alterations to expand narrow training sets and support more robust models.
- 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 you can do after
- Build additional training data that can help models become more robust when original examples are narrow.
Machine Learning for Tabular Data: XGBoost, Deep Learning, and AI
A technique that expands or diversifies training data by applying label-preserving transformations or generating additional examples.
- How it is used
- Mitigates deep learning’s difficulty with small training datasets.
- How deep it goes
- Defines augmentation’s purpose but stops short of describing transformations, implementation steps, or evaluation procedures.
- Examples
- Uses the Tokyo and New York City Airbnb datasets to illustrate contrasting data volumes and deep-learning challenges.
- What you can do after
- Evaluate when augmentation may help a deep-learning model facing limited training data.
Deep Learning with JAX
The practice of creating modified training examples to increase diversity and improve model robustness.
- How it is used
- Uses randomized image transformations to increase training diversity, improve robustness, and reduce overfitting.
- How deep it goes
- Explains augmentation through two image transformations and connects it to implicit regularization and random-number handling.
- Examples
- Demonstrates image classification with random noise and random rotations applied to training images.
- What you can do after
- Apply data augmentation with JAX's functional random-number model to diversify image-classification training data.
Read how Deep Learning with JAX explains data augmentation →
Inside Deep Learning: Math, Algorithms, Models
The practice of creating additional training examples by applying label-preserving transformations to existing data.
- How it is used
- Improves CNN training by increasing the diversity of spatial image inputs.
- How deep it goes
- Explains randomized transforms, probabilities, label preservation, and PyTorch pipelines, but stops short of transformation theory.
- Examples
- Uses MNIST affine transforms, sample transformation visualizations, and accuracy curves comparing augmented and unaugmented CNNs.
- What you can do after
- Apply it via torchvision.transforms by composing training augmentations and evaluating the resulting CNN.
Read how Inside Deep Learning: Math, Algorithms, Models explains data augmentation →
Generative AI in Action
Data augmentation is the creation or transformation of training examples to increase dataset variety while preserving task-relevant information.
- How it is used
- Uses synthetic visual data to cover edge cases and insufficient diversity in training datasets.
- How deep it goes
- Explains GAN-based synthetic-data augmentation and names mode collapse, instability, output control, and bias risks without implementation steps.
- Examples
- Uses education training simulations, rare edge cases, and datasets lacking sufficient diversity as augmentation scenarios.
- What you can do after
- Apply it via GANs to create synthetic training data for edge cases or scarce, insufficiently diverse datasets.
Read how Generative AI in Action explains data augmentation →
Sutskever's List
Data augmentation is the creation of varied training examples by applying label-preserving transformations to existing data.
- How it is used
- Expands the effective training set and supports generalization by exposing models to realistic variation.
- How deep it goes
- Distinguishes geometric transforms from PCA-based color jitter and their purposes, without specifying implementation parameters.
- Examples
- Uses random crops, horizontal reflections, and PCA-based color jitter on ImageNet training images.
- What you can do after
- Understand: Explain how geometric and color variations can support generalization in image classification.
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.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.