Mask
A mask is a representation or mechanism that selectively identifies, includes, excludes, reveals, or obscures specified parts of data, a system, or a computation.
GPU Programming with Triton
A mask is a per-element boolean predicate that enables or suppresses selected elements of a GPU memory operation.
- How it is used
- Protects Triton loads and stores when fixed-size blocks or tiles extend beyond valid data.
- How deep it goes
- Explains predicate construction, masked load and store behavior, multidimensional broadcasting, and replacement values for subsequent computations.
- Examples
- Works through a partially covered matrix tile and a fused softmax with negative-infinity replacement.
- What you can do after
- Apply it via Triton to guard boundary accesses and select replacement values appropriate to the computation.
Deep Learning with PyTorch, Second Edition
- How it is used
- Selects editable image regions, segmentation pixels, or sample subsets for classification metrics.
- How deep it goes
- Explains mask meanings and metric operations with pipeline examples, while using pretrained models rather than implementing inpainting or Segment Anything from scratch.
- Examples
- Uses a horse-to-zebra edit, astronaut segmentation, and CT nodule masks to illustrate pixel selection.
- What you can do after
- Apply masks via diffusers and Segment Anything, and use Boolean masks to calculate classification metrics.
Read how Deep Learning with PyTorch, Second Edition explains mask →
Inside Deep Learning: Math, Algorithms, Models
The use of masks to ignore invalid or padded elements when processing batched variable-length data.
- How it is used
- Keeps padded items from affecting attention and context averages in batches of variable-length inputs.
- How deep it goes
- Works through Boolean mask construction, attention-score suppression, and valid-item averaging, including an empty-bag division safeguard.
- Examples
- Uses zero-filled items in a five-bag tensor, a masked attention network, and merged nuclei pixel labels.
- What you can do after
- Build it from scratch with PyTorch to identify padded items and exclude them from attention and context averages.
Read how Inside Deep Learning: Math, Algorithms, Models explains mask →
Graph Neural Networks in Action
A mask is a Boolean array or tensor that selects a subset of data elements.
- How it is used
- Selects labeled nodes for training while keeping other node subsets distinct in semi-supervised graph learning.
- How deep it goes
- Defines Boolean masks and works through their construction, embedding filtering, and use in a GNN training loop.
- Examples
- Uses complementary NumPy masks for a 20% labeled subset and a PyTorch `train_mask` to select predictions.
- What you can do after
- Build it from scratch with NumPy masks, then apply a PyTorch training mask to compute loss on labeled nodes.
Build a Large Language Model (From Scratch)
A training-time method that prevents the model from attending to certain input positions, such as padding tokens.
- How it is used
- Uses masking during batched instruction fine-tuning to exclude padding and optionally prompt tokens from attention or loss.
- How deep it goes
- Implements padding, shifted targets, -100 replacement, and loss behavior, but leaves instruction masking optional.
- Examples
- Demonstrates masked target tensors, GPT end-of-text padding, and an exercise comparing instruction masking with unmasked training.
- What you can do after
- Apply masking via a custom PyTorch collate function and verify that ignored targets leave cross-entropy unchanged.
Read how Build a Large Language Model (From Scratch) explains mask →
Build a Text-to-Image Generator (from Scratch)
A mechanism that selectively blocks information from specified sequence positions during model computation.
- How it is used
- Constrains decoder self-attention so sequence predictions cannot use future positions.
- How deep it goes
- Explains purpose and architectural placement, while deferring detailed masking mechanics.
- Examples
- Uses German-to-English translation and one-token-at-a-time text generation as concrete applications.
- What you can do after
- Understand how masking preserves valid sequential dependencies during decoder prediction.
Read how Build a Text-to-Image Generator (from Scratch) explains mask →
CSS in Depth, Second Edition
Masking is a CSS visual technique that selectively reveals or hides portions of an element using a mask image.
- How it is used
- Creates irregular silhouettes, fades, and selective visibility when rectangular rendering or border-radius is insufficient.
- How deep it goes
- Explains the opacity principle, demonstrates image and gradient masks, and catalogs related properties without descending into rendering internals.
- Examples
- Uses a star-shaped image mask, a fading linear gradient, a radial vignette, and repeating-gradient stripes.
- What you can do after
- Build masked images, tune their placement and repetition, create gradient fades, and diagnose invisible or misaligned masks.
Deep Learning with R, Third Edition
Masking is the exclusion of designated sequence positions from downstream neural-network computations.
- How it is used
- Excludes padding tokens from recurrent computations so sequence predictions use meaningful tokens.
- How deep it goes
- Explains implicit embedding-layer masking and its effect on LSTM output, while deferring explicit mask construction.
- Examples
- Uses padded movie-review sequences and an R/Keras Embedding–bidirectional LSTM classifier.
- What you can do after
- Apply it via Keras by reserving zero for padding and enabling mask_zero on an Embedding layer.
Read how Deep Learning with R, Third Edition explains mask →
Deep Learning with Python, Third Edition
A technique for marking padded or irrelevant sequence positions so later layers ignore them.
- How it is used
- Uses masks in padded text sequences so recurrent classifiers ignore padding when producing final predictions.
- How deep it goes
- Explains implicit zero-based masking and optional explicit masks, without detailing custom mask construction.
- Examples
- Demonstrates a padded movie review ending in “awful” and a Keras Embedding-to-Bidirectional-LSTM model.
- What you can do after
- Operate/configure Keras sequence models that ignore zero-valued padding or accept explicit masks.
Read how Deep Learning with Python, Third Edition explains mask →
Also covered in
- Human-in-the-Loop Machine Learning: Active learning and annotation for human-centered AI
- Rearchitecting LLMsIn early access (MEAP)
- Real-World Cryptography
- Secure APIs
- Build a DeepSeek Model (From Scratch)In early access (MEAP)
- Grokking AI Algorithms, Second Edition
- DuckDB in Action
- Infrastructure as Code, Patterns and Practices: With examples in Python and Terraform
- Architecting an Apache Iceberg Lakehouse
- Generative AI in Action
- Knowledge Graphs and LLMs in Action
- Build AI-Enhanced Web Apps
What to know first
- language modelA language model is a computational model that represents patterns in linguistic sequences to assign probabilities, predict, generate, or evaluate language.
- accuracyAccuracy is the degree to which a result agrees with a specified truth, reference, or criterion, commonly measured as the proportion of evaluated predictions that are correct.
- softmaxSoftmax is a function that exponentiates each value in a score vector and normalizes the results to produce positive values summing to one, forming a probability distribution.
- backpropagationBackpropagation is reverse-mode differentiation that propagates derivatives backward through a computational graph to compute gradients of an output or loss with respect to inputs and parameters.
Related concepts
- cropalternative to · Cropping is an image-processing operation that selects a subregion of an image and excludes the portions outside it.
- LSTMuses · A long short-term memory (LSTM) network is a recurrent neural-network architecture with gated memory cells that control information flow and model dependencies in sequential data.
- boolean seriesused by · A Boolean series is an ordered, indexed sequence of true and false values corresponding to observations.
- fancy indexingcontrasts with · Fancy indexing is selecting multiple array elements or positions with a collection of integer, Boolean, or label-based selectors.
Questions about mask
What is mask?
A mask is a representation or mechanism that selectively identifies, includes, excludes, reveals, or obscures specified parts of data, a system, or a computation.
Which Manning books explain mask?
GPU Programming with Triton, Deep Learning with PyTorch, Second Edition, Inside Deep Learning: Math, Algorithms, Models, Graph Neural Networks in Action, Build a Large Language Model (From Scratch), Build a Text-to-Image Generator (from Scratch), CSS in Depth, Second Edition, Deep Learning with R, Third Edition and 1 more.
What should I know before learning mask?
Manning books that teach mask usually introduce language model, accuracy, softmax and backpropagation first.
What is the difference between mask and crop?
A mask is a representation or mechanism that selectively identifies, includes, excludes, reveals, or obscures specified parts of data, a system, or a computation. Cropping is an image-processing operation that selects a subregion of an image and excludes the portions outside it.
What kinds of mask are there?
Kinds of mask include response mask.
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.








