Aliasing
Aliasing is a sampling artifact in which insufficient resolution causes distinct or fine-grained signal features to become indistinguishable or visibly distorted.
Inside Deep Learning: Math, Algorithms, Models
A sampling artifact in which fine-grained detail is collapsed into a smaller representation, causing distinct inputs to become indistinguishable.
- How it is used
- Uses aliasing to explain why naive CNN downsampling loses detail and weakens translation invariance.
- How deep it goes
- Explains sampling failure, toy patterns, image effects, shift tests, and anti-aliased pooling implementation.
- Examples
- Uses zebra stripes, one-dimensional binary patterns, CIFAR-10 crops, pixel shifts, and PyTorch pooling modules.
- What you can do after
- Diagnose/debug with it by testing predictions across pixel shifts and replace naive downsampling with anti-aliased pooling.
Read how Inside Deep Learning: Math, Algorithms, Models explains aliasing →
How Computers Make Books
Jagged or otherwise distorted visual appearance caused by representing a high-resolution signal at insufficient resolution.
- How it is used
- Explains why lines and filled shapes look jagged when displayed at insufficient resolution.
- How deep it goes
- Defines aliasing through lines and shapes, then connects it to resolution, viewing distance, anti-aliasing, and LCD sub-pixels.
- Examples
- Uses diagonal lines, an ampersand, gray dots, microscope images, and screen, smartphone, and billboard viewing distances.
- What you can do after
- Apply anti-aliasing by using gray shades to make low-resolution lines and shape boundaries appear smoother.
Related concepts
- anti-aliasingaddressed by · Anti-aliasing is a smoothing technique that reduces visual artifacts caused by sampling or representing continuous shapes on discrete grids.
- poolingaddressed by · Pooling is an operation that aggregates values or representations across regions or axes to produce a reduced or summarized representation.
- local averagingaddressed by · Local averaging is a smoothing technique that replaces values with arithmetic means computed from nearby observations within successive or sliding windows.
- translation invariancecontrasts with · Translation invariance is the property that a function, representation, or prediction remains unchanged or equivalent when its input is shifted in position.
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.