Receptive field
A receptive field is the set or region of input elements whose information can influence a particular neural-network unit, representation, or output.
Deep Learning with PyTorch, Second Edition
The region of the original input that can influence a particular output unit in a convolutional network.
- How it is used
- Tracks which input pixels or voxels can influence an output during image recognition and suspected-tumor classification.
- How deep it goes
- Traces input influence through stacked 2D and 3D convolutions, padding, and pooling, without giving a general calculation formula.
- Examples
- Traces an 8 × 8 image through convolution and pooling, then compares one and two 3 × 3 × 3 convolutions.
- What you can do after
- Evaluate/understand how stacking convolutions and pooling changes the input area represented by an output pixel or voxel.
Read how Deep Learning with PyTorch, Second Edition explains receptive field →
Sutskever's List
The receptive field is the region of an input that can influence a particular neural network unit's activation.
- How it is used
- Explains how deeper AlexNet units integrate information across broader image regions.
- How deep it goes
- Defines the input region affecting a unit and qualitatively explains its expansion across AlexNet’s layers.
- Examples
- Contrasts AlexNet’s 11×11 early filters with 5×5 and 3×3 filters deeper in the network.
- What you can do after
- Understand how deeper AlexNet units can capture patterns across broader image regions.
Related concepts
- depthproduced by · Neural-network depth is the number of successive computational layers in a neural-network architecture.
- dilated convolutionproduced by · A dilated convolution applies a filter whose sampling positions are spaced apart, expanding the receptive field without proportionally increasing kernel size or parameters.
- poolingused by · Pooling is an operation that aggregates values or representations across regions or axes to produce a reduced or summarized representation.
- downsamplingproduced by · Downsampling is the reduction of data quantity, spatial resolution, temporal resolution, or sampling rate by retaining, selecting, or aggregating fewer representative values or observations.
- convolutional neural networkproduces · A convolutional neural network is a neural-network architecture that applies learned convolutional operations to extract patterns from structured data.
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.