Knowledge Maxxing← Back to topic library
ML MaxxingAdvanced75 sec
Convolution
A convolution slides a small learned filter across an input, such as an image, computing a weighted sum at each position to detect a specific local pattern.
Prerequisite: Artificial neuron
What matters
Because the same filter is reused at every position, convolutions can detect a pattern like an edge no matter where it appears in the image, using far fewer parameters than a fully connected layer.
Example
A convolutional filter trained to detect vertical edges produces a strong response wherever a vertical edge appears anywhere in an image.
Common misconception
A convolution filter is not manually designed by a person to detect a specific shape; the filter's values are learned automatically during training.
Recall question
Why can one convolutional filter detect a pattern anywhere in an image?
Keywords and related concepts
convolutionfilterlocal patternconvolutional neural network