# convolutions

Tags: math, ml

Convolutions is applying a filter to an input that results in an activation, such as with convolutional neural network (cnn)

For ml, a convolution is a linear operation that involves the multiplation of a set of weights with the input

• the filter is smaller than the input data and is mulitplied by the dot product
• smaller is intentional, allows the filter to be reused across multiple places in the image
• if we design the filter to detect a specific type of feature, then this can be discovered anywhere in the image (aka translation invariance, checking if the feature is present rather than where it was present)
• pytorch implements this with conv2d