CS231N Lec. 5 | Convolutional Neural Networks

18 Oct 2020 in Studies on Lecture-Review

please find lecture reference from here¹

CS231N Lec. 5 | Convolutional Neural Networks

Output size shall be number of filters

Summary of Conv. Layer

Common setting :

Number of filters(K) : power of 2 ( e.g. 32 , 64, 128, 512 … )
F = 3, S = 1, P = 1
F = 5, S = 1, P = 2
F = 5, S = 2, P = ? (Whatever fits)
F = 1, S = 1, P = 0

Q. What’s the intuition behind how you choose your stride.

it’s kind of resolution control. ( more stride, more downsizing )
similar with pooling in sense, but different.
impacts on # of parameters, size of model, overfittings... considering trade off.

one notable point is, this Conv net and Neural net only locally connected

Example of 5 filters

28 x 28 x 5 filter
–> 5 different filters looking at same input region
filters2

CNN overview

cnn-overview

Pooling Layer

make representations smaller, manageable
pooling

Max pooling

One way of pooling. Commonly used than Average pooling.
Cuz average pooling, model couldn’t show performance at image overall. (Because it’s average..)
To check corner case of image, Max pooling is better approach.
Max-pooling