Introduction to Neural Networks
Introduction to Neural Networks
Convolutional neural networks can learn features through supervised learning and can be categorized into three different types:
- Classification
- Object detection
- Segmentation
The underlying basis of all these neural networks is the convolutional layer.
A classification CNN takes in an input image and outputs a distribution of class scores. This is done by feeding the input image through convolutional layers, which is the backbone of a CNN. These layers are used to filter the input image, and the filters are also known as convolution kernels. A kernel has a certain (small) size and one or more channels, specifying its depth. Each filter extracts different features from an image, such as edges or colors. The output of a given convolutional layer is a set of feature maps (also called activation maps), which are filtered versions of an original input image. As a CNN trains it updates the weights that define the image filters in this convolutional layer using back propagation.
Activation Function
Activation functions, such as ReLU's, are typically placed after a convolutional layer to slightly transform the output so that it's more efficient to perform backpropagation and effectively train the network.
References
- Self-Driving Cars with Duckietown
- https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/
- Gradient Descent
- Yes you should understand backprop by Andrej Karpathy
- Hacker's guide to Neural Networks
- CS231n taught by Andrej Karpathy
Comments
Comments are GitHub Discussions, shown with giscus. Loading them connects to giscus.app (hosted by Vercel, USA) and GitHub (USA); you need a GitHub account to comment.