Which deep learning algorithms does MATLAB support?

I want to use deep learning with MATLAB. Since every release introduces new algorithms, I am confused which version supports which algorithms, and which toolboxes I need to have. Could you tell me about the supported version and required/recommended toolboxes of each deep learning algorithm?

Answer 1

[MathWorks Support Team]

As of R2019a, MATLAB supports the following deep neural network algorithms:

・Recurrent Neural Networks (RNN)

・Stacked Autoencoders (SAE)

・Convolutional Neural Networks (CNN)

・Regions with CNN (R-CNN)

・Fast R-CNN / Faster R-CNN

・Directed Acyclic Graph Networks (DAG)

・Long Short-Term Memory Networks (LSTM)

・Semantic Segmentation

Not only these algorithms, there are some related features such as transfer learning (using the pre-trained models based on MatConvNet/Caffe/Keras/Tensorflow) and deep dream (visualization of network features). Supported versions and required/recommended toolboxes are depending on which feature you want to use. Please check with the following table:

Table: Supported versions and required/recommended toolboxes of deep learning related algorithms

※1 Cannot import models with layers which are not supported by Neural Network Toolbox

※2 "Required" when GPU is used

※3 "Required" when the classification method is not Neural Network

Model importers and pre-trained models are provided via File Exchange by MathWorks Neural Network Toolbox Team.

https://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A1211826

The name of each toolbox is abbreviated in the table. The official name is as follows:

・PCT: Parallel Computing Toolbox

・NN: Neural Network Toolbox

・ST: Statistics and Machine Learning Toolbox

・IPT: Image Processing Toolbox

・CVST: Computer Vision System Toolbox