Many-to-one encoder training

Implementation of the many-to-one encoder training procedure using backpropagation.

THIS DISTRIBUTION CAN SOLELY BE USED FOR ACADEMIC OR PERSONAL RESEARCH.
Source code written and owned by Tiantian Xu at [e-mail].
This package contains the implementation of many-to-one encoder training
procedure used in the following papers:
1. Xu et.al, "Dual many-to-one-encoder-based transfer learning for
cross-dataset human action recognition", 2016, Image and Vision Computing
and
2. Xu and Wong, "Learning temporal structures for human activity
recognition", to appear in 2017 British Machine Vision Conference.
Please cite either or both the above papers if you use this package.
Thanks.

Documentation of this package
trainMOEncoder.m: train a Multiple-to-one encoder network with user
defined hiddenlayer size and trainning parameters.
ffnet.m: the autoencoder network.
activation_func.m and sigmoid.m: activation functions.

Before use:
Add this folder to your search path

Example:
net=trainEncoder('matlab.mat',10,5,5,20,0.001,0.9,0.0001,'tanh');

Please note that in 'trainMOEncoder.m', before training begins, input data
is standardized to have 0 mean and 1 variance.
Please standardize test data with mean and variance of training data
before use.

Any question, please contact Tiantian Xu at [e-mail]
Tiantian Xu, 7/11/2017.