Multiple order modeling for deep learning

Introduction
This is an example of multiple order modeling for accuracy improvement in deep neural networks.
Different approaches are shown on how to use the outputs of a category prediction model as predictors for a second model.
Time series instances of samples are used as multiple inputs (for example N frames of a video is used as N image inputs) for model,
and those N number of predicted output (probability density) is used as predictors for the second model.

Data
We attach a set of simulated data for testing this approach.
Details regarding the data is available in comment section of FE_DataLoad.m

Supporting function
A function (trainClassifier.m) attached here for training data with SVM algorithm is called by the scripts.
This function is generated using MATLAB's CalssifierLearner App's code generation functionality