mbttda

Multi Branch Tensor Train Disriminant Analysis

Code for our paper with the same title published in IEEE TIP

IEEE link

arxiv link

How to use:

mydemo.m is a script that runs most of the experimental settings. It uses:

1. A data loading utility load_data.m, with a wide variety of parameter options for algorithms.
2. A cross validation function optLambda.m that finds the best $\lambda$ value for various number of 'branches'.
3. Functions for each algorithm compared in the paper. See mdademo.m for Tucker based algorithms.
4. A visualization utility, figPlot.m, that can plot accuracy and computation time w.r.t. storage complexity.

The algorithm saves the results for each set of parameters, and random initialization of experiments in a new folder. See line 80 in mydemo.m for some details regarding file name convention. The visualization utility reads from these files to plot.

Acknowledgement:

This toolbox was created with heavy utilization of TTNPE toolbox, and other third-party toolboxes provided in the folder third-party.