tuc-proaut/vsa_toolbox

VSA_toolbox

The VSA_toolbox is a MATLAB implementation to develop and apply different kinds of VSA within one environment. It is also the source code to the paper [1].

[1] K. Schlegel, P. Neubert, and P. Protzel, “A comparison of vector symbolic architectures,” Artif. Intell. Rev., Dec. 2021.

Requirements

• MATLAB shadedErrorBar function to plot the results (https://de.mathworks.com/matlabcentral/fileexchange/26311-raacampbell-shadederrorbar) Clone the folder to your workspace:

  git clone https://github.com/raacampbell/shadedErrorBar
  

• [optional] MATLAB export_fig function to export the plots as pdf - if not installed, the script saves the plots as png (https://de.mathworks.com/matlabcentral/fileexchange/23629-export_fig)

Installation

• no specific installation is required
• start the scripts from the "VSA_toolbox" folder

Usage

Use the toolbox

• the script "demo.m" contains a demonstration of using the toolbox
• first, create an object with "vsa_env.m" and specify the architecture
• available architectures are: MAP-B, MAP-C, MAP-I, BSC, HRR, VTB, FHRR, BSDC, BSDC-S, BSDC-SEG, MBAT (see paper for explanation)
• methods of such an VSA environment object are the operations, like bundling, binding, unbinding and similarity measurement
• the folder "+operation" contains the implementations of the operators for each VSA

Reproduce the experiments

• start the experiment main script: experiments_main
• it contains subscripts for bundle capacity, binding pairs capacity, repetitive binding/unbinding and language recognition (place recognition will be published as soon as possible)
• default parameters are used in the paper
• after execution, a folder named "experimental_results" is created in the main folder with a sub-folder "plots", which contains the plotted curves

Plot the figures from the paper

• the folder "experimental_results" contains the original .mat files from the experiments in the paper
• to plot the figures, start the script "visualize_all.mat" with the default parameter setup: experimental_scripts.visualization.visualize_all

License

This code is released under the GNU General Public License version 3.

vsa_toolbox's Issues

Using MAP-I to fish out spatial co-ordinates

Imagine, I put together good face HD profile and keep it in my associative memory! Currently I am building query vector from a image with spatial co-ordinates of each feature bind to its HD. If now bind my face HD profile with query vector! Is there a way to extract spatial co-ordinates of features that matches with face profile HD?

Sorry If I bothered with my questions 🙏

Undefined variable 'prob_correct_answer' in boundpairs.m

First of all, I want to say that your paper is great, and really appreciate for sharing your code so others including me can try it.
Thanks to your paper, I have definitely been able to get a better understanding of VSAs.
I would like to inform you that an error occurred when trying to run your code.

At line 93 in boundpairs.m in the experimental_scripts folder,

prob_correct_array(d_idx,:) = prob_correct_answer;

should be changed as follows:

prob_correct_array(d_idx,:) = prob_correct;

because the prob_correct_answer is an undefined variable.
I confirmed that it worked properly after correcting it.

Why we are not normalize intermediate bundle in lang recongizition

Why we are not normalize intermediate bundled VSA in lang recognition?

VSA_Toolbox/+experimental_scripts/language_recognition.m

Line 85 in 5f60b6e

buffer = VSA.bundle(buffer,ngram,0);