Re-creating the results in ICASSP 2023 submission 
"IMPROVED AUTO-ENCODING USING DETERMINISTIC PROJECTED BELIEF
NETWORKS AND COMPOUND ACTIVATION FUNCTIONS":

1. Get data.
   Experiment 1:
     mn389_0.mat   : These are characters "3", "8", and "9" from MNIST with added dither.

   Experiment 2:
     ds45.mat  :  These are commands "three" and "tree" from Google keywords

2. Download the Aesara version of the PBN Toolkit at

       http://class-specific.com/pbntk/

Skim through the docs to familiarize yourself with the various sections of the 
graphical interface, and terminology.

3. Download the model definitions and event list files. 
    Experiment 1: mntestz.py, mntest.py
    Experiment 2: goog11z.py, goog11.py, goog11_dpbn.py
   Put these in a sub-directory calles "models"

4. Run the toolkit (see instructions at: http://class-specific.com/pbntk/  )
   python pbntk.py

  o If you are not training from scratch,  extract the pre-trained parameters:
    Experiment 1: mntestz_mn389_0.tgz
    Experiment 2: goog11z_ds45.tgz

To continue training or evaluate the D-PBN/TCA model:

  a. set batchsize (BS) to 288 (because this 
  b. enter 'ds45' into TRAIN DATA field, and press LOAD
  c. Check the "+1" checkbox on for bottom right of GUI.
  d. enter 'goog11z' into MODEL field and press LOAD
     If you have unpacked the pre-trained parameters, 
     they will be loaded in, otherwise randomly initialized.  
     To delete pre-trained parameters, press DELETE, then LOAD.
  e. Enter "7" into LYR field in DPBN Section.
  f. Enter "7" into LYR field in PBN-Synth Section.
  g. Compile DPBN by pressing DPBN.
  h. Compile DPBN synthesis pressing SYN in the PBN Synthesis Section.

  To continue training DPBN/TCA:

  a. Set learning rate (LR) to 1e-6 
  b. Check and uncheck "TRN" in DPBN Section to start and stop training.
  c. To avoid start-up transients, re-load parameters (press "LOAD" in MODEL Section),
     and re-start training several times until a stable change is seen, then
     training can proceed.
  

  To evaluate DPBN/TCA:

  a. Select desired data partition (Train/Valid/Test) in FWD Section.
  b. Press EVAL in DPBN Section, result gets printed.

  To visualize reconstructions:

  a. Check "PLT" on top right of GUI to enable plotting.
  b. Select desired data partition (Train/Valid/Test) in FWD Section.
  c. Press GO in PBN Synth Section.


For the D-PBN (without TCA), load the model "goog11_dpbn"

For the Auto-encoder, load the model "goog11"