Visualizing Properties of Encoders

Following up on this very old post, I finally got around to making an interactive tool that allows you to make these visualizations and configure arbitrary encoder setups. You can try it out here.

I haven’t yet put very much explanation on how things work, but if you have even a little background on the subject of encoders, you can play around with it and see how changing different parameters changes the visualizations.

Things you can do:

  1. configure scalar encoders that encode an interval or are periodic like grid cells.
  2. interactively change the parameters and have the plots immediately reflect this change.
  3. display the encoded bits, color-code them by sub-encoder, sort them by activation (I got this idea from Matt Taylor’s early visualization experiments), and show the 1/0 value or the index value.
  4. Display 7 different kinds of plots: encoder bins, point similarity, bit codes, self-similarity heatmap, point similarity heatmap, regional granulation, bit flip crossings.
  5. interactively move the a slider cursor to see how the input value changes the encoding
  6. turn plots on and off, and resort their order

There’s a bit more, but that should be enough to get you started.

This is a continuing effort from my previous post on my Interactive Visualization of Discrete Cortical Circuits (DCC).

The latest demo of which you can find on my research blog.

3 Likes