Feedback wanted: Hobbist-built Codec for Large Sparse Binary Vectors

Science Related Papers
1

My cousins and I have been developing a new codec for very large sparse binary vectors, and the initial results seems promising for parameter free encodings. Initially we thought it might be pretty novel, because the codec is power-of-two aligned (we didn’t find anything else like it). But the results aren’t amazing right now.

The only thing is, we’re not experts in HDC/VSA, only hobbyists with some claude credit. So I was wondering if anyone here could help with some initial feedback / direction etc?

Hope that’s ok to ask.

Full paper draft at: riley-c/RILEY_paper.pdf at main · NikolasBielski/riley-c · GitHub

2

I had to use chatGPT to figure out it is a gap encoding algorithm.

3

What does large and sparse means, how large and how sparse? And what are they supposed to be used for?

4

Basically the idea is compressing large sparse binary strings for Kanerva binary splatter code like systems.
Apparently the paper shows a concentration in the number of bits needed to encode the gaps between 1’s in the binary string, which you can use to get good compression.
The paper is 70% about compression and only small amount about ML according to chatGPT’s assessment.