Comment by espeed | Hacker News Reader

Comment by espeed | Hacker News Reader

A sheaf is a combinatorial, high-dimensional data structure on a topological space. You can use it to stitch together images, spatio-temporal target tracking, inference, data fusion, IoT time-series, continuous functions -- all kinds of things. Crazy cool stuff!

Applied sheaf theory is pretty new, and not yet widely known. This DARPA sheaf tutorial is a full two-day video series by Prof Michael Robinson (great teacher), and it's about the equivalent of a semester CS course compressed into two days. If you want to see an intro/big-picture overview video on what the sheaf data structure is, and how it's related to graph databases, topology, category theory, and data analysis, see:

"Sheaves for engineering problems" https://www.youtube.com/watch?v=223-0x2KNOg

And for a deep dive into a paper, see:

"Sheaves are the canonical data structure for sensor integration" https://arxiv.org/abs/1603.01446

Python Cellular Sheaf Library:

https://github.com/kb1dds/pysheaf

What kind of math background do you need to understand Sheaf theory? Are these lectures truly accessible for someone with a non-background?

I tried watching bits of the youtube video OP linked but never saw where he actually defined what a Sheaf was as oppose to building motivation of learning it for applications.

I am excited for any truly accessibly beginner resources. Want to get your opinion if you links are for the true beginner?

Have you found a book or a review/survey paper on this topic?

I just started Computational Homology by Kaczynski et al. but I just looked in the index and didn't find anything on sheaves.

Prof Robinson's book is Topological Signal Processing [1], and there are a ton of papers linked from his home page [2].

For a general overview, check out the wikipedia page: https://en.wikipedia.org/wiki/Sheaf_(mathematics)

[1] https://www.amazon.com/Topological-Signal-Processing-Mathema...

[2] http://www.drmichaelrobinson.net/publications.html

Robert Ghrist offers his Elementary Applied Topology for free on his site. It's useful more as a supplement or survey than a primary text to learn.

https://www.math.upenn.edu/~ghrist/notes.html

http://www.cis.upenn.edu/~jean/gbooks/sheaf-coho.html

Sheaf models are one of the most useful tools in logic and I'll definitely read this paper once I have more time again. One thing that seems strange to me based on the overview video is the focus on sheafs on a topological space. In mathematics it's typically easier and more flexible to consider sheafs on categories equipped with a Grothendieck coverage. Is there a reason why topological spaces are sufficient to model sensor integration?

Maybe because topological spaces are more familiar to many mathematicians than categories?

That youtube link is great, thanks.

Reply on news.ycombinator.com