search

leios / SoME_Topics

Collaboration / Topic requests for SoME
Other
212 stars 6 forks source link

Can we recast AI problems as geometric ones? / An applied introduction to geometric group theory and combinatorics #30

Closed tfburns closed 2 years ago

tfburns commented 2 years ago

About the author

I'm a PhD student studying mathematical/computational neuroscience and AI. My collaborator on this project from the academic side is a lecturer in pure math. I tried working on something for last SoME but ran out of time trying to 'prettify' my very unpretty PPT files, so it would be great to have someone with a complementary skillset to work with.

Quick Summary

How can we represent a typical AI environment using a single geometric or topological object? And if we could, what would it tell us?

This lesson aims at introducing tools and concepts from geometric group theory and combinatorics via a demonstration of their application in simple AI environments called gridworlds. For a high-level (and slightly more academic explanation), see here: https://twitter.com/tfburns/status/1531442853699670017.

The target audience for the SoME project would be high schoolers and early math/AI undergraduates. I'm hoping they will (1) be introduced to classical and new results in geometry (and be introduced to the fields of geometric group theory and combinatorics), and (2) be inspired to see how these results can be applied in exciting and unexpected ways in emerging fields like AI.

Target medium

The target medium should probably be a moving picture, ideally a video. An interactive webpage could also work, but there would be some computational limitations on what we could show, most likely, and so probably not well-suited for this project. It would be great to make some nice animations of gridworlds (inherently simple 2D environments, much like basic video games) and some useful 3D pictures of graph-like objects called complexes.

More details

Twitter thread: https://twitter.com/tfburns/status/1531442853699670017 (could be used as a rough guide/storyboard for the video, but ideally we'd add some more technical details about cube complexes and such)

Academic paper: https://arxiv.org/abs/2201.06274 (no need for you to fully read or understand this, as I will be more than happy to, but just FYI -- and the appendix of this paper has some background information on cube complexes and Gromov's Link Condition which should probably go into the video)

Contact details

Please contact me via Twitter (tfburns), or find my e-mail on my website: tfburns.com

Additional context

There is a GitHub repository with some tools for constructing simple gridworlds and their associated geometric objects: https://github.com/tfburns/State-Complexes-of-Gridworlds/

Shiviiii23 commented 2 years ago

Hello! I would love to help with this.. I've been trying to do more research in this field! I am just an undergraduate right now but I would love to connect, learn more about this, and definitely help with a presentation! Please comment back if you would be interested in working with me! My name is Shivani and I study Mathematics and Computer Science at UT Austin

ghost commented 2 years ago

I have contacted you via Twitter!

tfburns commented 2 years ago

@Shiviiii23 Could you send me an e-mail?

@r3dapple Thanks! I messaged you back.

For anyone else reading this, I'm organising a small group video call for people who contacted me, which will be held on 19 June at 3pm Japanese Standard Time. If you'd like to join, please contact me.

In the call I'll give a high-level overview of the content, and we can then go into questions. From there, hopefully we coalesce into a group of sorts, but also if some are not interested after hearing more, that's totally fine.

Shiviiii23 commented 2 years ago

Hello! I was wondering if you could give me your email address @tfburns

tfburns commented 2 years ago

@Shiviiii23, please find my e-mail on my website or contact me via Twitter. I would prefer not to post my e-mail here.

tfburns commented 2 years ago

Although a few people responded as coming/not coming, no one showed to the video call today.

Unfortunately I have quite limited time for such meetings at the moment as have some conferences and meetings planned up until early August. Please e-mail me directly if you are still interested after reading the paper.

ghost commented 2 years ago

@tfburns Please check your email. The call was scheduled for 3pm JST, which is in 23 minutes.

tfburns commented 2 years ago

@r3dapple Just sent you an e-mail.

tfburns commented 2 years ago

Embarrassingly, I mis-scheduled the meeting time on my calendar. Fortunately, @r3dapple indeed correctly pointed this out and we just had a wonderful chat :) I'll close this now since it seems we have a team. However if others wish to join us feel free to reach out.

tfburns commented 2 years ago

As a small conclusion to this story: there was some initial interest and conversations, but this ultimately didn't lead to a video/animation/explainer article.