[Theory] FW: Theory Lunch 2023-04-12T17:30:00.000Z

[Christopher Kang]

________________________________
From: noreply+automations at airtableemail.com <noreply+automations at airtableemail.com>On Behalf OfTheoryBot (via Airtable) <noreply+automations at airtableemail.com>
Sent: Wednesday, April 12, 2023 12:00:53 AM (UTC-06:00) Central Time (US & Canada)
To: Antares Chen <antaresc at uchicago.edu>
Cc: Christopher Kang <ctkang at uchicago.edu>
Subject: Theory Lunch 2023-04-12T17:30:00.000Z


Today's Theory Lunch talk:

Andrew Hands (University of Chicago): Permutation Equivariance in Higher Order Message Passing Networks and Beyond

https://uchicago.zoom.us/j/91616319229?pwd=dDdXQnFXeGNubFRkZy9hTDQrcWlXdz09<https://urldefense.com/v3/__https://uchicago.zoom.us/j/91616319229?pwd=dDdXQnFXeGNubFRkZy9hTDQrcWlXdz09__;!!BpyFHLRN4TMTrA!8iS5MioaE0jVsQX68yYWBAJ1ntFBPZ51Vdz2sOtKwv--OBq6P4wh7keNawXz4wEIX45gCJfKQtzLztw4WwfChUFe3zOW1xWenpY$>

Description: Message Passing Neural Networks (MPNNs) have proven to be a powerful tool for learning functions on graphs; however, they are inherently limited as they only propagate information across graphs in a purely local and isotropic fashion. Thus preventing them from being able to distinguish even simple structures such as k-regular graphs. To help address this issue, we introduce Permutation Equivariant Higher Graph Neural Networks, which are able to consider higher order message passing between not just vertices, but subgraphs as well. This allows us to generate a richer higher order representation, while still maintaining a low time complexity for lower density graphs, such as those arising from molecular learning.

Sent via Automations on         [Airtable] <https://urldefense.com/v3/__https://airtable.com?utm_medium=email&utm_source=product_team&utm_content=transactional-alerts__;!!BpyFHLRN4TMTrA!8iS5MioaE0jVsQX68yYWBAJ1ntFBPZ51Vdz2sOtKwv--OBq6P4wh7keNawXz4wEIX45gCJfKQtzLztw4WwfChUFe3zOWzlW_jPs$>
©2023 Airtable
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.uchicago.edu/pipermail/theory/attachments/20230412/a189c382/attachment-0001.html>