The relational inductive biases paper seems a very natural evolution of the “think like a vertex/edge” frameworks for graphs, with richer abstraction via 3 kinds of update/aggr functions. Do you think choice of specific functions leads to interesting theoretical properties?
In celebration to open sourcing Graph Networks, I assembled some slides (adapted from Yujia Li) which should serve as a nice tutorial -- feedback welcome! Tutorial: https://goo.gl/J4ybLq Open Source: https://github.com/deepmind/graph_nets …pic.twitter.com/kJcKtj3pQh
-
-
-
Probably! I think we are just starting to scratch the space for these kind of models.
Kraj razgovora
Novi razgovor -
-
-
It is wonderful!
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Don't forget to enable access for the video on slide 18! ;)
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
-
-
Great slides, but you mention that Deep Sets / Pointnet have no information on edges, but this is a bit misleading. Instead I think it makes more sense to treat is as fully connected where all edges are weighted equally so all nodes see the same 'global context'.
Hvala. Twitter će to iskoristiti za poboljšanje vaše vremenske crte. PoništiPoništi
-
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.