GAT: Graph Attention Networks (Graph ML Research Paper Walkthrough)

แชร์
ฝัง
  • เผยแพร่เมื่อ 2 ธ.ค. 2024

ความคิดเห็น • 6

  • @TechVizTheDataScienceGuy
    @TechVizTheDataScienceGuy  3 ปีที่แล้ว +2

    Hello, Binge more research paper walkthroughs at th-cam.com/video/ykClwtoLER8/w-d-xo.html
    cheers!

  • @anilkamat7872
    @anilkamat7872 2 ปีที่แล้ว

    Hi, How does GAT gets different attention scores for i->j and j->i, if the hi and hj remain the same for both operations?

  • @vinaychetnani
    @vinaychetnani 3 ปีที่แล้ว +1

    According to research you have done and insights you have gained, what do you think would perform better on real raw graph-structured data for Inductive Link Prediction - simple GraphSage or with attention ?

    • @TechVizTheDataScienceGuy
      @TechVizTheDataScienceGuy  3 ปีที่แล้ว

      In the first go, I would suggest w/ attention (as it has worked well in language domain). But again totally depends on the connectivity logic and if on average you know all nodes will have same impact on central node or not, computation/inference requirements and so on. Having said that, skip connections seem to work better (as expected most of the times) here as well. So consider trying that as well with either of the chosen methods.

  • @nastaranmarzban1419
    @nastaranmarzban1419 2 ปีที่แล้ว

    Hi hope you're doing well
    Is there any graph neural network architecture that receives multivariate dataset instead of graph-structured data as an input?
    I'll be very thankful if you answer me i really nead it
    Thanks in advanced

    • @TechVizTheDataScienceGuy
      @TechVizTheDataScienceGuy  2 ปีที่แล้ว

      Hi, for applying Gnn to any problem statement, the pre-requisite is to have data formatted and represented with nodes and edges. The graph need not always be directly available but for applying Gnn you will have to transform your data to one. Ex - textgraphs (these are also made up graphs). The nodes here can represent multivariate feature setting with edges between nodes based on certain logic. Hope that’s helpful!