hollow horizon
Our scientific objective is to improve the performance of link prediction on Text Attributed Graph in a scalable and computationally efficient manner and to generalize it to more
representative large scale real world datasets. In this pilot project, we aim to benchmark current GCN approaches and LLM methods to seek useful insights for future work.
As a preliminary step,
I have generalized current implementation of GCNs, https://github.com/ChenS676/TAPE, latest branch is wb_gcn, for parameter tuning.
Experiments are currently being run using CPUs, GPUs (4 A100) from Horeka.
Running some experiments with a guest account from my GPUs/TPUs is possible.
Help with proof reading is welcome.
Collaboration on the paper writing is welcome.
A paper will be written regardless of outcome, and all contributors will be credited, with permission.
This post is adapted from "Systematic Investigation of Mu Transfer"
full zodiac
storm pendant