You can sample nodes for scalable #GNN #training. But how do you do #scalable #inference?

In our latest paper (Oral #LogConference
) we introduce influence-based mini-batching (#IBMB) for both fast inference and training, achieving up to 130x and 17x speedups, respectively!

1/8 in 🧵

#Introduction

Hi, I’m Aleks :wave:

I’m faculty at the CISPA Helmholtz Center for Information Security where I lead the group on trustworthy #AI.

I’m broadly interested in #Trustworthy #MachineLearning, that is models and algorithms that are not only accurate or efficient but also robust, privacy-preserving, fair, uncertainty-aware, and interpretable. I have a special focus on graph-based models such as graph neural networks.

I'm happy to be here. Looking forward to great discussions 🎉

#introduction

Hi all!

I'm a research scientist joining #Google Research in a few weeks. I've written my PhD thesis at TU Munich, and interned at #DeepMind and #FAIR.

I've done research on #MachineLearning for graphs and molecules, but want to focus more on safety aspects in the future.

I am currently visiting #Berkeley, and will soon be based in #Zurich.