Hyperight

Self-supervised Learning on Ad Graphs – Levan Tsinadze, Eyeo Gmbh

Session Outline

Self-supervised learning is a new paradigm in machine learning. This paradigm uses variations of the input data for generating useful representations. Representations essential for downstream evaluation tasks (classification, regression, and any other type of pattern recognition in general). However, much of the early work in self-supervised learning has been limited to language modeling and computer vision. It has not been extended to other data modalities. In this session at the Data Innovation Summit 2023, we have Levan Tsinadze from Eyeo GmbH! In this talk, Levan explores the extension of self-supervised learning to graphs and the various algorithms which have been developed in this domain over the past few years. We also focus on data augmentation strategies for graph data. Finally, we dive into the use-case of applying these self-supervised learning techniques to ad graphs.

Key Takeaways:

  • Self-supervised learning for creating representations (embeddings) for graph Convolutional neural networks and graph neural networks in general
  • Data augmentation strategies for graph-based data
  • The optimal self-supervised learning strategies for the use case of online ad graphs and relative tradeoffs of node versus graph classification for the same

Add comment

Upcoming Events

Advertisement