This session will talk about recent developments in graph embeddings, inspired from word embeddings in NLP. Different methods to train node embeddings, edge embeddings through random and biased walks. It will show how you can train models for classification and regression purposes using graph embeddings as input. It will briefly mention the state of the art in this rapidly developing field – graph convolutional networks and gated graph neural networks and the advantages of using these techniques in production with a massive graph dataset.
We will see how well word embeddings work on NLP. and the popular implementations of their algorithms
We will take a sample small graph and generate walks and train node and edge embeddings on it
We will use these node embeddings and edge embeddings to train a classifier.
The talk will help people understand the latest development in capturing graph features node and edge embeddings rather than via feature engineering to generate manually curated features. This talk will also demonstrate how we can use embeddings to perform graphical problems like node classification.
Srishti works on training deep learning models and building end to end ML pipelines to deploy them at Hike for a variety of data science problems such as recommendations, image stylisation, growth and more. Srishti has a Masters in Machine Learning from Georgia Tech and used to work at Apple before joining Hike in her current role.