Machine Learning on Graphs
Joydeep Bhattacharjee (~infinite-Joy) |
Everything is a graph. Graphs power a lot of the applications that we love and use daily. On the other hand, machine learning is about efficiently identifying patterns and relationships in data. Many tasks such as finding associations among terms so that accurate search recommendations or locating individuals within a social network who have similar interests, are naturally expressed as graphs.
The primary challenge in this domain is finding a way to represent, or encode graph structure. The primary challenge in this domain is finding a way to represent, or encode graph structure so that it can be easily exploited by machine learning models.
The presenter would create an end to end solution starting from scraping data from the internet and forming a graph using Neo4J, a popular graph database. A collective edge prediction model would be created that can easily be converted to real world systems such as building a recommendation system. Finally we will deploy the model on an ec2 instance that will make predictions over an api.
Outline/structure of the workshop
- scraping data from the internet. 0.5 hours
- building the neo4j database. 0.5 hours.
- discusion on the different graph learning models. 0.5
- preparing a collective deep learning model using pytorch. 0.5
- evaluating the model and deploying on an ec2 instance. 0.5
- A linux or mac powered laptop
- basic understanding of python and databases.
- neo4j installed on your laptop and pytorch biggraph installed in the system
The content urls and presentation would be shared soon
Joydeep is a team lead at Nineleaps and his primary interests lie in data engineering and machine learning. He has authored a book "fastText Quick Start Guide", describing a practical approach to using the popular NLP library fastText.
Find him on twitter and linkedin and he is always interested in talking about a great topic.