Representation Learning in Knowledge Graphs - BC-737

Project type: Research
Desired discipline(s): Engineering - computer / electrical, Engineering, Computer science, Mathematical Sciences, Mathematics
Company: AI Garage, Mastercard India Services Pvt. Ltd.
Project Length: 6 months to 1 year
Preferred start date: As soon as possible.
Language requirement: English
Location(s): Toronto, ON, Canada; Canada; Canada
No. of positions: 1
Desired education level: CollegeUndergraduate/BachelorMaster'sPhDPostdoctoral fellowRecent graduate
Open to applicants registered at an institution outside of Canada: Yes

About the company: 

We work to connect and power an inclusive, digital economy that benefits everyone, everywhere by making transactions safe, simple, smart, and accessible. Using secure data and networks, partnerships, and passion, our innovations and solutions help individuals, financial institutions, governments, and businesses realize their greatest potential. Our decency quotient, or DQ, drives our culture and everything we do inside and outside of our company. We cultivate a culture of inclusion for all employees that respects their individual strengths, views, and experiences. We believe that our differences enable us to be a better team – one that makes better decisions, drives innovation, and delivers better business results. At AI Garage, we use state-of-the-art AI techniques to solve some of the most important problems in the financial world.

Describe the project.: 

Incorporating properties of a node in a heterogenous multi relation graph (Knowledge Graph) into a vector has an immense number of applications in the AI domain. Mastercard’s transaction data along with the attributes of entities (cardholder and merchant) involved in the transaction provides an opportunity to build a highly informational KG, which could further be used to learn the representation of each of those entities and use them for many downstream tasks like detecting fraud entities, anomalous communities, etc.

In recent literature, Graph Neural Networks have been primarily used for such a task. However, some studies show that the formulation of the traditional message passing function of GNNs assumes the Graph structure to be tree-like (no short length loops). It is important to note that most of the KG datasets in the literature have a large number of loops. There is very little if any, study done in extending the message passing function to our field relevant downstream tasks (link prediction, node classification, community detection, etc.) on KGs.

We believe that there is a scope for formulating a novel message passing function for KG which factors in the short length loops. This would help in learning optimal representations of entities resulting in better performance in downstream tasks and would promote a new track of research aimed at exploiting the information in KGs better.

Required expertise/skills: 

  • Good theoretical and practical familiarity with Deep Learn Models
  • Decent understanding of Graph Neural Networks formulations
  • Good understanding of Machine Learning theory
  • Decent understanding of probability theory and statistics
  • Experience in Knowledge Graphs is a plus
  • Good experience with packages such as Pytorch and Tensorflow