![Members of the Graph Neural Network Lab research team](http://www.uwindsor.ca/dailynews/visualarts/sites/uwindsor.ca.dailynews/files/styles/full/public/900_gnnlab.jpg?itok=4VoOVj41)
The School of Computer Science is about to expand its ability to crunch big data. Its acquisition of highly specialized computing equipment will allow researchers to better tackle problems revolving around information retrieval, social network analysis, and drug design.
Hossein Fani and co-applicants Ziad Kobti and Alioune Ngom received $150,000 from the Canada Foundation for Innovation John R. Evans Leaders Fund (CFI-JELF) and the Ontario Research Fund, along with $51,000 from the School of Computer Science, for their project, “Computing Infrastructure for Deep Learning on Graph-Structured Data.”
The three researchers will help establish a Graph Neural Network Lab (GNN-Lab), the first deep learning laboratory at the University of Windsor with a special interest in applying neural networks on graph-structured data.
Graphs are the cornerstone of modelling mutual relationships between discrete entities. Dr. Fani will focus on information retrieval, Dr. Ngom investigates drug design, and Dr. Kobti and Fani perform social network analysis.
Fani says information retrieval, search engine, and recommender systems currently do not take into consideration the person doing the search or making the request.
“When you use Google or Amazon and they are recommending products or bringing up web pages, they just look at the content of the query, but they don’t look at the user info in social networks.”
For example, Ngom says, by finding similarities between the protein networks of two drugs, researchers might find that an existing drug could possibly be repurposed to treat a new disease.
“This way we can avoid designing a new drug which can take a lot of time and is extremely expensive,” says Ngom.
“For instance, in COVID when they wanted to design a new treatment drug, instead of designing one from scratch, they just looked at the similarities of COVID to other diseases and then tried to re-use the drugs for those diseases — it is cost effective and fast.”
Fani says that because disparate problems are based on graph-structured data, it allows for them to be explored in the same computer science lab.
“If we find an algorithm that can find similarities between drugs, we can find similar algorithm that could find similarities between people or users,” he says.
Previously, the researchers worked independently in their own labs. Now, as a team, they can work together to find the best algorithms.
“This will provide the first steps towards realizing new cross-discipline linkages that come with new machine learning methods or AI,” says Kobti.
“With the GNN-Lab, UWindsor will serve as a hub for developing advanced techniques for all types of graph-structured data — from social and collaborative networks of people to protein-protein networks.”
The GNN-Lab will consist of a Graphics Processing Unit server that provides fast and flexible computing. In addition, a Natural Sciences and Engineering Research Council of Canada (NSERC) Research Tools and Instruments grant of $32,010 rounded out the lab with nine small-scale GPU-enabled computers.
“On the new computers, students can prototype their artificial neural network models to see if they’re viable. Normal computers don’t have this capability,” says Fani.
“Once they want to run these models with huge data, they’ll need the GPU server as the central computation server of the proposed lab.”
The GNN-Lab will provide opportunities for new interdisciplinary collaborations, both nationally and internationally, ultimately increasing the capacity and quality of student training opportunities.
“With a significant increase in research partnerships, with the proposed equipment, we will meet the growing industry demand for GNN expertise,” Fani says.
He adds the team is grateful to Nicole Noel, research funding officer in the Office of Research and Innovation Services, for her supplying helpful comments on their application and assisting in the submission process.
For more info, visit https://fani-lab.github.io/.