Consider a question: “How many students took each calculus class?” A seemingly simple inquiry that soon turns complicated if you are trying to teach an artificial neural network to produce a database query based on the question – something that could be usefully applied in an academic setting.
“We want neural models that can discover, that, for example, when predicting \Class.name” as the grouping attribute, the relevant part of the question to focus on is \each calculus class”, or even just \class””, says Dzmitry Bahdanau, a newly appointed Adjunct Professor in the School of Computer Science, describing one of his research focuses – what AI researcher refer to as ‘learning rationales to induce context-independence.’
One of the first to develop some of the first successful neural speech recognition systems, Bahdanau’s contributions to the deep learning revolution in natural language processing (NLP) are widely recognized.
Today, Bahdanau and three other emerging McGill AI research leaders – Adam Oberman, David Rolnick and Xujie Si – garnered the support of a dedicated research funding program through the Canadian Institute for Advanced Research (CIFAR). They are among the 29 world-class AI researchers named CIFAR AI Chairs.
AI researcher to watch
Rolnick, an Assistant Professor in the School of Computer Science, is named by CIFAR as one of four AI researchers to watch in 2021. He is building a field applying machine learning algorithms to tackle one of the greatest problems of our time: climate change. Rolnick is the co-founder of Climate Change AI, an initiative that brings together experts from industry, academia, and policy to use machine learning to help mitigate climate change and adapt to its consequences.
“I am thrilled to join the CIFAR AI Chairs program, which brings together some of the world’s foremost researchers in AI,” says Rolnick. “This community also represents an exceptional opportunity for synergy between research, industry, and policy, which is essential for enabling impactful AI work on topics such as climate change.”
Across Canada, the newly appointed AI Chairs are advancing research in a wide range of areas, including machine learning for health and responsible AI. The Canada CIFAR AI Chairs program, a cornerstone of the Pan-Canadian AI Strategy, aims to attract outstanding researchers to Canada by providing them with long-term, dedicated funding to pursue innovative ideas. Since 2017, 57 researchers have taken up their first faculty position in Canada as Canada CIFAR AI Chairs. 1,200 graduate and postdoctoral fellows have been trained at the AI institutes (Amii, Mila and the Vector Institute).
“I am delighted to congratulate the newest McGill cohort of Canada CIFAR AI Chairs”, says Martha Crago, Vice-Principal, Research and Innovation. “They will join a growing network of talented AI researchers at Mila, who are seeking to transform many human pursuits while also developing novel interactions between academia and industry. With CIFAR’s support, AI researchers are equipped to advance the field of Machine Learning and to tackle the problems we don’t yet know how to solve, such as climate change.”
Each of McGill’s four CIFAR AI Chairs will be affiliated with Mila a Montreal-based research institute and partnership between McGill and the Université de Montréal, closely linked with Polytechnique Montréal and HEC Montréal.
From climate change to health sciences
For Rolnick, innovations in deep learning can be driven by a mathematical understanding of which functions different neural networks are able to express and learn. Through collaboration with experts in domains such as electricity systems, ecology, and atmospheric science, Rolnick uses machine learning to help reduce greenhouse gas emissions and increase society’s resilience to the effects of climate change.
As a CIFAR AI Chair, Adam Oberman, Full Professor in the Department of Mathematics and Statistics, has set a research goal to develop a rigorous mathematical understanding of deep learning algorithms, through the lens of generalization, robustness and averaging. Resolving the issues of robustness and generalization of deep neural networks is essential before these methods may be adopted in applications where reliability and safety are the primary concern, for example in the example of self-driving cars. The theory, techniques, and applications, as well as the algorithms and models developed under this research project, will be broadly applicable to a wide variety of fields, including Health Sciences.
Xujie Si, Assistant Professor, School of Computer Science, brings an interesting new angle to Mila’s research efforts. Si’s research lies in the intersection of programming languages and machine learning. He is interested in developing learning-based techniques to help programmers build better software with less effort and apply program reasoning techniques to improve data-efficiency of machine learning. His expertise in programming languages, particularly, inductive logic programming, program synthesis, and program verification, will complement Mila’s research in several important sub-fields of AI, such as symbolic AI, interpretable learning, AI robustness and reliability.
Bahdanau looks forward to the research opportunities afforded by the CIFAR AI Chair program: “This is a one-of-a-kind opportunity for me as an industrial researcher to maintain active ties with AI academia in Montreal and across Canada,” he says. “I am looking forward to supervising graduate students and to interacting with other Chairs. I believe that the resulting collaborations will be uniquely fruitful for my research ambition to enable wide-spread use of AI-powered language user interfaces, all while being a great way for me to contribute back to the academic research community that I come from.”
Read the CIFAR announcement here