By: Justine Brooks
23 May, 2024
Six new CIFAR AI Catalyst Grants awarded
May 23, 2024
Projects address pressing questions in responsible AI, energy and the environment and fundamental research.
The latest round of CIFAR AI Catalyst Grants follows efforts of CIFAR’s Working Groups in Energy & Environment and Responsible AI, and addresses the “Advancing AI Science” priority area of the Pan-Canadian AI Strategy.
Six new CIFAR AI Catalyst Grants were selected as part of the 2023 round of applications in three streams: Fundamental AI, Energy and Environment and Responsible AI. The winning projects were chosen due to their creativity and innovation in pursuing high-risk, high-reward research with the potential for broad impacts in the fields of AI and machine learning. Each will each receive up to $50,000 per year for up to two years to support collaborative research and exchange with current Canada CIFAR AI Chairs. Three smaller grants were also awarded in support of conference activities.
Fundamental AI
Generating Images with Multimodal Instruction
Wenhu Chen (Vector Institute; University of Waterloo) & Aishwarya Agrawal (Mila; Université de Montréal)
Canada CIFAR AI Chairs Wenhu Chen and Aishwarya Agrawal are exploring a new method of creating image-based generative models. By combining both text and image inputs, they seek to create a model that can generalize to conditions other than those it was trained on. They expect to create both novel machine learning methods and a practical online demonstration for the public to generate images.
Fundamentals of Transformers: From Optimization to Generalization
Murat Erdogdu (Vector Institute; University of Toronto) & Christos Thrampoulidis (University of British Columbia)
Transformers have become the de facto architecture in various applications, ranging from natural language processing and generation, to computer vision and time-series forecasting. Canada CIFAR AI Chair Murat Erdogdu is partnering with Assistant Professor at the University of British Columbia Christos Thrampoulidis to provide transformers with concrete optimization and statistical guarantees, a comprehensive assessment of their limitations, and informed choices in the design of faster algorithms and efficient architectures.
Energy & Environment
Solving Simulators with Reinforcement Learning for Material and Process Design
Martha White (Amii; University of Alberta), Mouloud Amazouz (Natural Resources Canada; University of Waterloo), & Ahmed Ragab (Natural Resources Canada; Polytechnique Montréal)
Material and process design simulators for clean fuel production can be expensive. Applying reinforcement learning optimization could help solve this problem more efficiently. An outcome of the CIFAR Energy & Environment Working Group, this project brings together Canada CIFAR AI Chair Martha White with Natural Resources Canada researchers Mouloud Amazouz and Ahmed Ragab in partnership with Natural Resources Canada. Beyond improving simulators, this research could also have broader uses for AI models applied to energy and environmental issues.
Responsible AI
AI Auditing Through Exploration of Model Multiplicity
Golnoosh Farnadi (Mila; McGill University) & Elliot Creager (University of Waterloo; Vector Institute affiliate)
Canada CIFAR AI Chair Golnoosh Farnadi and Assistant Professor at the University of Waterloo Elliot Creager are aiming to develop better auditing tools for multimodal foundation models by understanding how user input personalizes a model’s behaviour. The goal of this project is to have a significant impact across fields where AI is being used for decision-making, promoting trust, robustness and reliability in emerging AI technologies.
Robust Strategic Classification and Causal Modelling for Long-Term Fairness
Nidhi Hegde (Amii; University of Alberta) & Dhanya Sridhar (Mila; Université de Montréal)
Machine learning algorithms are increasingly being used to make impactful and potentially life-changing decisions about individuals. When errors occur in these models, they disproportionately affect minority subgroups. Canada CIFAR AI Chairs Nidhi Hegde and Dhanya Sridhar are taking an alternative approach to fairness, the goal being to allow individuals labelled with ‘undesirable outcomes’ to achieve ‘desired’ outcomes long-term. Hegde and Sridhar will investigate this algorithmic recourse and improvability in outcomes to improve long-term justice.
Natural Language Processing for Users from Diverse Cultures
Vered Shwartz (Vector Institute; University of British Columbia) & Siva Reddy (Mila; McGill University)
Current English Large Language Models (LLMs) are trained on primarily American user data, leading to a narrow Western-centric lens. This rigid Western perspective can lead to models perpetuating and reinforcing stereotypes and inequalities, limiting their effectiveness for non-Western users. Canada CIFAR AI Chairs Vered Shwartz and Siva Reddy will evaluate the cultural awareness of current LLMs with the goal of enhancing their cultural competence.
In addition to the above grants, three proposals were accepted to support conference-related expenses, with particular consideration given to conferences that support opportunities for equity-deserving community members.
- Collaborative AI and Modelling of Humans, Matthew Taylor (Amii; University of Alberta) & Cristina Conati (University of British Columbia)
- IEEE Conference on Secure and Trustworthy Machine Learning 2024, Nicolas Papernot (Vector Institute; University of Toronto)
- International Conference on Automated Planning and Scheduling 2024, Sheila McIlraith (Vector Institute; University of Toronto) & Nathan Sturtevant (University of Alberta)
Related Articles
-
Three 2024 Nobel Laureates among CIFAR’s acclaimed community of researchers
October 15, 2024
-
Racing for cause
June 13, 2024
-
How can we make AI that works for everyone?
May 14, 2024
-
AI Canada! Premier global AI conference includes strong Canadian representation
December 15, 2023