Artificial intelligence (AI) is changing the way we do business and influencing many other facets of our everyday lives, leading to innovations like self-driving cars and virtual assistants, such as Siri and Alexa. Deep-learning algorithms are the driving force behind almost all recent AI innovations. Despite this, there are clear signs that the performance of these learning algorithms deteriorates after we move them outside of the lab and into the real world.
Dr. Aaron Courville, Canada Research Chair in Learning Representations that Generalize Systematically, is improving the ability of deep-learning algorithms to extrapolate learned behavior to unseen situations that are similar to their training data, a process known as “systematic generalization.” He and his research team are exploring two broad categories of learning algorithms: self-supervised learning methods and iterated learning. Ultimately, they aim to mathematically formalize the notion of systematic generalization and analyze promising avenues to achieve it.