PhD Candidate
Mila Institute, Quebec
Areas of Interest
- Artificial Intelligence
Poster
From "What" to "Why": towards causal deep learning
Abstract
Deep networks have become powerful tools for perceptual tasks across a range of domains: with state-of-the- art classification results in images, audio, and video. However, the question of whether they capture the underlying causal structure of reality – or instead only capture surface-level, potentially spurious correlations – has become a central and disconcerting question. This issue raises concerns about whether deep networks will generalize well to changing environments and also raises concerns about fairness and whether their judgements will potentially be biased or socially destructive. The central aim of my research is to investigate deep networks which can learn causal structure, and to do in a way that is both theoretically sound and practically relevant. By learning causal structures, we can guarantee systematic generalization, and build AI systems that are adaptable to changing circumstances and dynamic environments.
Bio
Rosemary is a final year PhD student at Mila advised by Yoshua Bengio and Chris Pal. Her primary research interests center around investigating deep networks which can learn causal structure in order to achieve systematic generalization. She is also a recipient of the Facebook PhD fellowship program.
