The recent success of machine learning has been in large part due to massive success in an area known as representation learning, where a computer algorithm finds a way to identify structure in data. Danica Sutherland’s research focuses on improvements to the process of representation learning, especially using ideas from a tool known as kernel methods. Integrating kernels into currently-popular approaches can, ideally, help learn effective representations with smaller training datasets, and which generalize well even to populations different from those used in training. A major line of her research focuses on representations which identify differences between datasets, such as whether medical images differ between treatment and control groups, or if a generative model has succeeded at matching its goal distribution. She tries to work both on practical problems informed by theoretical viewpoints, and on theoretical problems informed by practice.
- Kamath, P., Tangella, A., Sutherland, D.J., & Srebro, N. (2021). Does Invariant Risk Minimization Characterize Invariance? Artificial Intelligence and Statistics.
- Zhou, L., Sutherland, D.J., & Srebro, N. (2020). On Uniform Convergence and Low-Norm Interpolation Learning. Advances in Neural Information Processing Systems.
- Liu, F., Xu, W., Lu, J., Zhang, G., Gretton, A., & Sutherland, D.J. (2020). Learning Deep Kernels for Non-Parametric Two-Sample Tests. International Conference on Machine Learning.
- Bińkowski, M., Sutherland, D.J., Arbel, M., & Gretton, A. (2018). Demystifying MMD GANs. International Conference on Learning Representations.
- Sutherland, D.J. & Schneider, J. (2015). On the Error of Random Fourier Features. Uncertainty in Artificial Intelligence.
CIFAR is a registered charitable organization supported by the governments of Canada, Alberta and Quebec, as well as foundations, individuals, corporations and Canadian and international partner organizations.