Dr. Ashtiani’s research interests revolve around Machine Learning, Statistics, and Theoretical Computer Science. Some of the research directions that he is currently working on are:
Differentially Private Machine Learning
Statistically/Computationally Efficient Distribution Learning
Learning in Presence of Adversarial Perturbations or Data Poisoning
Unsupervised Domain Alignment
Machine Learning under Distribution Shift
Modern Generalization Bounds for Supervised Learning
Prospective students are encouraged to consult Dr. Ashtiani’s personal homepage about open positions.
Block Heading
Dr. Hassan Ashtiani is an Associate Professor in the Department of Computing and Software at McMaster University, and a faculty affiliate at the Vector institute. He joined McMaster University in 2018 after obtaining his Ph.D. in Computer Science at University of Waterloo in the same year. Before that, he received his master’s degree in AI and Robotics and his bachelor’s degree in computer engineering, both from University of Tehran. Broadly speaking, a major theme in his research is the design and analysis of sample-efficient learning algorithms. In recent years, he has focused on studying sample-efficient learning methods that are robust to (i) privacy-related attacks, (ii) data poisoning or test-time adversarial attacks, (iii) distributions shift, (iv) and model misspecification. Dr. Ashtiani has routinely served as an area chair or program committee for flagship machine learning conferences such as NeurIPS. He is one of the recipients of NeurIPS best paper award in 2018 for introduction of distribution compression schemes for learning Gaussian mixtures.
Related Courses
The main focus of the course is on the modern aspects of distribution learning. The topics include PAC learning of distributions, efficient learning of mixture models, information theoretic aspects of distribution learning, computational complexity distribution learning, models for robust distribution learning, differentially private distribution learning, open questions in modern distribution learning, generative adversarial networks, and variational autoencoders. The course will be mostly based on recent papers, but some introductory aspects can be found in the book titled “Combinatorial Methods in Density Estimation” by Luc Devroye and Gabor Lugosi.
