Our group is committed to conducting fundamental and transformative research that understands and possibly changes the world. While great papers will naturally emerge as a byproduct, publishing for the sake of volume is not our goal. If your focus is on publishing only, our group may not be the best fit for you.
Before you proceed, please read here.
We are looking for students, visitors, postdocs, and industry collaborators within the Department of Statistical Sciences and Department of Computer Science at University of Toronto and MBZUAI. We work on trustworthy AI, efficient GenAI with a focus on 3D and multimodal generation and editing, and foundations of AGI. Please fill this google form if you are interested. Please read the following instructions before getting in touch. I travel frequently, and I apologize in advance if I am slow in responding to your emails.
Your research interests and goals,
A brief outline of your proposed project that you want to explore with us,
How you believe our group could support your work.
You are expected to be strong in theory and/or computing.
Theoretically strong means that you have solid foundation in analysis and probability. You should be familiar with measure concentration, empirical process theory, or mean field asymptotics. Below are some references.
Measure concentration: First 5 chapters of High Dimensional Statstics (HDS) by Wainwright. In addition, you can also refer to Probability in High Dimension by van Handel or High Dimensional Probability (HDP) by Vershynin.
Empirical process theory by Pollard; otherwise you are welcome to read my lecture notes.
Mean-field asymptotics: Please also read the first two chapters of Random Matrix Theory for Machine Learning (RMT4ML) and related application chapters. You can learn more on RMT by referring to A First Course in Random Matrix Theory (reader friendly!) and more on equilibrium statistical mechanisms by referring to Statistical Mechanism of Lattice Systems (SMLS) .
Computationally strong means that you have strong deep learning engineering abilities. You are expected to have papers in top venues. Below are some references.
Reference: Dive into Deep Learning.
You are reasonably familiar with generative AI.
You already have a specific research area, such as model merging, that you are interested in. Please be specific.
You are expected to have worked on an open-source project such as FACIL.
As we continue to explore new areas that may better the society, you are also welcome to pursue your own research directions provided they are under our umbrella.
Postdoc fellows: I may have openings for postdoctoral fellows at MBZUAI and UofT. When inquiring about such positions, please send your CV along with a one-page research proposal and your best 2-3 papers in PDFs and fill in this google form. Joint supervisions are possible. There are also open calls for postdoctoral positions at UofT: Schmidt AI postdoc fellowship, Banting postdoc fellowship, SGS provost postdoctoral fellowship, UTSC provost postdoctoral fellowship, arts and science fellowship, UTSC postdoc fellowship. All fellowships require sponsorships. Feel free to contact me for details.
We currently have a postdoctoral fellow position in GenAI (flow matching and discrete diffusion models).
Visiting students whose research interests closely align with ours are welcome to contact us. Interested candidates should submit your CV, representative papers, and a research plan. Such positions are often externally funded, for example, by university exchange programs or by CSC. We can not provide funding unless in exceptional cases: For example, we can provide financial support for senior PhD students whose research closely aligns with ours. Please fill this google form if you are interested.
Industry collaborators are welcome to get in touch for potential collaborations.
If you are interested in joining our group as interns and students, please read the followings carefully.
Trustworthy AI: We are interested in using statistics to make AI reliable and trustworthy. For example, we can make AI more reliable by protecting it from random noise (and stats is really good at understating noise!) and other conditions such as background/lab effects, adversarial noise, and nonstationarity.
Generative AI: We are currently working on 3D/video/multimodal generation and editing. Students, interns, and visitors are needed!
Foundations of AGI: We are currently working on pushing the boundaries of statistics to understand and advance AI.
you are a UofT undergraduate or master student looking for some research experience and independent research projects for the first time. This course uses independent research projects as part of the evaluation. If you think you have a compelling story though, for example, you are an ICPC world finalist (strong in engineering), feel free to reach out. I apologize in advance if I am slow in responding to your emails.
So long, and thanks for all the tips by Witten D
Checklists for Stat-ML PhD students by Ramdas A
Efficient Writing by Marron JS
Write Statistics Right by Little R
Style and grammar tips for biostatistics and statistics students by Little R
How to be a modern scientist: https://leanpub.com/modernscientist
Grinding PhD: https://www.goodreads.com/en/book/show/15731248-the-phd-grind