website

Qiyao Wei

My website is https://qiyaowei.github.io/website/ Reach me at q.wei@mail.utoronto.ca

Hi there! You found my webpage! This is Qiyao (chi-yao), a University of Toronto undergrad inspired by the advances and accomplishments of Artificial Intelligence. My short-term research interest is to make AI more about theory and justification, rather than network design and hyperparameter tuning. My long-term goal is to crack the code of human intelligence, or to achieve human-level intelligence in our machines.

Publications and Presentations

• Here is a technical report on mirror descent, Bregman Divergences, and an analysis on Krichene and Bartlett (2017). Here is a presentation I gave to Professor Yinyu Ye at Stanford on this paper.
• Epidemic Control with Reinforcement Learning, only single-node for now. Here is a presentation I gave to our research group at Vector.

Favorite Readings

This is where I put some of the articles or papers I read, and will recommend to others. Specifically, they will either convey some inspiration that I could not have summarized better, or illustrated a theory very clearly

• Ever wondered where all the reparametrization in Stochastic Variational Inference came from?
• In my opinion, the epitome of combining theory and practice: how machine learning should be done. Variational Bayes
• Cute papers. Distillation, Rank AE, ReLU bottleneck, NIPS experiment

Favorite quotes

This is where I record some of the best (i.e. pithy but elucidating) words I came across. For a total stranger, they will require context to make sense. I always welcome discussions of stuff like this.

• “Why does the wind blow?” “Because the leaves move” —Yann LeCun interview with Lex Fridman
• E.T.Jaynes page 103 “how we solve a problem when it becomes too difficult”
• Von Neumann “You should call it entropy, for two reasons. In the first place your uncertainty function has been used in statistical mechanics under that name, so it already has a name. In the second place, and more important, nobody knows what entropy really is, so in a debate you will always have the advantage”

Miscellaneous

There are also beautiful words that are not AI related.

• The Great Gatsby and his “extraordinary gift of hope”
• Top secret

What if I am just getting started?

• Here is a nice introductory paper about MLPs and RNNs
• Here is my favorite deep reinforcement learning course. A word of caution about videos—come to them with questions and learning will go smoothly, if watched without active learning then not much can be recalled
• Here is a great introductory post on NTK, coming from one of my favorites blogs