Welcome to Mila Tea Talks!

What are tea talks ?

Tea talks are scientific keynote-like talks given at Mila on Friday mornings (10:30am EDT). We either have an internal or invited speaker give a ~1 hour talk with about 15 minutes of questions interspersed throughout. These talks are open to the public, but the vast majority of attendees are Mila students and professors, so the talks are aimed at that level of understanding. We also encourage our speakers to stay after the talk and interact with students in one-on-one or group meetings to discuss research interests and ideas, and we're happy to organize that as well.

What's in this webpage?

You will find the recording and slides (when available) of the past tea talks. We ordered them by session.

Would like to get schedule notifications ?

Simply click here with a quick description of who you are ( XX from YY is enough) to join the external mailing list.

The schedule is also publicly available on Mila website

Would like to suggest a speaker?

You can email us (admin-teatalks@googlegroups.com) and/or fill the suggestion form here!

LiveStream

Talks are Live-streamed on Google Meet. Link here.

Current Session>FALL 24

Friday, 25 October, 2024

Toby Dylan Hocking

Recording : TBD

two new algorithms for scientific applications of machine learning

In the last few years, I have maintained active collaborations with scientists who are not machine learning experts, but who want to use machine learning algorithms for their data analyses.

In many scientific applications of machine learning, two questions come up again and again.

Question 1. we have some data from one region (or time period), so if we use these data to train, will it work on a new region? (or time period)

Question 2. how do we deal with class imbalance?

Example A: forestry. When predicting forest properties based on objects in satellite images, if we train on one state (say Arizona), will it work in Quebec? How to deal with the fact that some objects of interest (trees, burn) are only a small minority of data?

Example B: medicine. When predicting autism diagnosis from other survey responses, if we train on one year of survey data (say 2019), will it work in another year? (say 2020) And can we combine the two years of data to get a better model? How to deal with the fact that autism represents only 3% of the total surveys? (97% of survey respondants did not have autism)

For Question 1, we propose a new algorithm called SOAK (Same/Other/All K-fold Cross-Validation), which can be used to quantify the extent to which it is possible to predict on a given data subset, after training on Same/Other/All data subsets. https://arxiv.org/abs/2410.08643

For Question 2, we propose a new differentiable loss function which can be used to optimize the ROC curve, https://jmlr.org/papers/v24/21-0751.html

Report abuse

Welcome to Mila Tea Talks!

Current Session>FALL 24

Friday, 25 October, 2024

two new algorithms for scientific applications of machine learning

Be our next SPEAKER