Title: Learning from Coarse Samples
| Speaker: | Felix Zhou |
| Affiliation: | University of Waterloo |
| Location: | MC 6029 |
Abstract:Coarsening occurs when the exact value of a sample is not observed.
Instead, only a subset of the sample space containing the exact value is known.
Coarse data naturally arises in diverse fields, including Economics, Engineering, Medical and Biological Sciences, and all areas of the Physical Sciences.
One of the simplest forms of coarsening is rounding, where data values are mapped to the nearest point on a specified lattice.
We survey applications of coarse learning to regression with self-selection bias, regression with second-price auction data, and present details of an SGD-based algorithm for coarse Gaussian mean estimation.
Based on joint work with Alkis Kalavasis and Anay Mehrotra (https://arxiv.org/abs/2504.07133)