Mathematics Colloquium - Yan Wang

Warning Icon This event is in the past.

When:
January 11, 2023
2:45 p.m. to 3:45 p.m.
Event category: Lecture
In-person
Speaker: Yan Wang (Iowa State University)

Title: Active Learning for Regression Guided by Out-of-Bag Error Prediction  
 
Abstract: Active learning aims to do machine learning in an economical way by actively picking the “most informative” data points in a pool of data. The feature (or covariate) vector of each data point in the pool is known “for free”, but the corresponding response (or label) will only be revealed for picked data points at some cost. Given a budget, how to select the informative data points stands at the core of active learning. We tackle this problem by first analyzing the constituents of the prediction error for a (stable) learning algorithm, and then propose a method to actively sample data points to reduce the prediction error in a bias-correcting way. In particular, by taking advantage of the out-of-bag (OOB) error of random forests, the implementation of our method is computationally efficient. We might also discuss how the OOB error can be helpful in constructing prediction intervals for active learning.  

Contact

Department of Mathematics
3135772479
math@wayne.edu

Cost

Free
January 2023
SU M TU W TH F SA
1234567
891011121314
15161718192021
22232425262728
2930311234