Presented By: Maize Pages Student Organizations
MSAIL #3: Active Learning and Global Bayesian Optimization
Dear Friends, Our next MSAIL meeting (#3) will focus on Active Learning andGlobal Bayesian Optimization --- check out these slides for asneak peek! We'll meet at: 3427 EECS, Wednesday, 2016-10-26, 18:00-19:00. The problem is that our loss function is a very complicated functionof our parameters. It's too complicated to compute except by samplingat a few points in parameter-space. Last time, we discussed a famousheuristic applicable when at each sample we have gradient information.This time, we'll discuss a learning framework applicable even when welack gradients. The basic idea is: estimate the loss function by a simpler one by doingregression on known (parameter, loss) pairs. The simpler function willguide queries for further parameters to try. We are led to active learning,wherein the algorithm asks "clarifying questions" by requesting moretraining data. Automatically Learning,MSAIL
Related Links
Co-Sponsored By
- Student Organization: Michigan Student Artificial Intelligence Lab
Explore Similar Events
-
Loading Similar Events...