Spring 2017

Exponential Computational Improvement by Reduction

Wednesday, May 3, 2017 2:45 pm3:30 pm

Add to Calendar

In Learning Reductions, you reduce to a simpler machine learning problem, apply a solver for that problem, and then use the solution on the simpler problem to solve a more complex problem.  Learning reductions have been used to create exponentially more efficient solutions to Multiclass classification, Contextual Bandit learning and exploration, and Active Learning.   I will discuss several families of learning reductions, their applications, and limits.