Booz Allen Hamilton Colloquium: "Learning to Benchmark," Alfred Hero, University of Michigan

Friday, November 16, 2018
3:30 p.m.-4:30 p.m.
1110 Jeong H. Kim Engineering Building
Kara Stamets
301 405 4471
stametsk@umd.edu

"Learning to Benchmark"

Alfred O. Hero, III
John H. Holland Distinguished University Professor of EECS; R. Jamison and Betty Williams Professor of Engineering
University of Michigan

Abstract: We propose a framework for learning the intrinsic difficulty of classifying a labeled training sample, based on empirical estimation of the minimal achievable classification error, i.e., the Bayes error rate. We call this meta-learning problem "learning to benchmark" with the objective of finding low complexity and statistically consistent estimates of the Bayes mis-classi fication error rate that bypasses the problem of approximating the Bayes-optimal classi fier. The Bayes classification error probability can be represented as an information divergence measure and an ensemble of geometric learners will be shown to solve the learning to benchmark problem with optimal (linear) rates of implementation complexity and mean squared error convergence.

Bio: Alfred Hero is the John H. Holland Distinguished University Professor of Electrical Engineering and Computer Science and the R. Jamison and Betty Williams Professor of Engineering at the University of Michigan. He is also co-Director of the University’s Michigan Institute for Data Science (MIDAS).

At the University of Michigan his primary appointment is in the Department of Electrical Engineering and Computer Science (EECS) and he has secondary appointments in the Department of Biomedical Engineering and the Department of Statistics.

He is also affiliated with the UM Center for Computational Medicine and Bioinformatics (CCMB), and the UM Graduate Program in Applied and Interdisciplinary Mathematics (AIM).

His research is on data science and developing theory and algorithms for data collection, analysis and visualization that use statistical machine learning and distributed optimization. These are being to applied to network data analysis, personalized health, multi-modality information fusion, data-driven physical simulation, materials science, dynamic social media, and database indexing and retrieval.

Audience: Clark School  Graduate  Undergraduate  Faculty  Post-Docs  Alumni 

 

October 2019

SU MO TU WE TH FR SA
29 30 1 2 3 4 5
6 7 8 9 10 11 12
13 14 15 16 17 18 19
20 21 22 23 24 25 26
27 28 29 30 31 1 2
Submit an Event