Machine Learning Seminar

Speaker: Baskar Ganapathysubramanian, Iowa State University

Title: Machine learning applications in plant phenotyping and materials-by-design

Refreshments available at 2 pm.

Date: Mon, 24 Sep 2018, 2:15 pm – 3:15 pm
Type: Seminar
Location: 1400 BPS Bldg.

Abstract:
Machine learning approaches have been deployed with signifi­cant suc­cess in a variety of applica­tions. This has been driven in part by the availa­bility of larger and larger data­sets that can be utilized to train the models. While collect­ing data no longer seems to be an issue, a bottle­neck now is the availa­bility (i.e., crea­tion) of good, labelled data. Label­ling data either re­quires human experts to pains­taking­ly assess the data (agri­culture applica­tion) or have a compu­tational­ly expen­sive simula­tor inter­rogate the data (mater­ials-by-design applic­a­tion). There has been in­creasing interest in doing ‘more with less’ with respect to availa­bility of labeled data. I will dis­cuss some of the col­labora­tive work per­formed in this area, which is driven by the applica­tions in agri­culture and materials science.

The first application (maize tassel detec­tion in fields) in­volves ob­ject detect­ion in a large set of clut­tered, noisy images. I will dis­cuss how crowd sourcing is used as a means of col­lect­ing noisy labels, and out­line ap­proaches used to uti­lize these labels to boot­strap to bet­ter labeled data. The second applica­tion (micro­struc­ture-sen­sitive proper­ty) involves micro­struc­ture design for (organic) photo­voltaic applica­tion. The micro­struc­ture → prop­erty simu­lator is compu­tational­ly expen­sive. I will dis­cuss how a small labeled data set is created by adap­tively samp­ling from the input space. The deep learn­ing model build using this data is sub­sequently used to rapid­ly design opti­mal micro­struc­tures exhibit­ing enhanced photo­voltaic per­formance.

This work is collaborative work with the Schnable group (agri­culture applica­tion) and the Sarkar group (both applica­tions)

Speaker Bio:
Baskar Ganapathy­subramanian is a Professor of Mechan­ical Engineering at Iowa State Univer­sity. He runs a curios­ity-driven compu­ta­tional science research group. His research inter­ests are in leverag­ing con­cepts in applied mathe­matics and scien­tific compu­ting to solve prob­lems in sustaina­bility science—with a specific focus on agri­culture, health and ener­gy. Ganapathy­subramanian com­pleted his PhD and MS from Cornell Univer­sity and holds a BS degree from the Indian Insti­tute of Tech­nology-Madras. He is a faculty fel­low of the Plant Science Insti­tute at Iowa State Univer­sity.