Start Date: | 5/2/2018 | Start Time: | 4:00 PM |
End Date: | 5/2/2018 | End Time: | 5:30 PM |
|
Event Description
BIOMED Seminar
Title:
How to Do Good Machine Learning
Speaker: Konrad P. Kording, PhD Penn Integrates Knowledge (PIK) Professor Department of Neuroscience, Perelman School of Medicine Department of Bioengineering, School of Engineering and Applied Science University of Pennsylvania
Abstract: Smart phones and other wearable sensors, increasingly embedded in everyday life, have spurred rapid accumulation of our daily life data. We also have exponentially growing databases of photographs and videos. We thus may hope that we can translate these datasets into insights that are scientifically exciting or medically useful. The approaches promise to improve fields as diverse as: cardiovascular disease, falls, measuring rehabilitation outcomes in stroke and amputees, monitoring Parkinson’s disease symptoms and detecting depression. And yet, despite all the excitement, machine learning's impact on healthcare is, at least so far, limited.
So where does this gap between promise and impact come from? I argue that the algorithms are often wrongly evaluated, and in a way that makes the promise seem too easy and fast. Worse, today's questions are mostly causal questions, and the standard predictive paradigm is quite limited in its ability to make causal prescriptions.
Biosketch: Konrad Kording, PhD, has appointments in the Perelman School of Medicine and the School of Engineering and Applied Science, both at the University of Pennsylvania. His work discovers the structure in big data sets to understand the brain, improve healthcare, and improve science.
Dr. Kording works on diverse areas, including the automatic detection of image fraud in millions of papers, personalized medicine using mobile phones, and algorithms to control prosthetic devices. His lab continues to seek out new areas in which data can yield scientific insights and improve lives. |
|
Location: Papadakis Integrated Sciences Building (PISB), Room 120 |
Audience: Undergraduate StudentsGraduate StudentsFacultyStaff |
|