About the talk
00:13 T1D Management
00:58 Modeling Approaches
01:43 Hybrid Approach
02:08 Glucose Predictions
02:36 Discussion & Future Work
I am a Research Engineer at Apple working on various aspects of machine learning. I also have a part-time position as a Research Scientist at the Paul G. Allen School of Computer Science & Engineering at the University of Washington where I work with Emily Fox in the Department of Statistics and Adrian KC Lee in the Institute for Learning and Brain Sciences. My research at UW focuses on developing statistical and machine learning methods to learn structure underlying data that arise from processes that exhibit complex dependencies such as neuroimaging data. In particular, I am interested in Bayesian nonparametric methods as well as sparse models for high-dimensional data. Additionally, I work on scalable inference algorithms for complex models of large data sets. Previously, I was a Washington Research Foundation Innovation Postdoctoral Fellow in Neuroengineering and Data Science at the University of Washington, jointly sponsored by the Institute for Neuroengineering and the eScience Institute. I received my PhD from the Computer Science Department at Dartmouth College where I was advised by Dan Rockmore. While in graduate schoool, I worked closely with Sinead Williamson on dependent Bayesian nonparametric models and efficient MCMC algorithms for them.View the profile
Emily Fox is a Professor in the Paul G. Allen School of Computer Science & Engineering and Department of Statistics at the University of Washington, and is the Amazon Professor of Machine Learning. She received an S.B. in 2004 and Ph.D. in 2009 from the Department of Electrical Engineering and Computer Science at the Massachusetts Institute of Technology (MIT). She has been awarded a Presidential Early Career Award for Scientists and Engineers (PECASE, 2017), Sloan Research Fellowship (2015), ONR Young Investigator award (2015), NSF CAREER award (2014), National Defense Science and Engineering Graduate (NDSEG) Fellowship, NSF Graduate Research Fellowship, NSF Mathematical Sciences Postdoctoral Research Fellowship, Leonard J. Savage Thesis Award in Applied Methodology (2009), and MIT EECS Jin-Au Kong Outstanding Doctoral Thesis Prize (2009). Her research interests are in large-scale Bayesian dynamic modeling and computations.View the profile
Thanks for being the short summary of the paper learning insulin, glucose Dynamics in the wild. This work was done with Nicholas voting and diabetes is a chronic condition in which little-to-no in frustrating the regulation of blood, glucose requires constant Management, Monterey glucose levels meticulously track meals. Dangerously low glucose levels. Or cast of like we both can play a role in the management of T20 here. We focus on using data recorded by a continuous glucose monitor
on an insulin, pump diet, log-in activity measurements. Existing coaches for forecast in blood glucose. Digital approaches like a classic Karma, X Series models. For machine learning models, like find pattern from observational data forecast. Alternatively physiological models. Such as the minimum model for the easy a Padawan t1d simulator, describe the Dynamics of glucose insulin in carbohydrates in the body often with a set of differential equation.
What is Mount? Olympus real estate Dynamics they can be described. We take a hybrid approach. We incorporate a sociological t1d simulator. DVA patawad TMD simulator component describes the individuals physiological state. In the Layton State. This model allows the physiological t1d simulator parameters, such as insulin, sensitivity carbohydrate, absorption rate to fluctuate in time. We find that will the static t1d simulator struggles to model real CGM? Data The
increased flexibility of the hybrid approach allows for more Additionally, the hybrid model, inherits realistic constraints from the t1d simulator while it clearly statistical modeling, physically inconsistent forecast, such as glucose Rising after bolus insulin, do we find that the hybrid model makes physiologically possible forecast in your contacts? To summarize. We should a hybrids distance to and physiological model can enjoy the best of both world. The expressively to describe a real-world CGM data and inductive is to generate physiologically consistent forecast, long-range blood
glucose forecast in remains a challenge in their analysis was limited to a small Coke work without spatial data, larger. Follow-up studies will be needed to work me up tomorrow meals and Incorporated. Thanks for listening.
Buy this talk
Buy this video
Our other topics
With ConferenceCast.tv, you get access to our library of the world's best conference talks.