Exploring Deep Learning in Epidemiology: A Look at Dr. Jing Li’s Upcoming Seminar
Join us on Wednesday, October 22, for an enlightening session in the “Works In Progress” Epidemiology Seminar Series. This time, the spotlight will be on Dr. Jing Li, a prominent Postdoctoral Research Fellow at the Harvard T.H. Chan School of Public Health. Dr. Li will present her groundbreaking research on a Deep Learning–Based Estimator for the Non-Iterative Conditional Expectation (NICE) g-Formula.
Understanding the NICE g-Formula
The g-formula is a powerful statistical tool that allows researchers to estimate the causal effects of treatment strategies using observational data. When utilized correctly, it provides insights into the impact of sustained treatments under certain assumptions: consistency, positivity, and exchangeability.
Dr. Li’s focus on the NICE estimator introduces a nuanced approach by utilizing deep learning techniques. Traditional methods that rely on parametric models to estimate the conditional distribution of assorted variables can often lead to model misspecification. This misalignment can bias causal estimates, especially in high-dimensional or nonlinear contexts, thereby skewing results and compromising the validity of research findings.
In her presentation, Dr. Li will unveil her unified deep learning framework tailored for the NICE g-formula. By incorporating recurrent neural networks, this innovative framework makes it easier to effectively model the joint conditional distribution of time-varying variables.
Unveiling the Research Framework
Through a series of simulation studies, Dr. Li and her team demonstrate how their deep learning–based estimator surpasses conventional parametric methods. They highlight key advantages such as improved accuracy in capturing complex relationships and an ability to quantify uncertainty within the estimates. This not only enhances statistical integrity but also opens new avenues for understanding treatment effects in epidemiological research.
Speaker Bio: Dr. Jing Li
Dr. Jing Li’s credentials reflect her expertise and significant contributions to both causal inference and machine learning. As a Postdoctoral Research Fellow at CAUSALab within the Department of Epidemiology at Harvard, she studies the intersection of these fields to refine how we comprehensively assess treatment outcomes in complex, longitudinal datasets.
Her ongoing investigations revolve around advancing the g-formula methods while developing deep learning techniques that better estimate sustained treatment effects. Dr. Li’s work has seen publication in esteemed forums such as CVPR, ICLR, AAAI, and more. Furthermore, she is the primary developer of the open-source Python package pygformula, contributing to several other causal inference packages as well.
Dr. Li earned her Ph.D. in Computer Science from Peking University. During her academic journey, she learned under the guidance of Prof. Yizhou Wang. Moreover, her solid foundation in statistics comes from her undergraduate studies at Wuhan University, where she received a bachelor’s degree in Statistics.
Event Details and Participation
Don’t miss out on the chance to engage with Dr. Li as she delves into an exciting blend of epidemiology and machine learning. The seminar promises to be an informative event for anyone interested in modern approaches to causal inference in public health settings.
The session not only highlights the importance of deep learning in enhancing traditional epidemiological methods but also paves the way for future innovations in the field. Be sure to mark your calendars for this insightful opportunity!