Join us for an invited lecture with Dr. Finale Doshi-Velez (School of Engineering and Applied Sciences; Harvard University):
- Title: “Interpretable and Predictive Generative Models: Case Study of Deriving Data-Driven Phenotype Trajectories in Autism Spectrum Disorders“.
- Event Date: Thursday, November 12; 2015, 16:00 am
- Location: 4.0.E05 Room; Torres Quevedo Building; Leganés Campus; Universidad Carlos III de Madrid.
Latent variable models provide a powerful tool for summarizing data through a set of hidden variables. These models are generally trained to maximize prediction accuracy, and modern latent variable models now do an excellent job of finding compact summaries of the data with high predictive power. However, there are many situations in which good predictions alone are not sufficient. Whether the hidden variables have inherent value by providing insights about the data, or whether we simply wish to improve a system, understanding what the discovered hidden variables mean is an important first step.
In this talk, I will discuss one particular model, GraphSparse LDA, for discovering interpretable latent structures without sacrificing (and sometimes improving upon) prediction accuracy. The model incorporates knowledge about the relationships between observed dimensions into a probabilistic framework to find a small set of human-interpretable “concepts” that summarize the observed data. This approach allows us to recover interpretable descriptions of clinically-relevant autism phenotypes from a medical dataset with thousands of dimensions. I’ll also talk about recent directions for learning the most distinguishing features of each phenotype and also how these phenotypes evolve over time.
About the speaker:
Finale Doshi-Velez is excited about methods to turn data into actionable knowledge. Her core research in machine learning, computational statistics, and data science is inspired by —and often applied to— the objective of accelerating scientific progress and practical impact in healthcare and other domains.
Specifically, she is interested in questions such as: How can we design robust, principled models to combine complex data sets with other knowledge sources? How can we design models that summarize and generate hypotheses from such data? How can we characterize the uncertainty in large, heterogeneous data to provide better support for decisions? Finale Doshi-Velez is interested in developing the probabilistic methods to address these questions.
Prior to joining SEAS, Finale Doshi-Velez was an NSF CI-TRaCS Postdoctoral Fellow at the Center for Biomedical Informatics at Harvard Medical School. She was a Marshall Scholar at Trinity College, Cambridge from 2007-2009, and she was named one of IEEE’s “AI Top 10 to Watch” in 2013.
Signal Processing Group (GTS-UC3M) and CASI-CAM-CM (S2013/ICE-2845).