Machine Learning and Data Science for Medicine - A Vision, Some Progress and Opportunities

Explore the intersection of machine learning, data science, and medicine in this 46-minute talk by Professor Mihaela van der Schaar at the Alan Turing Institute. Dive into the development of AutoPrognosis, an innovative automated system for creating tailored machine learning pipelines in medical settings. Learn how this approach addresses challenges in medical data analysis, including missing data imputation, feature selection, and classifier choice. Discover how AutoPrognosis outperforms existing clinical, statistical, and machine learning methods across various medical datasets. Gain insights into the potential of AI-driven prognostic modeling to revolutionize personalized medicine and improve patient outcomes.

Intro
Machine Learning & Clinical Practice
Machine Learning & Medicine: Vision
Machine Learning in Prognostic Research
Prognostic Modeling using Machine Learning
What is a pipeline?
Ensembles
Goal: predict the performance of ML pipelines
Interpretability
AutoPrognosis: Automating Prognostic Modeling
Related Works
AutoPrognosis: System Overview
AutoPrognosis: Pipeline Components
Automated Pipeline Configuration (1)
The Curse of Dimensionality
The Structured Nature of Hyperparameters
Bayesian Optimisation with Structured Kernel Learning
Benchmarks
Application to Cardiovascular Cohorts
AutoPrognosis: A Solution for MANY problems