Our lab focuses on several research directions, primarily representation learning, behavioral machine learning, machine learning for healthcare, and "healthy" machine learning.
We list a selection of representative work here, with a complete listing in Publications.
Representation learning has prompted great advances in machine learning; for example, the lower dimensional, qualitatively meaningful representations of imaging datasets learned by convolutional neural networks. Healthcare data lacks such obviously natural structures, and investigations into appropriate representations should include multi-source integration, and learning domain appropriate representations.
- Unfolding Physiological State: Mortality Modelling in Intensive Care Unit (KDD 2014);
- A Multivariate Timeseries Modeling Approach to Severity of Illness Assessment and Forecasting in ICU with Sparse, Heterogeneous Clinical Data (AAAI 2015);
- Clinical Intervention Prediction and Understanding using Deep Networks (MLHC 2017/JMLR W&C V68);
- Semi-supervised Biomedical Translation with Cycle Wasserstein Regression GANs (AAAI 2018);
Much of life happens outside a clinical environment. Phenotyping is an important goal in healthcare, and behavioral data provides an ongoing way for devices to collect continuous non-invasive data and provide meaningful classifications or alerts when the patient is in need of clinical attention. We target the development and deployment of machine learning on behavioral data, working to understand the variance of human behavior's impact on health.
- Learning to detect vocal hyperfunction from ambulatory necksurface acceleration features: Initial results for vocal fold nodules (IEEE TBME 2014);
- Uncovering Voice Misuse Using Symbolic Mismatch (MLHC 2016/JMLR W&C V56);
- Project Myalo with Verily
While healthcare is an inherently data-driven field, most clinicians operate with limited evidence guiding their decisions. Randomized trials estimate average treatment effects for a trial population, but many day-to-day clinical decisions are not based on high-quality randomized control trials. We develop machine learning models to understand the use of important interventions, and the development of outcomes.
- Predicting early psychiatric readmission with natural language processing of narrative discharge summaries (Trans Psych/ Nature 2016)
- Predicting Intervention Onset in the ICU with Switching State Space Models (AMIA-CRI 2017)
- Continuous State-Space Models for Optimal Sepsis Treatment-a Deep Reinforcement Learning Approach (MLHC 2017/JMLR W&C V68);
Machine learning models are powered by data, and bias can be encoded by data itself or modeling choices. With the expanding impact of machine learning in sensitive areas like healthcare, we work to identify the potential for bias in data, learning and deployment.
- Modeling Mistrust in End-of-Life Care; (MLHC 2018 Preprint). (FATML 2018 Workshop Preprint).
- Exploring Toxicity of Song Lyrics Using Machine Learning