Machine Learning for Health

我们的目标是开发健康表型，可用于改善医疗保健，识别相似的患者，并量化可能的干预措施的影响。用机器学习针对单个临床问题并不能很好地扩展;相反，我们专注于生理状态或表型的表征学习，因为潜在状态在多个任务中很有用。由于不同的数据可能对特定的患者和任务很重要，我们整合了笔记，程序，诊断，生命体征，实验室和其他信号，拟议的研究将对我们理解健康意味着什么以及人类表型内的差异产生强烈影响。

Our goal is to develop health phenotypes that could be used to improve healthcare, identify similar patients, and quantify the impact of possible interventions. Targeting individual clinical questions with machine learning does not scale well; instead we focus on representation learning of physiological state - or phenotype - because latent states are useful in multiple tasks. Because different data may be important for a particular patient and task, we integrate notes, procedures, diagnoses, vitals, labs and other signals, The proposed research will have a strong impact on our understanding of what it means to be healthy, and variance within the human phenotype.