Room: Owl & Eagle Suite

Abstracts

Intro

Abstract

Robust Machine Learning for Clinical (and other Critical) Settings

Abstract

While the predictive powers of (most recently, deep) learning algorithms have pushed the boundaries of what’s possible, the existing tools are limited in crucial ways. These models depend precariously on superficial statistics of the training data, it’s still not clear how to estimate their uncertainty, and they offer predictions without reasons—and are thus are generally misapplied when used to guide decisions. While concerns about these issues are often vacuously expressed (“I don’t trust the model’), and the proposed solutions similarly ill-defined (“add explanations!”), the problems are real and formidable. In this talk, I’ll focus on adapting under distribution shift, discussing challenges of applying supervised learning-based methods in critical settings, the limits of past approaches, and some of my recent work on building robust models.

Screening for medical conditions using search engine queries and search advertising

Abstract

Studies have shown that the traces people leave when browsing the internet are indicative of their medical condition. Recently, these traces have been used to screen for serious medical conditions including Parkinson’s disease, diabetes, and several types of cancer. In my talk I will begin with an overview of these studies, focusing on how an anonymous cohort of patients can be identified, and on the utility of these traces as screening tools. I will focus on our recent study which showed that the adaptive engines of advertising systems working in conjunction with clinically verified questionnaires can identify people who are suspected of having one of three types of solid tumor cancers. First, a classifier trained to predict suspected cancer inferred from questionnaire response using past queries on Bing reached an Area Under the Curve of 0.64. Second, using the Reinforcement Learning mechanism of the conversion optimization engine, the Google advertisement system learned to identify people who were likely to have symptoms consistent with cancer, such that after a training period of approximately 10 days, 11% of people it selected for showing of targeted campaign ads were found to have suspected cancer. People who received information that their symptoms were consistent with suspected cancer increased their searches for healthcare utilization and maintained it for longer than people whose symptoms were not associated with suspected cancer, indicating that the questionnaires provided useful information to people who completed them. These results demonstrate the utility of using search engine queries to screen for possible cancer and the application of modern advertising systems to identify people who are likely suffering from serious medical conditions.