Overview

The goal of the research roundtables is to foster discussion between the participants and senior researchers in the field on several topics of high relevance in the ML4H community. Each roundtable will have a small group of senior researchers and practitioners who are experts in the selected topics and a few junior chairs who will lead the discussion on open problems in ML4H which were crowdsourced from the community.

How to Participate

We will host in-person roundtables that will take place on December 10th, 2023 during ML4H. The session will be split into two parts, with a 5 minutes break in the middle where participants can move to a different roundtable. Participants are encouraged to pick at most two roundtables that they would like to join and engage in the discussion points proposed by the junior chairs and ask further questions to the senior chairs. To avoid the disruption of conversations, the participants should only move between roundtables during the break.

In-Person Research Roundtables

1. Health AI Collaborations, Deployment, and Regulation

One of the barriers to deploying AI models in healthcare is the ability to safely and effectively integrate models into clinical workflows. What are different factors one should consider in presenting AI models to clinicians that result in effective clinician-AI collaborations, and how do we know if these models truly have a significant impact within the healthcare setting? What are the desires of caregivers and clinicians, and what aspects are still lacking? Furthermore, there has been an increased focus on AI regulation by policymakers and industry players in the last few years. How do we ensure all stakeholders are considered in AI policy, and who should be in charge of writing such regulation, if at all?

Senior Chairs: Jason Fries, Parisa Rashidi
Junior Chairs: Hussein Mozannar, Rahul Thapta

2. Integrating AI into clinical workflows

The rate of AI progress in the last few years seems to have major implications for the types of models we train for healthcare purposes. With the types of models we train constantly changing, how can we develop model-agnostic methods to integrate AI into clinical workflows?

Senior Chairs: Brett Beaulieau-Jones, Xuhai Orson Xu
Junior Chairs: William Jongwon Ha, Nikita Mehandru

3. Health AI Foundation models

Foundation models train on large amounts of data, and there might be benefits to combining the data from multiple sources (hospitals) and over-training separate foundation models for each hospital. What are the pros and cons of doing so?

Senior Chairs: Matthew McDermott, Tristan Naumann
Junior Chairs: Michael Wornow, Vlad Lialin

4. Large language models and healthcare

Generative AI has shown promise in multiple applications. How can the promising advances in generative AI be translated into healthcare applications, and what opportunities or challenges does this innovative technology present in this field? How can we create effective and ethical data sharing protocols in the healthcare sector that respect individual privacy, promote research, and enhance patient care? What are some low-hanging fruit opportunities to use large language models in healthcare?

Senior Chairs: Monica Agarwal
Junior Chairs: Xin Liu, Alejandro Lozano

5. Multimodal AI for Health

Multimodal AI shows promise over leveraging data modalities independently. How can we effectively integrate multiple data sources (e.g., Electronic Health Records (EHRs), images, genomics) for ML applications in healthcare? How does this work in real-time in a hospital?

Senior Chairs: Marinka Zitnik
Junior Chairs: Jiacheng Zhu, Rafal Dariusz Kocielnik

6. Health AI model development and generalizability

Applying ML models in real practice could face multiple challenges including domain shift, annotation quality, and out-of-distribution. How can we ensure the robustness and generalizability of a model?

Senior Chairs: Berk Ustun
Junior Chairs: Haoran Zhang, Keith Harrigian

7. Health AI and Accessibility

Making AI accessible to all in healthcare is important, but “accessibility” could encompass many things such as infrastructure, compute resources, or access to healthcare in the first place. What are the different components of the healthcare system that could improve patients’ accessibility to health AI, and how do these different components play into the development of AI models?

Senior Chairs: Edward Choi, Kristen Yeomi
Junior Chairs: Edward Lee

8. Health AI and patient privacy

How can we preserve patient privacy and maintain data security while leveraging machine learning techniques in healthcare?

Senior Chairs: Gamze Gürsoy
Junior Chairs: Milos Vukadinovic

9. Bias/Fairness in Health AI

Despite its potential, the application of machine learning in healthcare has often resulted in models that reflect and reinforce existing health disparities. How can machine learning promote fairness and enhance global health outcomes?

Senior Chairs: Marzyeh Ghassemi, Emma Pierson
Junior Chairs: Aparna Balagopalan, Sarah Jabbour

10. ML for Survival Analysis & Epidemiology/Population Health

Where do we stand with ML's role in population health? How can ML be applied for time-to-event survival analysis? How ML is aiding in preventing and responding to outbreaks of infectious diseases?

Senior Chairs: George Chen, Sanjat Kanjilal
Junior Chairs: Vincent Jeanselme

11. Causality in Health AI

How can recent advances in AI/ML help discover causal relations using clinical data? To what extent can we use observational data to emulate randomized trials, to evaluate the causal effect of any treatment?

Senior Chairs: Michael Oberst, Linying Zhang
Junior Chairs: Katherine Matton, Ilker Demirel