# Michael Yao - Homepage ## About Michael Yao is an MD-PhD candidate at the University of Pennsylvania leveraging AI to improve human health. [CV](https://www.michaelsyao.com/public/cv.pdf) Currently: Medical student, Research scientist I am an MD-PhD candidate at the [University of Pennsylvania](https://ai.upenn.edu/) and Research Scientist at [Abridge](https://www.abridge.com). My research focuses on **trustworthy machine learning** and how we can reliably use clinical ML systems under distribution shift. I am interested in how to leverage [prior knowledge](https://arxiv.org/abs/2509.20975) and [statistical methods](https://proceedings.mlr.press/v267/yao25b.html) to help algorithms better [generalize to new distributions](https://openreview.net/forum?id=STrpbhrvt3), and how we can use these methods to improve **clinical decision support** in [rare disease diagnostics](https://www.nature.com/articles/s41467-025-58801-7) and [hospital workflows](https://www.nature.com/articles/s43856-025-01061-9). I also seek to better understand the [types of distribution shift](https://doi.org/10.1093/jamiaopen/ooag173) observed in clinical practice, and am also working to improve physician AI literacy in [medical education](https://mededu.jmir.org/2025/1/e63602). I was advised by [Osbert Bastani](https://trustml.github.io) and [James Gee](https://picsl.upenn.edu) during my PhD, and am grateful to be supported by an [NIH F30 NRSA Fellowship](https://reporter.nih.gov/project-details/11194234) from the National Institute on Minority Health and Health Disparities (NIMHD). - 2026 | Research Scientist at [Abridge](https://www.abridge.com) - 2025 | Received my [PhD in Bioengineering](https://www.proquest.com/docview/3298652850) and MS in Computer Science from the University of Pennsylvania - 2025 | ML Scientist Intern at [Genentech](https://www.gene.com/scientists/our-scientists/braid) Generative AI - 2025 | Human Frontier Collective Intern at [Scale AI](https://hfc.scale.com) - 2023 | AI Clinical Fellow at [Glass Health](https://glass.health) - 2022 | Research Scientist Intern at [Microsoft Research](https://www.microsoft.com/en-us/research/) Health Futures - 2021 | Software Engineering Intern at [Hyperfine Research](https://hyperfine.io) - 2021 | Graduated salutatorian from Caltech, BS Applied Physics ## Publications - 2026 | Can language models help us personalize treatment strategies for patients? Learn more about how we can use LLMs for precision medicine in our new [paper](https://openreview.net/forum?id=w025bYRVkO) accepted to ICLR 2026! - 2025 | Can generative language models like ChatGPT help clinicians order diagnostic imaging studies in the ED? Check out our new [paper](https://www.nature.com/articles/s43856-025-01061-9) in Communications Medicine to learn more! [Penn press release](https://www.linkedin.com/posts/pennengai_using-ai-to-support-smarter-imaging-decisions-activity-7425539042240360448-9wdP) [Aunt Minnie article](https://www.auntminnie.com/imaging-informatics/artificial-intelligence/article/15752498/generative-ai-improves-clinical-decisionmaking-in-the-ed) - 2025 | How can we ensure that offline optimization methods propose both high-quality *and* diverse sets of designs? Learn more about our method DynAMO in our new [paper](http://arxiv.org/abs/2501.18768) accepted to ICML 2025! - 2025 | Excited to share our work in Nature Communications on [multimodal concept bottleneck models](https://www.nature.com/articles/s41467-025-58801-7) for interpretable eye cancer diagnostics, led by the incredible [Yifan Wu](https://yifannnwu.github.io)! - 2025 | Can datathons be effective venues for teaching AI to med students? Check out our work on [trainee-led datathons](https://mededu.jmir.org/2025/1/e63602) now out in JMIR Medical Education! - 2024 | Can we reliably optimize against surrogate objectives in offline optimization problems? Learn more about our method for [Generative Adversarial Model-Based Optimization (GAMBO)](https://arxiv.org/abs/2402.06532) accepted to NeurIPS 2024. Check out our work in Vancouver! - 2024 | Grateful to have contributed to our NeurIPS Spotlight work on [Knowledge Bottlenecks](https://yueyang1996.github.io/knobo/) for improved interpretability and robustness of ML for healthcare, led by the fabulous [Yue Yang](https://yueyang1996.github.io/)! [Penn press release](https://blog.seas.upenn.edu/training-medical-ai-with-knowledge-not-shortcuts/) ## Contact - [LinkedIn](https://www.linkedin.com/in/michael-s-yao/) - [GitHub](https://github.com/michael-s-yao) - [X (Twitter)](https://twitter.com/michael_s_yao) - [Email [no spam]](hello [at] michaelsyao [dot] com) - [Google Scholar](https://scholar.google.com/citations?user=jz9IC2QAAAAJ&hl=en) ## Teaching - 2026 | **Course Author**: [Ethical Algorithms for the Modern Clinician](https://www.med.upenn.edu/eamc) - 2026 | **Head TA**: Health, Healthcare and Technology - 2025 | **TA**: Distributed Systems - 2024 | **TA**: Principles of Deep Learning - 2024 | **TA**: Imaging Informatics - 2024 | **TA**: Diagnostic Ultrasound for Medical Students - 2024 | **TA**: Clinical Reasoning for Medical Students ## Outreach I designed and run a short course on the [fundamentals of ML for medical students](https://www.med.upenn.edu/eamc). I currently serve as a president of the [Penn HealthX](https://www.pennhealthx.com/) student group, and have previously served as Vice Chair of the Technology Committee for the [American Physician Scientists Association (APSA)](https://www.physicianscientists.org) and as Director of Data Science and AI for [MDplus](http://ai.mdplus.community). At Penn, I am involved in a number of mentorship and outreach initiatives and have served on both the Admissions Committee and AI Curriculum Steering Committee for the School of Medicine. I am actively involved in Penn's interview and recruitment process for medical school admissions. I set aside half an hour a week to meet with current students for pro bono feedback on applications and general college advice, especially for underrepresented students from minority backgrounds. I also enjoy working with and mentoring students on interesting research projects. If you're interested in connecting, please reach out to me via email. In my free time, I enjoy solving [fun math problems](https://github.com/michael-s-yao/ProjectEuler) and [festive programming puzzles](https://github.com/michael-s-yao/AdventOfCode2025) with code, and contributing to open-source software. I also enjoy solving puzzles like the New York Times Games, and also building their corresponding medical knockoffs (check out [Clerkship Connections](https://www.clerkshipconnections.com) and [Pimping Rounds](https://www.pimpingrounds.com)). Finally, I am a (mediocre) chess hobbyist and occasionally [write on my blog](https://www.michaelsyao.com/substack).