Andrea Wynn

I am a second-year PhD student in Computer Science at the Johns Hopkins University Whiting School of Engineering. I am advised by Dr. Eric Nalisnick and Dr. Anqi Liu, and I work closely with Dr. Gillian Hadfield. My research interests span AI safety and robustness, AI alignment, and human-AI collaboration. My work brings together ideas from machine learning and cognitive science to create more reliable, trustworthy, and human-centered AI systems.

Before that, I was a MSE student and teaching assistant in the Computer Science department at Princeton University, working with Dr. Tom Griffiths in the Princeton Computational Cognitive Science Lab.

And before that, I received a B.S. with a double major in Computer Science and Mathematics from Rose-Hulman Institute of Technology in May 2022.

I worked with LinkedIn in summer 2025 as a PhD AI Research Intern, and at Expedia Group in summer 2023 as a Machine Learning Science Intern. I have previously completed software engineering internships at Amazon Web Services, Kratos Defense & Security Solutions, Collins Aerospace, and Circa.

Email  /  CV  /  Resume  /  LinkedIn  /  Google Scholar  /  Github  /  OrcID

September 2025: Our poster, Safe and Efficient In-Context Learning via Risk Control, was accepted to the JHU DSAI Fall 2025 AI Symposium!

August 2025: I am honored to be selected for associate membership in Sigma Xi, the Scientific Research Honor Society.

June 2025: Our workshop paper, Talk Isn't Always Cheap: Understanding Failure Modes in Multi-Agent Debate, was accepted to the 2025 ICML MAS workshop in Vancouver!

May 2025: I will continue my AI safety work during an internship as a PhD Trust & Responsible AI intern at LinkedIn!

February 2025: Our work, Learning Human-like Representations to Enable Learning Human Values, has been selected as one of only four student talks at the 2025 DSAI Human Alignment of AI Symposium!

January 2025: I was named a Junior Member of the Future of Life Institute's AI Existential Safety community for my work on human-centric AI!

October 2024: I received the Jun Wu and Yan Zhang Graduate Student Fellowship, as well as the Louis M. Brown Engineering Fellowship, from Johns Hopkins University.

September 2024: I received the two-year Percy Pierre Doctoral Fellowship to fund my work!

September 2024: Our paper, Learning Human-like Representations to Enable Learning Human Values, has been accepted to NeurIPS 2024!

Preprint.

Large language models (LLMs) demonstrate a remarkable ability to learn new tasks from a few in-context examples. However, this flexibility introduces safety concerns: LLMs can be influenced by incorrect or malicious demonstrations. This motivates principled designs in which the system itself includes built-in mechanisms to guard against such attacks. We propose a novel approach to limit the degree to which harmful demonstrations can degrade model performance. We present both theoretical and empirical results showing that our approach can effectively control this risk for harmful in-context demonstrations, while simultaneously achieving substantial performance and efficiency gains with helpful demonstrations.

Accepted to ICML 2025 Multi-Agent Systems Workshop.

Multi-agent debate, often proposed to improve AI reasoning, can sometimes harm performance. Prior work has only studied homogeneous agents, but this work examines how diverse tasks and model capabilities affect debate dynamics. Experiments show that accuracy may decrease over time, even when stronger models outnumber weaker ones, as agents tend to adopt peers' incorrect reasoning for the sake of agreement. This reveals key failure modes of multi-agent debate, suggesting that naive applications of debate risk degrading performance when agents cannot resist persuasive but flawed arguments.

Accepted to NeurIPS 2024.

We study a previously overlooked factor that influences a ML agent's ability to learn human values (representational alignment), drawing on insights from cognitive science. We demonstrate that aligning AI representations with humans can improve safety, sample efficiency, and generalization ability when learning a wide range of human values in personalization tasks. This introduces a new avenue for pursuing scalable, robust, and personalizable alignment of AI agents with human values.

I find and prove the existence of Dirac conical points in multiple 2D materials under certain conditions on electric potential, a property which has been conjectured to be related to the unique properties of graphene. I additionally discovered and proved the existence of a new type of spectral touching point, which I named the mesa touching point.

LinkedIn Core AI Team, PhD Research Intern, May - August 2025.

Expedia Group Vacation Rental Dynamic Pricing Team, Machine Learning Science Intern, May - July 2023.

Amazon Web Services Pool Balancing & Demand Forecasting Team, Software Development Engineer Intern, June - August 2022.

Circa, Research Intern, November 2021 - May 2022.

Amazon Web Services Outposts Team, Software Development Engineer Intern, June - August 2021.

Impact Snacks, Software Engineering Intern, March - May 2021.

Kratos Defense & Security Solutions, Software Engineering Intern, June - August 2020.

Collins Aerospace, Software Engineering Intern, June - August 2019.

Backpat Tutoring, Co-Founder & Volunteer Coordination Manager, May 2020 - May 2022.

Society of Asian Scientists & Engineers, Midwest Regional Coordinator, June 2022 - Present.

Society of Asian Scientists & Engineers, Midwest Regional Conference Chair, August 2021 - March 2022.

Society of Asian Scientists & Engineers, Rose-Hulman Chapter President, February 2020 - May 2022.

Society of Women Engineers, Rose-Hulman Chapter Secretary, March 2019 - March 2021.

Society of Asian Scientists & Engineers, Rose-Hulman Chapter Secretary, February 2019 - February 2020.

Society of Asian Scientists & Engineers, Rose-Hulman Chapter Outreach Chair, October 2018 - February 2019.

COS126 Computer Science: An Interdisciplinary Approach, Princeton University, Fall 2022 - Spring 2024.

CSSE333 Database Systems, Rose-Hulman Institute of Technology, Spring Quarter 2021.

CSSE333 Database Systems, Rose-Hulman Institute of Technology, Winter Quarter 2020-21.

CSSE304 Programming Language Concepts, Rose-Hulman Institute of Technology, Fall Quarter 2020.

CSSE333 Database Systems, Rose-Hulman Institute of Technology, Spring Quarter 2020.

CSSE333 Database Systems, Rose-Hulman Institute of Technology, Winter Quarter 2019-20.

CSSE333 Database Systems, Rose-Hulman Institute of Technology, Fall Quarter 2019.

CSSE230 Data Structures & Algorithm Analysis, Rose-Hulman Institute of Technology, Fall Quarter 2019.

Lead Science Educator in Training, Seattle Pacific Science Center, Summers 2017-2018.

Volunteer English Teachers in China, Little Masters Club, Ankang, China, Summer 2017.

Science Educator in Training, Seattle Pacific Science Center, Summers 2014-2016.

Sigma Xi Scientific Honor Society Associate Membership, 2025-26.

Percy Pierre Doctoral Fellowship, 2024-26.

Louis M. Brown Engineering Fellowship, 2024-25.

Jun Wu and Yan Zhang Graduate Student Fellowship, 2024-25.

Herman A. Moench Distinguished Senior Commendation, Rose-Hulman Institute of Technology, May 2022.

Frank Young Outstanding Service Award, Rose-Hulman Institute of Technology, May 2022.

Summa Cum Laude Graduate, Rose-Hulman Institute of Technology, May 2022.

Center for Diversity Student Ambassador Award, Rose-Hulman Institute of Technology, May 2022.

Rose-Hulman Independent Research Project Grant Recipient, Rose-Hulman Institute of Technology, February 2022.

Society of Women Engineers Conference Scholarship (sponsored by BorgWarner), October 2021.

2021 Chevron Scholar & Scholarship Recipient, Society of Asian Scientists and Engineers, September 2021.

Henry Turner Eddy Award for Applied Mathematics, Rose-Hulman Institute of Technology, May 2021.

TechPointX SOS Tech Challenge First Place Winner, August 2020.

Student Leader of the Quarter, Rose-Hulman Institute of Technology, Spring 2020.

Diversity Connect Engineering Design Challenge First Place Winner, Rose-Hulman Institute of Technology, October 2018.

Rose-Hulman Dean's List, 12x (All Quarters Attending), August 2018 - May 2022.