Luise Ge EmailCV

I am a Computer Science PhD candidate at Washington University in St. Louis, coadvised by Prof. Yevgeniy Vorobeychik and Prof. Gregory Kehne. My current research interests cover the foundations of AI (e.g., integrating learning and reasoning, and automated discovery) and collective decision making (e.g., AI alignment, and computational social choice).

I started out as a cognitive science student at the University of Edinburgh, with the passion to figure out how the mind works. While this curiosity has never diminished, it just happened that the beauty of maths struck me greatly. I ended up getting an MSc in pure maths at Imperial College London under my amazing advisor Prof. Paolo Cascini. My master's thesis was on algebraic geometry, inspired by a fantastic idea called "the periodic table of shapes". I have turned back to CS due to both its philosophical considerations and societal impact. Outside of research, you may find me reading, writing, watching films and plays, making art, or traveling.

Research

C = conference, W = workshop, T = thesis. (α-β) denotes alphabetical author order.

Publications

• C9
  Q: If each voter answers only one or two comparison questions, can we still learn enough about the whole population to make good collective decisions?

  Linear Social Choice with Few Queries: A Moment-Based Approach

  Luise Ge, Daniel Halpern, Gregory Kehne, Yevgeniy Vorobeychik

  SODA 2027 (previously presented at the 2026 ESIF Economics and AI+ML Meeting)

  [arXiv] [Slides]
• C8
  Q: Instead of learning a full model, can we learn from partially observed data just what a given query needs, and still perform lifted reasoning efficiently?

  Query-Driven Learning for Lifted Reasoning

  (α-β) Luise Ge, Brendan Juba

  IJCLR 2026 (International Joint Conference on Learning & Reasoning)

  [Slides]
• C7
  Q: Can we efficiently perform lifted inference under uncertainty, reasoning about many objects at once, from incomplete knowledge and partially observed data, without ever building an explicit model?

  Implicit Learning for Reasoning in First-Order Probabilistic Logic

  (α-β) Luise Ge, Brendan Juba, Kris Nilsson, Alison Shao

  UAI 2026

  [Paper] [arXiv]
• C6
  Q: In federated learning, when clients use different models and have different data, how can they still benefit from each other while sharing only their predictions?

  COSMOS: Model-Agnostic Personalized Federated Learning with Clustered Server Models and Pseudo-Label-Only Communication

  Ben Rachmut, Luise Ge, William Yeoh, Ning Zhang, Yevgeniy Vorobeychik

  ECML PKDD 2026 ✦ Best Research Paper Nominee ✦

  [arXiv] [Slides]
• C5
  Q: When LLMs make decisions under risk, do they act like rational agents or like humans, and why do reasoning and conversational models behave so differently?

  Mind the (DH) Gap! A Contrast in Risky Choices Between Reasoning and Conversational LLMs

  Luise Ge, Yongyan Zhang, Yevgeniy Vorobeychik

  ACL 2026 ✦ Outstanding Paper Award ✦

  [arXiv] [Slides]
• C4
  Q: When voters and candidates are described by features, how much welfare do we lose by choosing a winner from rankings alone, and how can we lose as little as possible?

  Optimized Distortion in Linear Social Choice

  Luise Ge, Gregory Kehne, Yevgeniy Vorobeychik

  AAAI 2026 ✦ Oral ✦

  [arXiv] [Slides]
• C3
  Q: Can we reason efficiently under uncertainty about objects and their relations, even when we don't know how many objects there are, or there are infinitely many?

  Polynomial-Time Relational Probabilistic Inference in Open Universes

  (α-β) Luise Ge, Brendan Juba, Kris Nilsson

  IJCAI 2025

  [arXiv]
• C2
  Q: When tasks are too diverse for a single policy, can a small committee of policies ensure that most tasks, even unseen ones, have a near-optimal option?

  Learning Policy Committees for Effective Personalization in MDPs with Diverse Tasks

  Luise Ge, Michael Lanier, Anindya Sarkar, Bengisu Guresti, Yevgeniy Vorobeychik, Chongjie Zhang

  ICML 2025

  [arXiv] [Slides]
• C1
  Q: When we learn one reward model from many people's feedback, does the standard approach respect basic principles of collective decision making, and if not, what does?

  Axioms for AI Alignment from Human Feedback

  (α-β) Luise Ge, Daniel Halpern, Evi Micha, Ariel D. Procaccia, Itai Shapira, Yevgeniy Vorobeychik, Junlin Wu

  NeurIPS 2024 ✦ Spotlight ✦

  [PDF]
• W1
  Q: When can we learn a person's preferences from pairwise comparisons, either to predict their choices or recover their true utilities, and how much does choosing the questions help?

  Learning Linear Utility Functions From Pairwise Comparison Queries

  Luise Ge, Brendan Juba, Yevgeniy Vorobeychik

  AAMAS 2024 Workshop / ADT 2024 (Abstract)

  [arXiv] [Slides]
• T1
  Q: Can a market-based method for dividing chores fairly work in practice, and how does it hold up in running time, privacy, and incentives to misreport?

  Fair Allocation: Implementing and Evaluating an Algorithm for Competitive Allocation of Chores

  Luise Ge, supervised by Kousha Etessami

  Undergraduate Thesis, University of Edinburgh, 2022 ✦ Outstanding Thesis ✦

  [PDF]

Ongoing Projects

• Q: How can the theory and practice of pluralistic alignment inform each other?

  Pluralistic Alignment: A Survey Bridging Theory and Practice

  with Elinor Pool-Dayan

Misc