Hazel Kim

I am a DPhil student in computer science at the University of Oxford, conducting research in Natural Language Processing and Machine Learning. I am grateful to be advised by Philip Torr and Yarin Gal from Oxford and Hinrich Schütze from LMU Munich as a student of the European Laboratory for Learning and Intelligent Systems (ELLIS Society).

Much of the direction I've taken was shaped by my early interaction with Jae C. Choe and his advisor E.O. Wilson's book "Consilience: The Unity of Knowledge". I cannot express enough gratitude that Jae took my curiosity seriously at such a young age.

Their encouragement opened more doors than I could have expected. My interests have wandered widely, from preparing for mathematics olympiads (AIME 10 & 12, and IMO), to a growing curiosity about law while studying Latin and Roman law, to human rights work with Amnesty International's youth events. My early interest in biodiversity and climate change owes a great deal to Jae's close collaborator Jane Goodall, whose stories deepened my empathy for animals and their behavior. Through my undergraduate years, alongside my computer science studies, which included competing in the ICPC programming contest (North America Division, 2018), I pursued political science (minor), philosophy (first-, second-, and senior-year seminars), economics (advanced macroeconomics), and biology (evolution and gentics). I was also fortunate to study violin with conductor and violinist Andrew Koehler; one of our most memorable performances was a concert-requiem at Harris Theater Chicago commemorating the Holodomor, the Ukrainian famine-genocide of the 1930s. Music tells stories that are not far from subjects we study today; math, emotions, human beings, history, philosophy, politics, etc.

These experiences now converge in my research on human cognition-inspired artificial intelligence. My starting point has always been language, out of a belief that reasoning grows from language acquisition; words shape our thoughts. That conviction has guided me since I began working in deep learning in 2019, and still crafts the questions I ask today.

Research Areas:
Interpretable and Controllable Language Models

Language Acquisition from Limited Sources

Language models shape the world with given data. They work well with quality and abundant data while providing biased perspectives with scarce, poor data. However, quality data is often expensive to collect or private to get access. This challenge has inspired me to study how to encourage language models to acquire language proficiency and reasoning ability from a limited quantity or quality of resources.

Causal Inference, not Correlation

The heavy reliance on correlations between input observation and output predictions limits language models to reason about cause-and-effect relationships that are not explicitly present in the text. High correlations do not always uncover the causality of events. Causal inference is a crucial topic to control language models against biased data.

Information Quantification

Computation creates valuable information. The usable information for models varies depending on their interaction between input and output resources. Exploring how to quantify information is essential for interpreting model behaviors.

Emergent Knowledge of Language Models

Language models emerge new knowledge that is not present in smaller models when the scale grows. The emergent knowledge is a valuable yet underexplored resource. This has motivated me to investigate them regarding the model capabilities of how trustworthy they are for transparent usage in real-world scenarios.

Publications

• Selected and Recent Papers

Google Scholar

Resume

• University of Oxford 2023 - now

  DPhil Candidate in Computer Science
  ELLIS PhD Student

  • Admitted in 2022 (Recruited by Yarin Gal)
  • Deferred enrollment to 2023
• CLASSTING AI RESEARCH 2022 - 2023

  NLP/ML Research Lead

  • Rationales and Reasoning with Large Language Models (COLING25)
• NAVER AI LAB 2021 - 2022

  NLP/ML Research Intern
  (Foundation Model Commonsense Reasoning)

  • Hosted by Seong Joon Oh
  • Collaborated with Seong Joon Oh, Sangdoo Yoon, and Kangmin Yoo
• Yonsei University 2020 - 2022

  Master in Artificial Intelligence

  • Advisor: Yo-Sub Han
  • Learning with Limited Data Using Compositionality in Language (thesis, AAAI22)
  • Mathematical reasoning (EMNLP23)
  • Legal reasoning (co-worked with Yulchon LLC.)
• Kalamazoo College 2014 - 2019

  B.A. in Computer Science and Political Science
  The Index (Staff Writer)/ Ultimate Frisbee Team

  • Founder's Scholar
  • Advisors: Alyce Brady (CS), John Dugas (PS)
  • The International Collegiate Programming Contest (ICPC) North America Division Championship 2018
• Seoul National University 2012 - 2014

  Graduate School of Public Health

  • Research Team Founder and Leader
  • Collaboration with Intergovernmental Panel on Climate Change (IPCC)
• Hankuk Academy of Foreign Studies 2011 - 2014

  High School - International Track
  Latin Honor Society
  HAFS Lacrosse/ Basketball/ Archery(Team Captain - Seoul)

  • American Invitational Mathematics Examination (AIME) 10 & 12
  • International Mathematical Olympiad (IMO) Selected Training Team
  • AP Calculus AB/BC (5)
  • AP Statistics (5)
  • AP Macro/Microeconomics (5)
  • AP Chemistry (4)
  • AP Biology (5)
  • AP English Literature and Composition (5)
  • AP World History (5)
  • AP Psychology (5)
  • AP US Government and Politics (4)
  • SAT Latin