Machine Learning and Economics Group, Oxford

Overview

This webpage provides information for the informal Machine Learning (ML) and Economics group at the Department of Economics, University of Oxford, coordinated by Maximilian Kasy.

The goal of this group is to discuss research and develop a common research agenda at the intersection of ML and economics. The focus is on conceptual and methodological contributions of economics to ML and of ML to economics. These two fields share a common language in the frameworks of optimization, probability, and decision theory. Economics has much to contribute to ML with its considerations of multiple agents, inequality and conflicting interests, and private information. ML has much to contribute to economics with its insights on supervised and active learning, considerations of non-traditional data types, and adaptive decision-making. Special emphasis will be put on the social impact of ML, and on non-commercial applications of ML.

Activities

Upcoming

• May 20-22, 2024: Workshop Economic Analyses of Science.
• Bi-weekly reading and discussion group on ML and economics, meeting in even weeks of term.
• Youtube channel of workshops and discussions.

Previous

• In May 2023, we organized a workshop on Social foundations for statistics and machine learning, featuring tutorials and frontier talks, as well as a panel discussion. All talks are available on Youtube.
• In June 2022, we organized a small cross-departmental workshop on ML and economics, with presentations of ongoing work and discussants, as well as a keynote by Michael Jordan from UC Berkeley. Information on this workshop can be found at Machine Learning and Economics Jamboree 2022.
• I started teaching a new MPhil course on the foundations of ML for economists in HT 2022.
  All faculty, post-docs, and doctoral students are invited to audit this course and participate in discussions.
  Syllabus, slides and readers can be found here: Foundations of Machine Learning
• In April 2021, we organized a conference on Machine learning and economic inequality. Recordings of the talks are available on Youtube.

Reading and discussion group

Hilary term 2024

• Time: 2:30pm, Tuesdays in even weeks. (That is: 23 Jan,6 Feb, 20 Feb, and 5 Mar)
• Location: SR E, Manor Road Building in week 2, 6 and 8, and SR C, Manor Road Building in week 4 (There will be coffee and pastries!).
• For those who cannot make in person, please join via Zoom.

Topic: Applications of AI
Curation this term thanks to Friedrich Geiecke!

Tentative outline:

• W2: Reinforcement learning from human feedback / AI feedback
  Training language models to follow instructions with human feedback,
  Scaling Reinforcement Learning from Human Feedback with AI Feedback
  Presenter: Friedrich Geiecke.
• W4: Labor market impact of LLMs
  GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models
  Generative AI at Work
  Presenter: Jacob Greenspon.
• W6: Diffusion models (image generation)
  Understanding Diffusion Models: A Unified Perspective,
  Diffusion Models: A Comprehensive Survey of Methods and Applications
  Presenter: Maximilian Kasy. Slides
• W8: AI in science
  Learning skillful medium-range global weather forecasting
  Highly accurate protein structure prediction with AlphaFold

Michaelmas term 2023

• Time: 2:30pm, Tuesdays in even weeks. (That is: 17 Oct, 31 Oct, 14 Nov, 28 Nov)
• Location: SR A, Manor Road Building (There will be coffee and pastries!)
• For those who cannot make in person, please join via Zoom.

Topic: Large Language Models.

• W2: Foundations (Neural networks and natural language processing)
  Chapters 6,7, and 9 of Speech and Language Processing
  Presenter: James Dufy. Slides
• W4: Foundations (Transformers)
  Chapters 10 and 11 of Speech and Language Processing
  Presenter: Maximilian Kasy. Slides
• W6: Practical implementation
  Huggingface NLP course
  Presenter: Jeremy Large. Slides, Python notebook
• W8: The impact of large language models
  ChatGPT Is a Blurry JPEG of the Web
  The Debate Over Understanding in AI’s Large Language Models
  Presenter: Giulia Caprini

Hilary term 2023

Meeting dates:

• 2:30pm, Tuesdays in even weeks
  (That is: 24 Jan, 7 Feb, 21 Feb, 7 Mar)
  Manor Road Building, Seminar Room D

Topics:

This term we will be discussing several chapters of the book Prediction, learning, and games by Nicolò Cesa-Bianchi and Gábor Lugosi. This book builds on ideas in game theory and learning theory, and provides a comprehensive framework for online decision-making and adversarial learning.

We will cover the following chapters:

• W2: Chapter 2 (Prediction with expert advice)
• W4: Chapter 4 (Randomized prediction)
• W6: Chapter 6 (Prediction with limited feedback)
• W8: Chapter 7 (Prediction and playing games)

Michaelmas term 2022

No meetings (Max on sabbatical at MIT).

Trinity term 2022

Meeting dates:

• 2:30pm, Tuesdays in even weeks
  (That is: 3 May, 17 May, 31 May, 14 June)
  Manor Road Building, Seminar Room C

Topics:

• W2
  • Algorithmic Fairness
  • Sociological Perspectives on Racial Discrimination
  • Bonus material: Fairness, Equality, and Power in Algorithmic Decision-Making
• W4
  • Multitask principal-agent analyses
  • Human Compatible
• W6: (Martin Weidner guest edition)
  • Matrix Completion Methods for Causal Panel Data Models
  • Reevaluating COVID-19 Mandates using Tensor Completion

Hilary term 2022

• Slides for the coordinating meeting: Slides

Topics:

• W2:
  • Introduction of Understanding machine learning: From theory to algorithms
  • Roles for Computing in Social Change
  • Discussion questions, Slides Michael, Slides Binta
• W4:
  • Learning Equilibria in Matching Markets from Bandit Feedback
  • Digital Socialism?
  • Discussion questions
• W6
  • The Limits of Price Discrimination
  • AI’s Future Doesn’t Have to Be Dystopian
  • Discussion questions, Slides Jeremy, Slides Ben, Slides Samira
• W8
  • Chapters 1-3 of The Algorithmic Foundations of Differential Privacy
  • Democratic Data: A Relational Theory For Data Governance
  • Discussion questions