David Abel
Welcome!
I am a Senior Research Scientist at DeepMind on the Agency team led by Will Dabney, and an Honorary Fellow at the University of Edinburgh where I work closely with the Autonomous Agents Research Group.
I will be co-advising a small number of Ph.D students at the University of Edinburgh: for more info, see here
If you are interested in working together at Google DeepMind, see open roles here. Unfortunately, I do not have any current openings for direct reports or interns.
News
- 2025: Workshop chair for RLDM 2025.
- 2025: Associate program chair for CoLLAs 2025.
- Dec. 2024: Talk at Penn State.
- Nov. 2024: Guest Lecture at U. of Alberta.
- Nov. 2024: Visit and talk at UCL's Gatsby.
- Nov. 2024: Talk at "Industry insights" event at UoE.
- Oct. 2024: Talk at Purdue.
- Oct. 2024: Talk at RTX.
- Sep. 2024: I started as an Honorary Fellow at the University of Edinburgh.
- Sep. 2024: Attending the Dagstuhl Seminar on Explainable AI.
- Aug. 2024: Co-organized a workshop at RLC.
- Aug. 2024: On a panel at RLC and gave a talk at the RL safety workshop on Agency.
- Jul. 2024: Attended ICML.
- Jul. 2024: Gave a talk at the RSS workshop on task specification.
- Jun. 2024: Gave a (virtual) talk at Oregon State
- May 2024: Visited Montreal and gave a talk at Mila.
- May 2024: Three Dogmas of Reinforcement Learning accepted to RLC.
- May 2024: New paper, led by Andi Peng, accepted to ICML.
- Apr. 2024: Gave a talk at the UoE RL reading group.
- Apr. 2024: Visited the Autonomous Agents group at the University of Edinburgh.
- Mar. 2024: I moved to Edinburgh, Scotland.
- Feb. 2024: Podcast with Ather.
- Dec. 2023: Attended NeurIPS.
- Oct. 2023: Recording of Three Dogmas of Reinforcement Learning available.
- Oct. 2023: Podcast with Ron.
- Sep. 2023: A Definition of Continual Reinforcement Learning accepted to NeurIPS.
- Jul. 2023: Attended ICML and gave a talk at the interactive learning workshop.
- Jun. 2023: Attended a workshop in California.
- Apr. 2023: Settling the Reward Hypothesis accepted to ICML
- Apr. 2023: Guest lecture in Prof. Ben Van Roy's course at Stanford.
Interests
My research focuses on bringing clarity to the central philosophical questions surrounding agency, computation, and learning.
I value research that provides new understanding, and tend to get excited by simple but foundational questions. I typically work with the reinforcement learning problem, drawing on tools and perspectives from across philosophy, math, and computer science.
I am currently interested in better defining the main concepts of AI, such as learning, agency, and goals. Previously, my dissertation studied how agents model the worlds they inhabit, focusing on the representational practices that underly effective learning and planning.
Featured Research
|
We reflect on the paradigm of RL and suggest three departures from our current thinking.
Joint with Mark Ho and Anna Harutyunyan.
|
|
A Definition of Continual Reinforcement Learning
NeurIPS 2023
We present a precise definition of the continual reinforcement learning problem.
|
|
Settling the Reward Hypothesis
ICML 2023
We illustrate the implicit requirements on goals and purposes under which the reward hypothesis holds.
|
|
We develop a new theory describing how people simplify and represent problems when planning.
Led by Mark K. Ho, joint with Carlos G. Correa, Jonathan D. Cohen, Michael L. Littman, Thomas L. Griffiths.
|
|
On the Expressivity of Markov Reward
NeurIPS 2021 (Outstanding Paper Award)
We study the expressivity of Markov reward functions in finite environments by analysing what kinds of tasks such functions can express.
Joint work with Will Dabney, Anna Harutyunyan, Mark K. Ho, Michael L. Littman, Doina Precup, Satinder Singh.
|
|
A Theory of Abstraction in Reinforcement Learning
Ph.D Thesis, 2020
My dissertation, aimed at understanding abstraction and its role in effective reinforcement learning.
Advised by Michael L. Littman.
|
|
Value Preserving State-Action Abstractions
AISTATS 2020
We prove which combinations of state abstractions and options are guaranteed to preserve representation of near-optimal policies in any finite Markov Decision Process.
|
|
The Value of Abstraction
Current Opinions in Behavioral Science 2019
We discuss the vital role that abstraction plays in efficient decision making.
|
|
We prove that the problem of finding options that minimize planning time is NP-Hard.
|
Active Service
- JMLR Editorial Board (2020-)
- NeurIPS Area Chair (2023-)
- RLC Senior PC Member (2024-)
- RLDM Workshop Chair (2025)
- CoLLAs Senior PC Member (2022-)
- Recent Reviewing: AAAI, ACM, AISTATS, ICML, ICLR, JMLR, Nature, NeurIPS, OpenMind, TMLR, RLC.
Selected Awards
- Outstanding Paper Award, NeurIPS 2021, On the Expressivity of Markov Reward
- Presidential Award for Excellence in Teaching, Brown University
- Runner-Up, 2020 AAAI/ACM SIGAI Dissertation Award
- 7x Top Reviewer: ICML (x4); NeurIPS (x2), AISTATS (x1).
About Me
Before joining DeepMind, I completed my Ph.D in Computer Science at Brown University where I was fortunate to be advised by Prof. Michael Littman. I got my start in research working with Prof. Stefanie Tellex at Brown, and before that studied Philosophy and Computer Science at Carleton College.
I'm a big fan of basketball, lifting, baking, reading, games, snowboarding, and music (I play guitar/piano/violin and love listening to just about everything). I live in Edinburgh, Scotland with my wife Elizabeth and our dog Barley.
Always up for a chat -- please reach out over email if you'd like to discuss anything. If you would like to arrange a call, I have a recurring open slot in my calendar here.
If you have feedback of any kind, please feel free to fill out this anonymous feedback form.