Doina Precup has spent over two decades making reinforcement learning practical. Her foundational work on temporal abstraction shapes how AI agents learn complex behaviors, and her leadership of DeepMind Montreal has made Canada a global center for RL research.
This is the complete guide to Doina Precup—her journey from Romania to Montreal, the options framework that transformed hierarchical RL, and her mission to bring more women into AI.
Who Is Doina Precup?
Doina Precup is a Romanian-Canadian computer scientist specializing in reinforcement learning, machine learning, and artificial intelligence. She holds joint positions as a professor at McGill University and Research Director at Google DeepMind, where she co-founded and leads the Montreal research team.
Quick Facts
Precup is one of the most cited reinforcement learning researchers in the world. Her work on options and temporal abstraction has been foundational to hierarchical RL, enabling agents to learn and plan at multiple levels of complexity.
What Is Doina Precup's Background?
Precup's path from Romania to leading one of the world's top AI research labs shows how academic excellence and strategic positioning can create global impact.
Education and Early Career
At UMass Amherst, Precup worked with Richard Sutton and Andrew Barto—two of the founders of modern reinforcement learning. Her doctoral thesis on temporal abstraction laid the groundwork for her most influential contributions.
The Move to Montreal
Precup joined McGill's School of Computer Science in 2000, just as Montreal's AI ecosystem was beginning to form. She became a core member of what would become Mila, working alongside Yoshua Bengio and other Montreal AI pioneers.
"I only became aware of the gender imbalance in sciences and technology when I moved to North America. In Romania, the science program in my high school was well attended by girls." — Doina Precup, on her experience moving to Canada
What Is the Options Framework?
Precup's most influential contribution is the options framework, developed with Richard Sutton and Satinder Singh. Published in 1999, this work fundamentally changed how AI researchers think about learning complex behaviors.
The Problem Options Solve
Traditional reinforcement learning operates on primitive actions—individual steps like "move left" or "press button." But real tasks require sequences of actions: "navigate to the kitchen," "pick up the cup," "make coffee."
The options framework introduces temporally extended actions—reusable skills that can span multiple time steps.
How Options Work
Options bridge the gap between Markov Decision Processes (MDPs), which use primitive actions, and Semi-MDPs, which abstract away timing. This enables agents to learn and plan at multiple levels simultaneously.
Impact of the Options Framework
The 1999 paper "Between MDPs and Semi-MDPs: A Framework for Temporal Abstraction in Reinforcement Learning" has been cited thousands of times and remains foundational to:
- Hierarchical reinforcement learning
- Skill discovery and reuse
- Transfer learning in RL
- Robot learning and control
"Learning, planning, and representing knowledge at multiple levels of temporal abstraction are key, longstanding challenges for AI." — Sutton, Precup, and Singh, opening of the options paper
What Is DeepMind Montreal?
In 2017, DeepMind announced it was opening a research lab in Montreal, with Doina Precup as founding director. This was DeepMind's second lab in Canada, following Edmonton (led by Rich Sutton).
Why Montreal?
DeepMind chose Montreal for several reasons:
DeepMind Montreal's Focus
The Montreal lab focuses on fundamental reinforcement learning research, including:
- Hierarchical RL and temporal abstraction
- Deep reinforcement learning
- Continual learning
- RL theory and algorithms
Precup maintains her McGill professorship while leading DeepMind Montreal, bridging academic research and industrial application.
Impact on Montreal's AI Ecosystem
DeepMind Montreal, alongside Mila, Google Brain Montreal, Meta AI Montreal, and Microsoft Research Montreal, has made the city one of the world's top destinations for AI research talent.
What Are Precup's Key Research Contributions?
Beyond the options framework, Precup has made foundational contributions across reinforcement learning.
Major Research Areas
Notable Papers
- "Between MDPs and Semi-MDPs" (1999) — The foundational options paper
- "The Option-Critic Architecture" (2017) — Learning options end-to-end
- "Eligibility Traces" — Core RL algorithms with Sutton
- Healthcare applications — RL for treatment optimization
Citation Impact
With over 45,000 citations on Google Scholar, Precup is among the most-cited reinforcement learning researchers globally. Semantic Scholar identifies 2,560 highly influential citations across her 400+ publications.
What Is the AI4Good Lab?
In 2017, Precup co-founded the AI4Good Lab with Angelique Mannella to address the gender disparity in AI and machine learning.
Program Structure
The program is now part of the Pan-Canadian AI Strategy and is hosted by Mila, with CIFAR and the OSMO Foundation as founding partners.
Why Precup Founded AI4Good
"I also want to do my part to ensure that AI is developed in an equitable and inclusive way. For this purpose, I have co-founded the AI4Good lab, which aims to improve the participation of women in machine learning." — Doina Precup
Precup noticed the gender imbalance only after moving to North America—in Romania, her high school science programs had strong female participation. AI4Good aims to change the culture in tech through mentorship and education.
What Positions Does Doina Precup Hold?
Precup maintains multiple leadership roles across academia, industry, and the broader AI community.
Current Positions
Conference Leadership
Precup has served in leadership roles at major AI conferences:
- ICML (International Conference on Machine Learning)
- NeurIPS (Neural Information Processing Systems)
- AAAI (Association for the Advancement of AI)
How Does Precup Fit in Montreal's AI Ecosystem?
Precup is central to Montreal's position as a global AI research hub.
Key Relationships
Montreal's RL Strength
Montreal has unusual depth in reinforcement learning research:
- Mila — World's largest academic AI institute
- DeepMind Montreal — Led by Precup
- Meta AI Montreal — Founded by Joelle Pineau
- McGill Reasoning and Learning Lab — Co-directed by Precup
This concentration makes Montreal arguably the world's top city for RL research outside the UK.
Frequently Asked Questions
Who is Doina Precup?
Doina Precup is a Romanian-Canadian computer scientist and one of the world's leading reinforcement learning researchers. She is a professor at McGill University, Research Director at Google DeepMind Montreal, and co-founder of the AI4Good Lab. Her work on the options framework for temporal abstraction has been foundational to hierarchical reinforcement learning.
What is Doina Precup known for?
Precup is best known for co-inventing the options framework with Richard Sutton and Satinder Singh, which enables AI agents to learn and plan at multiple levels of temporal abstraction. She also leads DeepMind Montreal, co-founded the AI4Good Lab for women in AI, and has over 45,000 citations for her research contributions.
What is DeepMind Montreal?
DeepMind Montreal is Google DeepMind's research lab in Montreal, founded in 2017 with Doina Precup as director. The lab focuses on fundamental reinforcement learning research, including hierarchical RL, continual learning, and deep RL. It was DeepMind's second Canadian lab after Edmonton.
What is the options framework in reinforcement learning?
The options framework is a theoretical foundation for hierarchical reinforcement learning that extends standard actions to include temporally extended behaviors. An "option" is a policy that executes over multiple time steps—like "go to the kitchen" rather than "move left." This enables agents to learn complex, reusable skills.
What is the AI4Good Lab?
The AI4Good Lab is a 7-week machine learning program for women and gender-diverse people, co-founded by Doina Precup in 2017. It operates in Montreal (Mila), Toronto (Vector Institute), and Edmonton (Amii), and is part of the Pan-Canadian AI Strategy. The program addresses gender disparity in AI through mentorship and hands-on project work.
Where did Doina Precup study?
Precup earned her BSc and MSc at the Technical University of Cluj-Napoca in Romania, then received a Fulbright Scholarship to study at the University of Massachusetts Amherst, where she earned her MSc (1997) and PhD (2000) working with Richard Sutton and Andrew Barto on reinforcement learning.
How many citations does Doina Precup have?
Doina Precup has over 45,000 citations on Google Scholar, making her one of the most-cited reinforcement learning researchers in the world. Semantic Scholar identifies 2,560 highly influential citations across her 400+ publications.
What is Doina Precup's research focus?
Precup's research focuses on reinforcement learning, particularly temporal abstraction, hierarchical RL, off-policy learning, representation learning, and continual learning. She also works on AI applications in healthcare and other domains with social impact.
Related Reading
Montreal AI Ecosystem
- Yoshua Bengio Complete Guide — Montreal's Turing Award winner
- Mila Quebec AI Institute Guide — The research powerhouse
- Montreal AI Ecosystem Guide — The city's AI landscape
- Joelle Pineau Complete Guide — McGill colleague, FAIR to Cohere
Reinforcement Learning
- Rich Sutton Complete Guide — RL pioneer in Edmonton
- Edmonton RL Capital Guide — Canada's RL hub
- DeepMind Edmonton Story — The Edmonton lab
Canadian AI Research
- CIFAR Complete Guide — The organization that funded deep learning
- Vector Institute Guide — Toronto's AI institute
- Canadian AI Strategy Guide — Government investment
Diversity in AI
- Women in Canadian AI — Leaders shaping the field
Join AGI House Canada to connect with Montreal's AI research community. Scan the QR code to join our WhatsApp group or subscribe to our newsletter.
Data current as of January 2026. Information from McGill University, DeepMind, CIFAR, Mila, Google Scholar, Wikipedia, and AI4Good Lab.
