Skip to content
View DonnaVakalis's full-sized avatar

Block or report DonnaVakalis

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don't include any personal information such as legal names or email addresses. Markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
DonnaVakalis/README.md

Hi, I'm Donna

Postdoctoral Fellow at Mila – Quebec AI Institute, co-supervised by Yoshua Bengio and David Rolnick.

I develop AI systems to reduce carbon emissions from buildings—which account for ~37% of global CO₂ emissions. My research focuses on graph-structured world models and reinforcement learning for HVAC control, with an emphasis on building transferable systems that generalize across different building types, layouts, and climates. Previously I studied causal inference and Bayesian methods for decision-making, in the context of indoor air quality and energy use in cities.


Current Research

My work sits at the intersection of graph neural networks, model-based RL, and building physics:

  • Graph Dreamer — A world model architecture for variable-size, heterogeneous graph environments that learns structural relationships governing thermal dynamics, enabling zero-shot transfer across buildings with different topologies
  • HVAC-GRACE — Graph-based Bayesian forecasting and control for buildings
  • HOT Dataset — A dataset of ~150,000 simulated controllable buildings and gym for transfer learning research

I'm particularly interested in how we can learn generalizable physics from simulations


Selected Publications

  • HVAC Spice — UrbanAI @ NeurIPS 2025 and newer version under review @ ICLR [OpenReview]
  • Graph Dreamer: Temporal Graph World Models for Sample-Efficient and Generalisable RL — WiML @ NeurIPS 2025 [OpenReview]
  • A HOT Dataset: 150,000 Buildings for HVAC Operations Transfer Research — BuildSys 2025 [HuggingFace]
  • Challenges and Opportunities of IoT in Occupant-Centric Building Operations — Current Opinion in Environmental Sustainability, 2023 [Paper]

📚 Full publication list on Google Scholar


Technical Focus

Graph Neural Networks  •  Model-Based RL  •  EnergyPlus  •  PyTorch Geometric  •  Spatiotemporal Modeling

Most of my recent code lives in private repositories (pre-publication research), but I'm working on releasing components as papers are published.


Beyond Research

Two-time Olympian in modern pentathlon (London 2012, Rio 2016). Currently based in Nice, France. Board member at Racing to Zero, a nonprofit focused on sustainability in sport.


📫 Connect

Website LinkedIn Google Scholar ORCID

Pinned Loading

  1. GNN-BuildingDataGen GNN-BuildingDataGen Public

    Generating datasets of time-series data with spatial information and other meta-data of simulated buildings for use by graph neural networks.

    HTML 1

  2. stochastic-control stochastic-control Public

    Forked from adityam/stochastic-control

    Course notes for ECSE 506: Stochastic Control and Decision Theory

    Jupyter Notebook 1

  3. CCAI CCAI Public

    Summer School Tutorials and Team Work

    Jupyter Notebook

  4. forecasting-MRT-gap forecasting-MRT-gap Public

    Data-driven approach to determine ‘what is the right technology to implement for cooling’ given a changing climate

    Jupyter Notebook

  5. Livability Livability Public

    Initial proof-of-concept for a personalized livability web dashboard, one that lets the user weigh weather, transportation access, population density, air quality and a few other metrics. This proj…

    Jupyter Notebook