Marlos C. Machado

I am a senior research scientist at DeepMind Alberta, an adjunct professor at the University of Alberta, and a Canada CIFAR AI Chair through Amii. My research interests lie broadly in machine learning, specifically in reinforcement learning, representation learning, optimization, and real-world applications of all the above.
[Google Scholar] [DBLP]


Prospective students, click here.

Marlos C. Machado

Recent News

    • Jan. 2022: We had a paper accepted at AISTATS’22.
      • A General Class of Surrogate Functions for Stable and Efficient Reinforcement Learning
        (w/ Vaswani, Bachem, Totaro, Mueller, Garg, Geist, Castro, & Le Roux).
    • Oct. 2021: Our recent paper is now available on arXiv.
      • Temporal Abstraction in Reinforcement Learning with the Successor Representation (w/ Barreto & Precup).
    • Oct. 2021: I gave a talk and participated in a panel at the Microsoft Summit Workshop on RL, Forwards and Backwards: Insights from Neuroscience.
      • Temporal Abstraction in RL with the Successor Representation.
    • Jun. 2021: I’ve been appointed Canada CIFAR AI Chair.
    • Jun. 2021: I’ve been appointed Amii Fellow.
    • May 2021: We had a paper accepted at ICML’21.
      • Beyond Variance Reduction: Understanding the True Impact of Baselines on Policy Optimization (w/ Chung*, Thomas*, & Le Roux).
    • Jan. 2021: We had a paper accepted at ICLR’21.
      • Contrastive Behavioral Similarity Embeddings for Generalization in Reinforcement Learning (w/ Agarwal, Castro, & Bellemare).
    • Jan. 2021: I started as a Research Scientist at DeepMind.
    • Jan. 2021: I became an adjunct professor at the University of Alberta.
    • Dec. 2020: We had a paper accepted at Nature.
      • Autonomous Navigation of Stratospheric Balloons using Reinforcement Learning (w/ Bellemare, Candido, Castro, Gong, Moitra, Ponda, & Wang).
    • Sep. 2020: We had a paper accepted at NeurIPS’20.
      • An Operator View of Policy Gradient Methods (w/ Ghosh & Le Roux).
    • Feb. 2020: I gave a talk at Stanford University.
      • Temporal Abstraction in RL with the Successor Representation.

Blog Posts

Different behaviours for baselines that induce the same variance

The True Impact of Baselines in Policy Gradient Methods

Looking at Policy Gradient Methods from an Operators Perspective

Hello, world!