Meire Fortunato
Meire Fortunato is a research scientist at DeepMind, which she joined in 2016. Meire received her undergraduate and master's degrees in Mathematics from University of Campinas in Brazil (in 2008 and 2010, respectively); and her PhD in Mathematics in 2016 from the University of California, Berkeley.
Meire is co-founder and a member of the general committee of Khipu.ai, an effort to support and strengthen AI research in Latin America.
Meire has worked on a variety of machine learning topics, such as sequence modeling and exploration and understanding the role of memory in reinforcement learning agents. In the last two years, Meire's research focus has been on how to use machine learning for physics simulations -- with emphasis on graph neural networks architectures. This links to her mathematics background, as her PhD thesis was on the generation of high-order meshes by solving partial differential equations with finite element methods.
Research Areas
Authored Publications
Sort By
Noisy Networks for Exploration
Mohammad Gheshlaghi Azar
Bilal Piot
Jacob Menick
Ian Osband
Alexander Graves
Vlad Mnih
Remi Munos
Demis Hassabis
Olivier Pietquin
Charles Blundell
Shane Legg
Proceedings of the International Conference on Representation Learning (ICLR 2018), Vancouver (Canada)
Preview abstract
We introduce NoisyNet, a deep reinforcement learning agent with parametric noise added to its weights, and show that the induced stochasticity of the agent's policy can be used to aid efficient exploration.
The parameters of the noise are learned with gradient descent along with the remaining network weights. NoisyNet is straightforward to implement and adds little computational overhead.
We find that replacing the conventional exploration heuristics for A3C, DQN and dueling agents (entropy reward and epsilon-greedy respectively) with NoisyNet yields substantially higher scores for a wide range of Atari games, in some cases advancing the agent from sub to super-human performance.
View details
