Two-Timescale Networks for Nonlinear Value Function Approximation

By Wesley Chung, Somjit Nath, Ajin George Joseph, Martha White. 2019

51 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Networks and Communications, Reinforcement Learning, Optimization & Training

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Sergey Levine
2. Ofir Nachum
3. Marc G. Bellemare
4. Shimon Whiteson
5. Dale Schuurmans
6. Pablo Samuel Castro
7. Pieter Abbeel
8. Yoshua Bengio

In Computer Networks and Communications

1. Martha White
2. Dale Schuurmans
3. Marc G. Bellemare
4. Will Dabney
5. Tor Lattimore
6. Adam White
7. Pablo Samuel Castro
8. Clare Lyle

In Reinforcement Learning

1. Sergey Levine
2. Marc G. Bellemare
3. Ofir Nachum
4. Dale Schuurmans
5. Shimon Whiteson
6. Pablo Samuel Castro
7. Pieter Abbeel
8. Mohammad Norouzi

In Optimization & Training

1. Martha White
2. Adam White
3. Ofir Nachum
4. Dale Schuurmans
5. Will Dabney
6. Richard S. Sutton
7. Marc G. Bellemare
8. Tor Lattimore

Citations in the corpus

• DeepMDP: Learning Continuous Latent Space Models for Representation Learning (2019, background): “Francois-Lavet et al. (2018) and Chung et al. (2019) have considered similar latent losses, but to the best of our knowledge ours is the first theoretical analysis of these models.”
• Planning with Expectation Models (2019, methodology): “An alternative is to learn the mapping x using auxiliary tasks and approximate the value function using the learned state-features [Chung et al. , 2018; Jaderberg et al. , 2016].”
• A Geometric Perspective on Optimal Representations for Reinforcement Learning (2019, background): “erest, the weights can also be optimized more efficiently than the representation itself, warranting the view that the representation should be adapted using a different process (Levine et al., 2017; Chung et al., 2019). Note that a trivial “value-as-feature” representation exists for the single-poli”