Incremental Off-policy Reinforcement Learning Algorithms

By A. Mahmood. 2017

22 citations recorded by the source corpus.

Subfields: Computational Theory and Mathematics, Electrical and Electronic Engineering

How researchers see this paper

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

In Computational Theory and Mathematics

1. Hado van Hasselt
2. David Silver
3. Zhongwen Xu
4. Wojciech M. Czarnecki
5. Simon Schmitt
6. Thomas Brox
7. A. Mahmood
8. James Bergstra

In Electrical and Electronic Engineering

1. Hado van Hasselt
2. Zhongwen Xu
3. Matteo Hessel
4. David Silver
5. Wojciech M. Czarnecki
6. Thomas Brox
7. A. Mahmood
8. James Bergstra

Citations in the corpus

• Meta-Gradient Reinforcement Learning (2018, methodology): “A similar scheme was proposed by Mahmood [2017].”
• Multi-task Deep Reinforcement Learning with PopArt (2018, background): “This leads to the v -trace return (Espeholt et al. 2018) A very similar target was proposed for the ABQ( ζ ) algo-rithm (Mahmood 2017), where the product ρ t λ t was considered and then the trace parameter λ t was chosen to be adaptive to lead to exactly the same behaviour that c t = ρ t λ t =…”
• CrossNorm: Normalization for Off-Policy TD Reinforcement Learning (2019, background): “A proof can be found in Mahmood (2017) that whileK is positive-definite only in the on-policy case, in the off-policy setting K still always has positive real parts in its eigenvalues.”
• Benchmarking Reinforcement Learning Algorithms on Real-World Robots (2018, background): “However, to learn finer policies with faster action cycle times, moving toward efficient ordering of computations (Travnik et al. 2018, Mahmood et al. 2018) together with inexpensive incremental updates (Mahmood 2017) or asynchronous off-policy updates (Gu et al. 2017) would be essential.”