Neural Network Ensembles in Reinforcement Learning

By Stefan Faußer, Friedhelm Schwenker. 2015

59 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Management Science and Operations Research

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. Quoc V. Le
3. Steffen Udluft
4. Thomas A. Runkler
5. Mohammad Norouzi
6. Dale Schuurmans
7. Pieter Abbeel
8. Eiji Uchibe

In Management Science and Operations Research

1. Mohammad Norouzi
2. Craig Boutilier
3. Steffen Udluft
4. Dale Schuurmans
5. Kenji Doya
6. Thomas A. Runkler
7. Eiji Uchibe
8. Keith W. Ross

Citations in the corpus

• Striving for Simplicity in Off-policy Deep Reinforcement Learning (2019, methodology): “Ensemble-DQN is a simple extension of DQN which approximates the Q -values via an ensemble of Q-functions [18, 45, 1].”
• Randomized Ensembled Double Q-Learning: Learning Fast Without a Model (2021, background): “It has also long been recognized that using ensembles can improve the performance of DRL algorithms (Faußer & Schwenker, 2015; Osband et al., 2016).”
• Particle swarm optimization for generating interpretable fuzzy reinforcement learning policies (2016, methodology): “on Q iteration algorithms with neural networks have been published in past years. In [11] an RL method that monitors the learning process is presented. Furthermore, Hans et al. [16] and Fauˇer et al. [10] applied ensembles of neural networks to form a committee of multiple agents and showed that thi”
• Online meta-learning by parallel algorithm competition (2017, methodology): “ the differences in column heights instead of the absolute differences. The binary state vectors were of length 460in SZ-Tetris and 260in 10×10 Tetris. We used the following reward function proposed by Faußer and Schwenker (2013): r(s)=e−(number of holes in s)/z. (17) Here, zwas set to 33for SZ-Tetris”
• ConQUR: Mitigating Delusional Bias in Deep Q-learning (2020, methodology): “One direct approach uses ensembling, with multiple Q-approximators updated in parallel (Faußer & Schwenker, 2015; Osband et al., 2016; Anschel et al., 2017) and combined to reduce instability and variance.”
• An Optimistic Perspective on Offline Reinforcement Learning (2019, methodology): “Ensemble-DQN is a simple extension of DQN that approximates the Q-values via an ensemble of parameterized Qfunctions (Faußer & Schwenker, 2015; Osband et al., 2016; Anschel et al., 2017).”