SOLVING PARTIALLY OBSERVABLE REINFORCEMENT LEARNING PROBLEMS WITH RECURRENT NEURAL NETWORKS

How researchers see this paper

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

State Representation Learning in Robotics: Using Prior Knowledge about Physical Interaction (2014, background): “[5] learn these functions to predict future rewards conditioned on future actions.”
Learning States Representations in POMDP (2013, background): “, 2011) propose a model based on a recurrent auto-associative memory with history of arbitrary depth, while (Duell et al., 2012) present an extension of RNN for unsupervised learning.”
Learning state representations with robotic priors (2015, background): “Duell et al. [6] learn these functions to predict future rewards conditioned on future actions.”
Batch reinforcement learning on the industrial benchmark: First experiences (2017, methodology): “The topology of these RNNs is described in [21] as ”Markov Decision Process Extraction Network”.”
Interpretable Policies for Reinforcement Learning by Genetic Programming (2017, methodology): “For the IB benchmark two recurrent neural networks (RNNs) have been trained: Note that the Markov decision process extraction topology (Duell, Udluft, and Sterzing, 2012) of the RNNs that we applied here is well-suited for partially observable problems like the IB.”