Supervised Actor‐Critic Reinforcement Learning

By Michael T. Rosenstein, Andrew Barto, Jennie Si, Warrren B Powell, Donald C. Wunsch II. 2012

101 citations recorded by the source corpus.

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

Citations in the corpus

Supervised Reinforcement Learning with Recurrent Neural Network for Dynamic Treatment Recommendation (2018, background): “Su-pervised actor-critic [5, 6, 10] uses the indicator signal to pre-train a “guardian” and then combines “actor” output and “guardian” output to send low-risk actions for robots.”
Training Agents using Upside-Down Reinforcement Learning (2019, background): “While there is a rich history of techniques that incorporate supervised learning (SL) into reinforcement learning (RL) algorithms, it is believed that fully solving RL problems using SL is not possible, because feedback from the environment provides error signals in SL but evaluation signals in RL [”
Reinforcement Learning Via Practice and Critique Advice (2010, unknown): “Supervised actor-critic learning (Rosenstein and Barto 2004) follows a similar approach where online policy updates are driven by a combined reward and supervisory signal.”
A comprehensive survey on safe reinforcement learning (2015, methodology): “Risk is defined as the probability of visiting any of the constrained states.”
Interactively shaping robot behaviour with unlabeled human instructions (2019, background): “In [37], the authors present an Actor-Critic architecture that uses instructions for both decision-making and learning.”
Imitation Learning with Demonstrations and Shaping Rewards (2014, result): “Similar comments hold for other work on advice in RL (Rosenstein and Barto 2004; Maclin et al. 2005).”
Constructing Skill Trees for Reinforcement Learning Agents from Demonstration Trajectories (2010, methodology): “Moreover, with the growing realization that learning policies completely from scratch in such domains is infeasible, we may also need to bootstrap learning through a method that provides a reasonable initial policy such as learning from demonstration [5], sequencing existing controllers [6], using a”
Policy gradient in Lipschitz Markov Decision Processes (2015, methodology): “3 Ship steering with water current In this scenario we adapted a standard control problem where the task is to steer a ship, which is cruising at constant speed, to a goal in minimum time (Rosenstein and Barto 2004).”