Reinforcement Learning: : An Introduction

By A. Andrew. 1998

How researchers see this paper

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

In Reinforcement Learning

Citations in the corpus

Playing Atari with Deep Reinforcement Learning (2013, methodology): “ue function, by using the Bellman equation as an iterative update, Q i+1(s;a) = E[r+ 0max a0 Q i(s;a0)js;a]. Such value iteration algorithms converge to the optimal actionvalue function, Q i !Q as i!1[23]. In practice, this basic approach is totally impractical, because the action-value function is ”
Towards Deep Symbolic Reinforcement Learning (2016, background): “m on a stochastic variant of the game. 1 Introduction Deep reinforcement learning (DRL), wherein a deep neural network [1, 2] is used as a function approximator within a reinforcement learning system [3], has recently been shown to be effective in a number of domains, including Atari video games [4]”
Counterfactual reasoning and learning systems: the example of computational advertising (2012, methodology): “The tools described in the following sections are closely related to methods of reinforcement learning (Sutton and Barto, 1998) and methods proposed for various special cases such as multi-armed bandits (Robbins, 1952) and contextual bandits (Langford and Zhang, 2008) problems (see also the bibliogr”
Doubly Robust Off-policy Value Evaluation for Reinforcement Learning (2015, background): “For example, it is critical in intraoption learning when an agent tries to optimize the policies of different options (that is, temporally abstracted actions, sometimes referred to as skills) using the same stream of experience generated by some behavior policy (Sutton et al., 1998).”
Factorized Q-learning for large-scale multi-agent systems (2018, background): “for single-agent, and researchers have utilized various approximation schemes to compactly represent the Q-function, e.g., Radial Basis Function (RBF), Cerebellar Model Articulation Controller (CMAC) [22], Product of Experts (PoE) [20], and Robust Principle Component Analysis (RPCA) [16]. It is reas”
Ranking Sentences for Extractive Summarization with Reinforcement Learning (2018, methodology): “ping content and form verbose and redundant summaries. We next explain how we address these issues using reinforcement learning. 4 Sentence Ranking with Reinforcement Learning Reinforcement learning (Sutton and Barto, 1998) has been proposed as a way of training sequenceto-sequence generation models”
Optimal Completion Distillation for Sequence Learning (2018, background): “sequence. Edit distance is a common task metric. Our goal in sequence learning is to train a model, which achieves high scores of R(y ;ey). Drawing connection with the goal of reinforcement learning (Sutton and Barto, 1998), let us recall the notion of optimal Q-values. Optimal Q-values for a state-”
AI Safety Gridworlds (2017, methodology): “ deep RL algorithms for discrete action spaces. A noteworthy distinction is that Rainbow is an off-policy algorithm (when not using the n-step returns extension), while A2C is an on-policy algorithm (Sutton and Barto, 1998). This helps illustrate the difference between the two classes of algorithms ”
3D Simulation for Robot Arm Control with Deep Q-Learning (2016, neutral): “I. APPROACH A. Deep Q-learning In a standard reinforcement learning problem, a task describes the agent’s interaction with an environment by taking actions and then receiving observations and rewards [13]. Within a task, learning is split into episodes, where each episode represents one “attempt” to”