Adaptive Tile Coding for Value Function Approximation

By Shimon Whiteson. 2007

95 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.

Citations in the corpus

Fitted Q-Learning for Relational Domains (2020, methodology): “Indeed, a related method for solving ADP is that of tile-coding (Whiteson et al. 2007), a linear piece-wise approxi-mator that partitions the state space into (potentially overlapping) regions called tiles .”
Exploiting Structure and Agent-Centric Rewards to Promote Coordination in Large Multiagent Systems (2014, background): “Learning in large multiagent systems with large stateaction spaces can be extremely challenging, if not infeasible, without some form of exploitation of the problem structure to reduce the agent-to-agent coordination requirements [12, 32, 33, 38].”
Online Discovery of Feature Dependencies (2011, methodology): “While manually specifying important feature dependencies has been pursued in the literature (Albus, 1971; Sturtevant & White, 2006), learning dependencies online and expanding the representation automatically is becoming more popular as it simplifies the design process (e.g., Munos & Moore, 2002; Ra”
Reinforcement learning with misspecified model classes (2013, methodology): “ing learning bias, there has been a great deal of work in reducing representational bias by growing the representation using nonparameteric dynamics models [30], [31], Bayesian nonparametric dynamics models [25], nonparametric value functions [32], [2], and kernel-based methods [33], [34].”
Growing Action Spaces (2019, methodology): “In reinforcement learning, this has often taken the form of adaptively growing the resolution of the state space considered by a piecewise constant discretised approximation (Moore, 1994; Munos & Moore, 2002; Whiteson et al., 2007).”
Exploiting structure and utilizing agent-centric rewards to promote coordination in large multiagent systems (2013, background): “Learning in large multiagent systems with large stateaction spaces can be extremely challenging, if not infeasible, without some form of exploitation of the problem structure to reduce the agent-to-agent coordination requirements [12, 32, 33, 38].”