Policy analysis using Q-learning Conference Paper uri icon

abstract

  • Building of complex white box models brings the need to build tools that guide validation, verification and analysis processes. The goal of this study is to develop an automated tool for policy analysis. The tool utilizes an approximate reinforcement algorithm to improve the behavior of the simulation according to predefined objectives. The Stochastic and complex nature of the model makes approximate learning algorithms a good fit for the problem. The approximation technique requires a summary of information about the model that the user finds essential. This information is subjective. Hence, depending on the run results, user may verify whether user's understanding of the model overlaps with the model's representation of the system. Therefore, the tool merges a policy analysis phase with the verification and validation phases.

author list (cited authors)

  • Eksin, C.

publication date

  • December 2010