English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Conventional reinforcement learning methods for Markov decision processesrely on weakly-guided, stochastic searches to drive the learning process. Itcan therefore be difficult to predict what agent behaviors might emerge. Inthis paper, we consider an information-theoretic cost function for performingconstrained stochastic searches that promote the formation of risk-averse torisk-favoring behaviors. This cost function is the value of information, whichprovides the optimal trade-off between the expected return of a policy and thepolicy's complexity; policy complexity is measured by number of bits andcontrolled by a single hyperparameter on the cost function. As the policycomplexity is reduced, the agents will increasingly eschew risky actions. Thisreduces the potential for high accrued rewards. As the policy complexityincreases, the agents will take actions, regardless of the risk, that can raisethe long-term rewards. The obtainable reward depends on a single, tunablehyperparameter that regulates the degree of policy complexity.  We evaluate the performance of value-of-information-based policies on astochastic version of Ms. Pac-Man. A major component of this paper is thedemonstration that ranges of policy complexity values yield different game-playstyles and explaining why this occurs. We also show that ourreinforcement-learning search mechanism is more efficient than the others weutilize. This result implies that the value of information theory isappropriate for framing the exploitation-exploration trade-off in reinforcementlearning.

Analysis of Agent Expertise in Ms. Pac-Man using Value-of-Information-based Policies