Decisions under scarce resources and bounded optimality

Bounded optimality refers to the goal of optimizing the expected utility of a reasoning system, given the environment in which the system is immersed. This goal differs from the related work of investigators (including John March and Herbert Simon) who have proposed heuristic approaches to rationality under bounded resources. To compute the expected utility of an agent's behavior at design time, we must consider the set of responses a system makes to a sequence of challenges over time. We must also identify the design actions that are available for optimizing an agent's behavior by considering the set of controllable parameters and invariant constraints in the hardware or software architecture of the agent. Specifying different sets of constraints on the constitution of a reasoning system defines different classes of bounded optimality.

Bounded optimality, and the goal of optimizing the expected utility of an agent given a sequence of challenges over time (e.g., an agent's lifetime), was introduced in:
Click here to access a brief excerpt on bounded optimality from this article.

Other early work at Stanford examining aspects of rationality under bounded resources and bounded optimality is described in:

Accessing Relevant Papers

The Stanford technical reports list contains related earlier papers.

Link to Other Papers, Abstracts

Back to Eric Horvitz's home page.