Approximate dynamic programming in knowledge discovery for rapid response

Peter Frazier, Warren Powell, Savas Dayanik, Paul Kantor

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

One knowledge discovery problem in the rapid response setting is the cost of learning which patterns are indicative of a threat. This typically involves a detailed follow-through, such as review of documents and information by a skilled analyst, or detailed examination of a vehicle at a border crossing point, in deciding which suspicious vehicles require investigation. Assessing various strategies and decision rules means we must compare not only the short term effectiveness of interrupting a specific traveler, or forwarding a specific document to an analyst, but we must also weigh the potential improvement in our profiles that results even from sending a "false alarm". We show that this problem can be recast as a dynamic programming problem with, unfortunately, a huge state space. Several specific heuristics are introduced to provide improved approximations to the solution. The problems of obtaining real-world data to sharpen the analysis are discussed briefly.

Original languageEnglish (US)
Title of host publicationProceedings of the 42nd Annual Hawaii International Conference on System Sciences, HICSS
DOIs
StatePublished - 2009
Event42nd Annual Hawaii International Conference on System Sciences, HICSS - Waikoloa, HI, United States
Duration: Jan 5 2009Jan 9 2009

Publication series

NameProceedings of the 42nd Annual Hawaii International Conference on System Sciences, HICSS

Other

Other42nd Annual Hawaii International Conference on System Sciences, HICSS
Country/TerritoryUnited States
CityWaikoloa, HI
Period1/5/091/9/09

All Science Journal Classification (ASJC) codes

  • Computer Science Applications
  • Information Systems

Fingerprint

Dive into the research topics of 'Approximate dynamic programming in knowledge discovery for rapid response'. Together they form a unique fingerprint.

Cite this