A multiagent approach to identifying innovation in design components Public Deposited

http://ir.library.oregonstate.edu/concern/graduate_thesis_or_dissertations/c247dw63m

Descriptions

Attribute NameValues
Creator
Abstract or Summary
  • Innovation is a key element for a product to achieve market success, but identifying it within product or even defining the term is a difficult task. Identifying innovation has been approached in many different ways. Experts in design engineering may identify innovative designs based on an analysis of a product's functions, and statistical techniques may be used to evaluate innovation within a set of products. While there are numerous ways to recognize innovation in a product, there is no straightforward way of identifying how much each component within a product contributes to its innovation. Multiagent systems face an analogous problem; though the performance of a system may be easily assessed, the complex interactions of the agents makes using this system performance to reward each agent ineffective. Difference rewards provide a mechanism for a multiagent system to better quantify the impact of an agent on the system's performance. We introduce the Creative Agents for Repository-Reliant Innovation Engineering (CARRIE) algorithm, which frames the problem of creating a design as supervised learning within a multiagent system. Agents simulate the design process by selecting components to create a product from their training data, and receive external evaluations based on the product-level innovation score. In order to propagate this score to the component selections, the CARRIE algorithm incorporates difference rewards to identify components that positively or negatively impact the overall innovation score within a set of products. Traditional application of the difference reward requires a way to calculate a system’s performance, and then a way to recalculate this performance when an agent is removed in simulation. This presents a problem when we only have the numerical evaluation of the innovation in a product to use as a system performance score, and no indication of how this innovation score was obtained. For this reason, the CARRIE algorithm uses a method by which we can calculate the system score based on the novelty scores of the components in a product. This enables the computation of the difference reward in this domain without actually having a mathematical formulation of an arbitrary system reward.
Resource Type
Date Available
Date Copyright
Date Issued
Degree Level
Degree Name
Degree Field
Degree Grantor
Commencement Year
Advisor
Committee Member
Academic Affiliation
Non-Academic Affiliation
Keyword
Subject
Rights Statement
Peer Reviewed
Language
Replaces
Additional Information
  • description.provenance : Approved for entry into archive by Julie Kurtz(julie.kurtz@oregonstate.edu) on 2013-05-17T22:32:01Z (GMT) No. of bitstreams: 2 license_rdf: 1370 bytes, checksum: cd1af5ab51bcc7a5280cf305303530e9 (MD5) RebhuhnCarrieM2013.pdf: 1452791 bytes, checksum: 80b37ecb34927df776f3aa8c973744d4 (MD5)
  • description.provenance : Approved for entry into archive by Laura Wilson(laura.wilson@oregonstate.edu) on 2013-05-20T19:34:59Z (GMT) No. of bitstreams: 2 license_rdf: 1370 bytes, checksum: cd1af5ab51bcc7a5280cf305303530e9 (MD5) RebhuhnCarrieM2013.pdf: 1452791 bytes, checksum: 80b37ecb34927df776f3aa8c973744d4 (MD5)
  • description.provenance : Made available in DSpace on 2013-05-20T19:34:59Z (GMT). No. of bitstreams: 2 license_rdf: 1370 bytes, checksum: cd1af5ab51bcc7a5280cf305303530e9 (MD5) RebhuhnCarrieM2013.pdf: 1452791 bytes, checksum: 80b37ecb34927df776f3aa8c973744d4 (MD5) Previous issue date: 2013-05-13
  • description.provenance : Submitted by Carrie Rebhuhn (rebhuhnc@onid.orst.edu) on 2013-05-17T00:14:04Z No. of bitstreams: 2 license_rdf: 1370 bytes, checksum: cd1af5ab51bcc7a5280cf305303530e9 (MD5) RebhuhnCarrieM2013.pdf: 1452791 bytes, checksum: 80b37ecb34927df776f3aa8c973744d4 (MD5)

Relationships

In Administrative Set:
Last modified: 08/22/2017

Downloadable Content

Download PDF
Citations:

EndNote | Zotero | Mendeley

Items