Proposed metrics for transfer learning Public Deposited

http://ir.library.oregonstate.edu/concern/technical_reports/st74cr803

Version 4

Descriptions

Attribute NameValues
Creator
Abstract or Summary
  • Summary: Four proposed metrics: [1] average relative reduction in training time (sample size, number of training experiences) [2] jumpstart (initial advantage of transfer algorithm) [3] handicap (how long it takes the no-transfer algorithm to overcome the jumpstart) [4] asymptotic advantage (how much better the transfer learning algorithm does in the limit of large sample sizes)
Resource Type
Date Available
Date Issued
Series
Keyword
Table of Contents
  • Shortcomings of Transfer Ratio -- Proposed Metrics -- Notation -- Learning Speedup and Relative Reduction -- Statistical Interpretation of Speedup -- Integrated Speedup -- Issues -- Partial Solution: Remove the Problem Regions -- There is Still a Problem -- Solution: Use Relative Reduction Instead -- One More Quantity of Interest -- Corner Cases -- Additional Notes -- Summary -- Relative Reduction vs. Transfer Ratio.
Rights Statement
Publisher
Peer Reviewed
Language
Replaces
Additional Information
  • description.provenance : Made available in DSpace on 2012-12-26T23:03:18Z (GMT). No. of bitstreams: 1 2007-5.pdf: 405945 bytes, checksum: b0f74b02f420c50213b90b055b630bf5 (MD5) Previous issue date: 2006-04-30
  • description.provenance : Approved for entry into archive by Laura Wilson(laura.wilson@oregonstate.edu) on 2012-12-26T23:03:17Z (GMT) No. of bitstreams: 1 2007-5.pdf: 405945 bytes, checksum: b0f74b02f420c50213b90b055b630bf5 (MD5)
  • description.provenance : Submitted by Laura Wilson (laura.wilson@oregonstate.edu) on 2012-12-26T23:02:18Z No. of bitstreams: 1 2007-5.pdf: 405945 bytes, checksum: b0f74b02f420c50213b90b055b630bf5 (MD5)

Relationships

Parents:

This work has no parents.

Last modified

Downloadable Content

Download PDF

Items