TITLE: Cross-Validation with LULOO

AUTHORS: Paul Haase S{\o}rensen & Magnus N{\o}rg{\aa}rd
         Department of Automation, B326
         Technical University of Denmark
         DK-2800 Lyngby, Denmark
         Phones: (+45) 45253581, (+45) 45253565
         Fax: (+45) 45881295
         Emails: phs,pmn@iau.dtu.dk

         Lars Kai Hansen & Jan Larsen
         Department of Mathematical Modelling B349
         Technical University of Denmark
         DK-2800 Lyngby, Denmark
         Phones: (+45) 45253889, (+45) 45253923
         Fax: (+45) 45880117
         Emails: lkhansen@ei.dtu.dk jl@imm.dtu.dk

The leave-one-out cross-validation scheme for generalization assessment of
neural network models is computationally expensive due to replicated training
sessions. Linear unlearning of examples has recently been suggested
as an approach to approximative cross-validation. Here we briefly review
the linear unlearning scheme, dubbed LULOO, and we illustrate it on a system
identification example. Further, we address the possibility of extracting
confidence information (error bars) from the LULOO ensemble.

Accepted for 1996 International Conference on Neural Information Processing,
ICONIP'96, Hong Kong, Sept. 24-27, 1996.