Re: DM: Small data sets

Warren Sarle <saswss@unx.sas.com> disse:
 >
 > No, neither neural nets nor regression
 > will fail to learn if they are programmed
 > correctly. The danger is that they will
 > learn too well and overfit. [...] The most
 > serious problem is that extrapolation outside
 > the subspace spanned by the training set
 > may fail miserably.

Well, in terms of DM, that cannot be called "learning". Generalization is 
probably the most important issue in a DM learning solution. Maybe only 
behind accuracy. Maybe. :)


[]s
Rafael Figueira
-------
"Knowledge is the antidote to fear" - Ralph Waldo Emerson

• Prev by Date: DM: Extended Deadline: ECAI 2000 WS on Balancing Reactivity and Social
• Next by Date: DM: CFP: CCIA-2000 Cooperative Information Agents
• Prev by thread: Re: DM: Small data sets
• Next by thread: DM: Data mining
• Index(es):
  • Date
  • Thread