Inductive bias

From Lesswrongwiki
Revision as of 09:57, 5 June 2009 by Eliezer Yudkowsky (talk | contribs) (fleshed out)
Jump to: navigation, search
Wikipedia has an article about

The inductive bias of a learning algorithm is the set of assumptions that the learner uses to predict outputs given inputs that it has not encountered

Tom Mitchell[1]

"Inductive bias" refers to your suspicion that if the sun has risen for the last billion days in a row, then it may rise tomorrow as well. Since it is logically possible that the laws of physics will arbitrarily cease to work and that the sun will *not* rise tomorrow, coming to this conclusion requires an inductively biased prior.

This sort of bias is not a bad thing - without "inductive bias" you can't draw any conclusion at all from the data. It's just a different technical meaning attached to the same word.

Primary article

See also