Literature/1995/Yamanishi


 * ACM Digital Library @

Authors

 * Kenji Yamanishi
 * NEC Research Institute, Inc., 4 Independence Way, Princeton, NJ

Comments


The document retrieval system S is to predict the document(s) D most similar to the user query E, and the user U is to discriminate the outcome.

As the one-to-one-scaled map is impractical and implausible, so may be the full text D for System-User communication S-U so that it used to be necessarily surrogated, say, into an abstract or a set of keywords d.


 * Prediction and discrimination
 * Excerpt from Michael Ramscar, Research Statement, October 2010 http://psych.stanford.edu/~ramscarlab/Research_Statement.pdf

An aspect of learning theory that has been frequently misunderstood is the role that informativity plays in learning. Although learning models are usually referred to as "associative" -- which might seem to suggest that learning works by simply noting where cues and events co-occur in the environment -- this is not the case. Indeed, for over 40 years the process of "associative learning" has been formally conceived of in terms of prediction and discrimination: learning processes discriminate between predictive cues on the basis of the information they provide. What this means in practice is that learning is driven far more by predicted events that don't occur, than it is by positive pairings of cues and events.