Summary: Estimating Learning Performance
Using Hints
Zehra Cataltepe 1 and Yaser S. Abu­Mostafa
California Institute of Technology
Computer Science Department
Pasadena, CA 91125 USA
zehra@csvax.cs.caltech.edu
Learning from hints is a learning mechanism that unifies learning from examples and learning from
explicit rules. Performance of the learning system on different hints may be used to estimate the learning
performance of the system on the function. We develop a formula that estimates learning performance,
without using a validation error, but only errors on hints. We give the derivation and test examples for
this specific formula.
LEARNING FROM HINTS
Learning from hints [1], is a learning mechanism that unifies learning from examples and learning from
explicit rules. It allows us to express rules in the form of examples and then use a learning­from­examples
algorithm to teach this information to the system. Since usually we know more than just a finite set of
input/output examples about a function, hints allow us to use the additional information, such as invariance
properties, monotonicity, being binary, etc., in order to be able to achieve better solutions.
In the learning­from­examples paradigm, in order to measure how well a system has learned a function,
the usual method is to set aside some examples of the function as a validation set, use the remaining

Source: Abu-Mostafa, Yaser S. - Department of Mechanical Engineering & Computer Science Department, California Institute of Technology

Collections: Computer Technologies and Information Sciences