A simple mechanism that can be used to ``learn'' parameter values over time is the digital low-pass filter. A simple digital low-pass filter is defined by:

out_i+1 = &alpha * in_i + (1 - &alpha) * out_i, where &alpha < < 1.

This filter will remove most short-term ``noise'' in the input, while passing through the long-term trend.

A filter like this was used to adjust weights assigned to heuristic feature detectors in Samuel's checker-player program.

An advantage of such a mechanism is that multiple parameter values can be ``learned'' simultaneously, despite the fact that the system's performance changes as the parameter values change.

Contents    Page-10    Prev    Next    Page+10    Index