ji:s paper 1081--1088


1]
1) Pick up 1000 points at random
2) If at least one of the detecors created by the algorithm
   detect the point as non-self then plot it with black circle;
   non-self points not detected is plotted with blanc circle:
   all the self points are plotted with points.
   


self radius is termed by the author but it is somehow misunderstandable.
self cell, per ce, is difined as point and radius implys the area in which
detector can sense the point.

Why the number of detectors depend on self radius?
 => because algorithm terminate in three difernt ways.

experiments are carried out
not performed

They claimed the variable-sized detector outperformed
their other previous proposals
(1) constant-cized detector

constant-sized detector does not have a boundary-dilemma


All of these tecniques can train themselves only with using self/normal patterns.

if we converge a -> 0 the problem is a needle