🚀 go-pugleaf

	     The International Computer Science Institute
		    is pleased to present a talk:

		Thursday, January 16, 1992  2:00 p.m.

			  CHRISTOPH BREGLER
		       University of Karlsruhe

      ``Pattern Matching with Frequency Based Neural Networks''

   This work uses a connectionist network, consisting of binary units,
that are easy to implement in hardware.  In  these units analog values
are represented as ``firing''-frequencies.  Relaxation networks can be
constructed  of these  frequency   based  units to solve  optimization
problems, like the  Task  Assignment Problem.  Hopfield  and Tank have
already shown  this  using analog  units. With  the    frequency based
approach, and a different ``energy  function'', the network will  find
the best assignments  more   frequently. To do  pattern  matching,  we
transformed the  pixel  to pixel matching   into   the task assignment
problem.  In   addition  we   added  ``neighborhood  constraints''  to
encourage pattern rigidity.


       This talk will be held in the Main Lecture Hall at ICSI.
	 1947 Center Street, Sixth Floor, Berkeley, CA 94704
      (On Center between Milvia and Martin Luther King Jr. Way)