A: I don't know of any that works too well (and invariantly, including
recognizing mirrored images & different orientation etc... This is
fairly important for a system functioning in a dynamic world, and it
is obviously a tough nut to crack)


Well, "too well" is ambiguous...

There are plenty of image classification algorithms with published
precision and recall results on test databases.

The Poggio et al software seems to work about as well as humans
do on the task of classifying images based on very rapid viewing
(e.g. recognizing if there is a cat in an image flashed very rapidly
in front of you).  This makes sense because they have simulated
only feedforward connections.

Question is whether, when they add feedback connections, they
will get a system that works about as well as humans do on the
task of classifying images based on more leisurely viewing.  No
existing software system, including Poggio's or Hawkins', can
equal humans on this task.

But anyway, in terms of practical performance on real images,
Hawkins' is nowhere near the best image classification system
out there.  Its ultimate potential when further developed is a whole
other, and deeper, question, of course.

I find Hawkins' approach lies in a funny middle-area between
computer science and brain science.  Poggio's work actually
tries to simulate visual cortex.  Hawkins' incorporates Bayes
net learning which is known NOT to be how cortex works, but in an
architecture inspired closely by cortex ...
tries in this way to emulate the essence of what visual cortex
does.  Whereas my inclination tends to be toward approaches
that stray even further from brain science...

-- Ben G

-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?member_id=231415&user_secret=fabd7936

Reply via email to