Binding Problem - a Possible Solution

Philip Dorrell p at dorrell.demon.co.uk
Thu Dec 14 16:45:21 EST 1995


I propose a possible solution to the "binding problem", i.e. how the brain 
links together different perceptions relating to a single object, in my
web page -

http://www.xmission.com/~gastown/dorrell/binding.htm

Summary:

Solving the binding problem is related to solving the figure/ground
problem because both relate to the perceptions of objects. To solve
the figure/ground problem for a given object, the brain needs to know
what the object is. And to know what the object is it has to know which
of its current perceptions relate to the current object and which don't.
Those perceptions identified as coming from the current object can be
emphasised, by firing appropriate backprojection neurons, i.e. excitatory 
neurons that take signals from areas representing more processed 
perceptions back to those representing less processed perceptions. With this 
emphasis, the brain has a better idea what the object is, and can use this 
information to more accurately separate figure and ground. And so on, 
performing an iterative calculation, eventually settling on a (locally) 
optimal answer.

Implicit in this scheme is the existence of a pre-ordained motive to 
identify objects and separate them from the background.

I am suggesting that the sole purpose of backprojection neurons is to solve
the figure/ground separation problem. In solving this problem, the brain
is necessarily learning to identify different sorts of objects. When a new
type of object appears, the brain will use its experience of similar objects
to try and solve the figure/ground problem, as well as using general 
principles of figure/ground separation (e.g. common movement, sharp edges,
clues from similarity between different parts of an object). At the same time
it learns about this new object and adds it to its repertoire of object
identification skills.

The object identification expertise will be distributed around all neurons 
that are able to affect the activity of the backprojection neurons. There may 
be no central point of control, but for a given object, neurons in different
areas will communicate with each other if this helps them. (For ideas on
learning from the point of view of the individual neuron, see my web page 
"Meaning of a Neuron".)

Visual perception is very object-oriented. So is language perception.
Words are objects, and phrases, clauses and sentences are aggregate objects.
Perhaps speech object perception has evolved by using existing functionality 
employed in visual perception. But spoken language objects have a strictly
temporal existence. To map from spoken language object perception to visual
perception requires a mapping of the dimension of time into a spatial 
dimension. That is, as if words coming into our ears were being written down
and we were visually scanning them as they went past.

This may explain why we learn to read so easily.

If our perception of the world is so strongly oriented towards the perception
of objects, then this suggests mechanisms by which we can make things that
are difficult to understand easier to understand. We can do this by making 
them more object-like, and presenting them in such a way that solving a 
figure/ground problem where they are the figure requires us to understand what 
they are.

Philip Dorrell   Email: p at dorrell.demon.co.uk
Articles on consciousness, dreams, music, evolution, anthropic principle at
http://www.xmission.com/~gastown/dorrell/

                 



More information about the Neur-sci mailing list