Transcript Optional extra slides on the Binding Problem

The Binding Problem
Red Circle & Green Square or Green Circle and Red Square?
Binding likely takes place across small and large
anatomical scales.
Static vs. Dynamic Binding
Static binding => conjunctive representation; collections of visual
elements and properties map onto individual units of the
representation, one unit per conjunction
Hummel (2003)
Potential Solution 1:
Distributive Encoding of Conjunctions:
no need to have all possible conjunctions separately represented.
Problem: Ambiguity solved at one level, but still present at another
=> Still a need for Grandmother cells?
Potential Solution 2: Dynamic Binding:
separate properties represented by separate units,
conjunctions occur via some changing tag external to
the units themselves.
Hummel (2003)
Neural synchrony: things that fire together go together.
Engels et al. (1992)
Trujillo et al. (2002)
Static vs. Dynamic Binding often equated with….
Rate Coding: information encoding via firing rates
Vs.
Fries et al., 2001
Temporal Coding:
information encoding
via timing relationships
McNaughton, 2001
Problems with the Synchrony Hypothesis
1) Observed synchrony may be an epiphenomenon of neural
firings.
Neurons can fire in synchrony by random chance
(Shadlen and Movshon, 1999).
Synchrony might simply reflect the bidirectional
connectivity of the cortex (O’Reilly and Munakata, 2000; Lamme,
2001).
Counter-evidence against epiphenomenal view:
impaired odor detection in bees with a reduction in
synchrony of olfactory cells while preserving high firing
rates (Laurent et al., 1997).
2) Further processing of bound information requires novel
and unknown dynamic mechanisms.
McNaughton, 2001
Maybe temporal Coding might provide a dynamic mechanism
for processing? More research is needed…..
One problem: timing codes highly sensitive to noise
(O’Reilly and Munakata, 2000; Mazurek and Shadlen, 2000).
Attention and Binding
Sometime binding fails…..
“illusory conjunctions”: human subjects presented with an
may perceive incorrect feature conjunctions when
attention is divided over different objects and features.
(Triesman, 1999; Wolfe and Cave, 1999).
Attention may be the “tag” that binds features together.
Neural Synchrony and Attention
Human electrophysiological evidence:
Greater fronto-central gamma-range spectral power to
attended than unattended auditory stimuli.
Tiitinen, Sinkkonen, Reinikainen, Alho, Lavikainen, and Naatanen (1993)
Primate electrophysiological evidence:
Synchronization of pre-synaptic neuron inputs upon postsynaptic cells is increased with increased attention.
(Fries, Reynolds, Rorie, and Desimone, 2001).
This synchronization may boost temporal and spatial summation
of EPSPs and IPSPs, and thus increase the chance to
depolarize post-synaptic cell.
Conjecture:
McNaughton,2001
Neurons within the loci of attention may selectively gate
signals depending upon whether input cells are
synchronized.
IMHO: Binding by convergence (Singer, 1999)
binding occurs via convergence of signals among neuron
in feedforward, feedback, and lateral connections.
Information is irreversibly (i.e. nonlinearly) convolved
(“blended” or transformed) at each synaptic step,
i.e. nonlinear transformation.
Attention findings: Synchrony facilitates communication,
=> high synchrony can be a marker of good
communication
Synchrony might also contribute to synaptic plasticity :
Dynamic interplay between LTD and LTP could work to
create new connections in response to a stimulus that
are reset to near initial conditions when the stimulus is
removed (e.g. Loebel and Tsyodyks, 2002).
Ngezahayo,
Schachner,
and Artola
(2000)
Speculation: such dynamic LTD/LTP interplay may resemble
phenomena of “metaplasticity”: changes in the ability to
induce subsequent LTD/LTP.