Transcript Document

The distinct modes of vision
offered by feedforward and
recurrent processing Victor
A.F. Lamme and Pieter R. Roelfsema
Dichotomies in the Visual System?
• What are three dichotomies that Lamme identifies in
the visual system?
Dichotomies in the Visual System?
• What are three dichotomies that Lamme identifies in
the visual system?
• Dorsal vs. Ventral stream
Dichotomies in the Visual System?
• What are three dichotomies that Lamme identifies in
the visual system?
• Dorsal vs. Ventral stream
• Pre-attentive vs. Attentive
Dichotomies in the Visual System?
• What are three dichotomies that Lamme identifies in
the visual system?
• Dorsal vs. Ventral stream
• Pre-attentive vs. Attentive
• Conscious vs. Unconscious
The Feed-Forward Sweep
• What is the feed-forward sweep?
The Feed-Forward Sweep
• The feed-forward sweep is the initial response of
each visual area “in turn” as information is passed to
it from a “lower” area
– a single spike per synapse
– no time for lateral connections
– no time for feedback connections
The Feed-Forward Sweep
• The feed-forward sweep is the initial response of
each visual area “in turn” as information is passed to
it from a “lower” area
• Consider the latencies of the first responses in
various areas
QuickTime™ and a
decompressor
are needed to see this picture.
The Feed-Forward Sweep
• Thus the “hierarchy” of visual areas differs depending
on temporal or anatomical features
• Three aspects of the visual system account for this
fact:
– Some neurons in an area don’t receive direct connections
from the next “lower” area
– multiple feed-forward sweeps progressing at different rates
(I.e. magno and parvo pathways) in parallel
– signals arrive at cortex via routes other than the Geniculostriate pathway (LGN to V1)
The Feed-Forward Sweep
• The feed-forward sweep gives rise to the “classical”
receptive field properties
– tuning properties exhibited in very first spikes
– think of cortical neurons as “detectors” only during feedforward sweep
After the Forward Sweep
• By 150 ms, virtually every visual brain area has
responded to the onset of a visual stimulus
• But visual cortex neurons continue to fire for
hundreds of milliseconds!
After the Forward Sweep
• By 150 ms, virtually every visual brain area has
responded to the onset of a visual stimulus
• But visual cortex neurons continue to fire for
hundreds of milliseconds!
• What are they doing?
After the Forward Sweep
• By 150 ms, virtually every visual brain area has
responded to the onset of a visual stimulus
• But visual cortex neurons continue to fire for
hundreds of milliseconds!
• What are they doing?
• with sufficient time (a few tens of ms) neurons begin
to reflect aspects of cognition other than “detection”
Extra-RF Influences
• One thing they seem to be doing is helping each
other figure out what aspects of the entire scene
each RF contains
– That is, the responses of visual neurons begin to change to
reflect global rather than local features of the scene
– recurrent signals sent via feedback projections are thought
to mediate these later properties
Extra-RF Influences
• consider texture-defined
boundaries
– classical RF tuning
properties do not allow
neuron to know if RF
contains figure or
background
– At progressively later
latencies, the neuron
response differently
depending on whether it is
encoding boundaries,
surfaces, the background,
etc.
QuickTime™ and a
decompressor
are needed to see this picture.
Recurrent Signals in Object
Perception
• Can a neuron represent whether or not its receptive
field is on part of an attended object?
• What if attention is initially directed to a different part
of the object?
Recurrent Signals in Object
Perception
• Can a neuron represent whether or not its receptive
field is on part of an attended object?
• What if attention is initially directed to a different part
of the object?
Yes, but not during the feed-forward sweep
Recurrent Signals in Object
Perception
• curve tracing
– monkey indicates whether a
particular segment is on a
particular curve
– requires attention to scan
the curve and “select” all
segments that belong
together
– that is: make a
representation of the entire
curve
– takes time
QuickTime™ and a
decompressor
are needed to see this picture.
Recurrent Signals in Object
Perception
• curve tracing
– neuron begins to respond
differently at about 200 ms
– enhanced firing rate if
neuron is on the attended
curve
Feedback Signals and the binding
problem
• What is the binding problem?
Feedback Signals and the binding
problem
• What is the binding problem?
• curve tracing and the binding problem:
– if all neurons with RFs over the attended curve spike
faster/at a specific frequency/in synchrony, this might be the
binding signal
Feedback Signals and the binding
problem
• What is the binding problem?
• curve tracing and the binding problem:
– if all neurons with RFs over the attended curve spike
faster/at a specific frequency/in synchrony, this might be the
binding signal
But attention is supposed to
solve the binding problem, right?
Feedback Signals and the binding
problem
• So what’s the connection between Attention and
Recurrent Signals?
Feedback Signals and Attention
• One theory is that attention (attentive processing)
entails the establishing of recurrent “loops”
• This explains why attentive processing takes time feed-forward sweep is insufficient
Feedback Signals and Attention
• Instruction cues (for exaple in the Posner Cue-Target
paradigm) may cause feedback signal prior to
stimulus onset (thus prior to feed-forward sweep)
• think of this as pre-setting the system for the
upcoming stimulus
Feedback Signals and Attention
• We’ll consider the role of feedback signals in
attention in more detail as we discuss the
neuroscience of attention