Multisensory integration - cue combination Flashcards
tactile info when engaging/touching/manipulating an object =
haptic feedback
different senses might provide conflicting information about a sensory stimulus =
sensory conflicts
sensory conflicts need to be resolved. How?
we need to make a choice to resolve the conflict
does vision or touch dominate judgement of object size?
vision dominates perceived object size
what is it called when visual input dominates to provide a more accurate judgement of object size?
visual capture > this is the VISUAL CAPTURE EFFECT
does touch have any influence on the visual capture effect?
yes but only small but consistent influence
what were the 3 ways Rock and victor assessed visual capture?
pointing, feeling, drawing
is there a strict sensory hierarchy?
no, audition can dominate vision
who conducted a study to show that audition can dominate vision by testing for sensory conflict?
Shams et al, 2000
what were the results from Shams et al, 2000 study?
number of auditory beeps determined the number of reported visual flashes, sound was followed more closely than sight e.g. 1 flash + 2 beeps = 2 beeps reported. SHOWS WE TRUST AUDITION MORE THAN VISUAL SENSE
what is the modality precision hypothesis?
modality with highest precision and lowest uncertainty is chosen depending on the task type
following the modality precision hypothesis > spatial task uses _____, temporal task uses ______
vision, audition
what 3 things is sensory uncertainty due to?
perceptual limits, neural noise, cognitive resource limits
communication between neurons at the synapse doesn’t work > leads to uncertainty. this is due to?
neural noise
different senses have different levels of ______ uncertainty
noise
how did Ernst and Banks (2002) shift ppts attention to different senses?
by manipulating the level of sensory uncertainty
what does MLE stand for?
Maximum Likelihood Estimation
what is a normative model?
optimal solution to how a problem should be solved
what is a normative model based on?
theory
what is a process model?
how a problem is actually solved
what is a process model based on?
data
how should we solve problem of sensory integration?
pick integration method that minimises sensory uncertainty > maximum likelihood estimator
when making an estimate based on haptic and visual inputs what is the best way to get a maximum likelihood estimation?
combine senses and their uncertainty to get an INTEGRATED SIGNAL with the smallest possible combined uncertainty
a large distribution means high ______ which means high sensory _______ and a lower _______
variance, uncertainty, probability
a ______ distribution means low variance which means ___ sensory uncertainty and a _____ probability
small, low, higher
different senses have different amounts of what?
sensory noise
when there are 2 different sense inputs (visual and haptic) there are 2 distributions produced because of 2 different estimates. why are 2 estimates produced?
because sensory conflict is created
how are 2 probabilities integrated if both estimates have the same amount of variance and uncertainty level?
place the integrated estimate exactly in the middle
does a combined estimate of 2 sense inputs have lower or higher uncertainty?
lower! smaller variance = lower uncertainty > higher probability
when integrating probabilities, if visual uncertainty is lower than haptic uncertainty what does theory suggest we should trust?
vision
if visual uncertainty is smaller than haptic how are the estimates combined?
biased towards the visual estimate
visual weight of 1 and haptic weight of 0 = only trust ____ input
visual
visual weight of 0 and haptic weight of 1 = only trust _____ input
haptic
according to MLE what can be calculated from the variance (sensory uncertainty) of the visual and haptic distribution?
optimal weights
integrating info from multiple sources always causes uncertainty to _______
decrease (as variance is decreased when combining)
combining variance leads to a ______ in variance (minimal variance)
decrease
after converting sensory inputs from different senses into a coherent reference frame, info needs to be _______ and sensory conflicts need to be _______
integrated, resolved
MLE is a ________ model that allows us to calculate the _______ way to integrate info
normative, optimal
what is shown with perceived bar length when no visual noise is added? (Ernst & Banks, 2002)
biased towards visual input. shows visual dominance > visual capture (we trust visual over haptic input)
what happens when we add visual noise? (Ernst & Banks, 2002)
perception of bar length moves more towards haptic estimate and becomes determined by both visual and haptic inputs
explain why adding visual noise means theres a shift in our estimate towards haptic estimate
visual input is more noisy > we have more uncertainty and can’t trust it > shift estimate towards haptic feedback as can’t fully rely on visual
what happens when even more (high) visual noise is added?
perception of bar length is determined by haptic inputs and we rely on haptic over visual
as visual noise increases we move toward haptic capture as visual weight _______ and haptic weight ________
decreases, increases
more visual noise = more ________ which means we can’t trust its estimate
uncertainty
do we always integrate info optimally?
idea that we don’t integrate it optimally as heuristics aren’t optimal (but still work well enough)
is it easier to know the uncertainty for optimal integration in sensory perception or cognitive reasoning?
sensory perception! can generally estimate noise and uncertainty levels
why is it hard we always calculate uncertainty levels in our brains?
calculations are intractable and take a long time
what do we use instead of integrating info optimally and why?
use HEURISTICS! suboptimal but fast and provide good enough solutions
we are good at estimating ______ noise but bad at estimating _______ noise
sensory, cognitive
are probabilities encoded in the brain? discuss.
little direct electrophysiological evidence but several plausible theories > may be down to neurons to tell us the variance? different neurons may encode different probability values? different neurons may have different estimates? (active research area with lots of debate)
