Lecture 4
Pursuit gain = proportion. E.g. a pursuit gain of 1 means perfect pursuit, 0 means
eyes are not moving at all. >1 means eye is moving faster than target, <1 means
eye is moving slower than target.
Psychological gain = estimate of how fast the target is going, separate from
pursuit gain.
Bayesian models of perception
Signals measure speeds, one measuring a retinal stimulus
and one measuring the eye movements (ER) and then
there are estimates.
The signals are wrong, they have different gains. Eye
movement, pursuit, is not always accurate. Retinal slip
occurs and saccades are used to adapt to this.
By why do we have measurement errors?
Bayesian’s claim that the errors are at the estimates
rather than the signals.
‘Imprecise-signals’ model
Bayesian model claims:
“hard claim” or first claim: sensory evidence is accurate,
the early visual signals, are as accurate as each other. The
difference between them is the difference in precision of
each of them, and there is some ambiguity.
Precision: things are variable. Neurons are variable,
fluctuate. But the world is not. Interpret an invariable
stimulus in the “face” of variability.
Von Helm Hotlz: must be that the brain is having to draw inferences about what
is going on in the world based on variable activity.
Perception is a best guess based on this and prior expectations about the state of
the world. A model is needed to help interpret the activity.
Gives a balance between the fluctuation and prior
expectations.
Sensory evidence is also ambiguous because of various
geometric facts. The short line on the back of the eye
could correspond to any line out in the “real world”.
Don’t know which line it corresponds to – do not see
infinite options, just see one thing. Gives a constant
percept and sometimes gets it wrong.
Two options: in or out. Instead of random – use prior
expectation of light coming from above, so the discs with
shading at the bottom must be sticking out and those
with shading at the top must be concave. So this is an
example of drawing inferences based on an expectation
of something that’s going on in the world.
Pursuit gain = proportion. E.g. a pursuit gain of 1 means perfect pursuit, 0 means
eyes are not moving at all. >1 means eye is moving faster than target, <1 means
eye is moving slower than target.
Psychological gain = estimate of how fast the target is going, separate from
pursuit gain.
Bayesian models of perception
Signals measure speeds, one measuring a retinal stimulus
and one measuring the eye movements (ER) and then
there are estimates.
The signals are wrong, they have different gains. Eye
movement, pursuit, is not always accurate. Retinal slip
occurs and saccades are used to adapt to this.
By why do we have measurement errors?
Bayesian’s claim that the errors are at the estimates
rather than the signals.
‘Imprecise-signals’ model
Bayesian model claims:
“hard claim” or first claim: sensory evidence is accurate,
the early visual signals, are as accurate as each other. The
difference between them is the difference in precision of
each of them, and there is some ambiguity.
Precision: things are variable. Neurons are variable,
fluctuate. But the world is not. Interpret an invariable
stimulus in the “face” of variability.
Von Helm Hotlz: must be that the brain is having to draw inferences about what
is going on in the world based on variable activity.
Perception is a best guess based on this and prior expectations about the state of
the world. A model is needed to help interpret the activity.
Gives a balance between the fluctuation and prior
expectations.
Sensory evidence is also ambiguous because of various
geometric facts. The short line on the back of the eye
could correspond to any line out in the “real world”.
Don’t know which line it corresponds to – do not see
infinite options, just see one thing. Gives a constant
percept and sometimes gets it wrong.
Two options: in or out. Instead of random – use prior
expectation of light coming from above, so the discs with
shading at the bottom must be sticking out and those
with shading at the top must be concave. So this is an
example of drawing inferences based on an expectation
of something that’s going on in the world.
