Visual Neuro

Question 1

why are saccadic latencies so long? what is occult motor procrastination?

Answer 1

Although low-level centres can quickly determine where object
is, but not whether it is worthwhile looking at they are held in check until higher-level cortical processes decide suitability of target. This is oculomotor procrastination. Saccadic latency is therefore reflecting decision making time of higher level centres.

Question 2

latency distributios: what is a raw latency like as a diagram?

Answer 2

it is a skewed histogram usually.

Question 3

How do we remove the skewed latency

Answer 3

we plot the reciprocal of the latency 1/T which ends up looking like a normal distribution.

Question 4

Sigmoid curve illustrates what?

Answer 4

when the reciprocal latency is plotted as cumulative. At 50%, demonstrates 50% of area of normallly dsitributed reciprocal of latency graph.

Question 5

What does the

Question 6

probit graph

Answer 6

plots normally distributed reciprocal of latency as cumilative percent probability as a straight line.

Question 7

what is LATER?

Answer 7

Model of latency data. stands for linear approach to threshold with ergodic rate. explains why reciprocal of reaction times is normally distributed.

Question 8

Explain the LATER process.

Answer 8

in response to a stimulus coming on, we have
some sort of internal decision signal which we’re gonna call S that starts from a baseline.
* And then, in response to that stimulus coming on, it starts to rise linearly where baseline level is called S0 .
* It’s going to rise with a rate R.
* And then there’s gonna be this threshold ST,
- And when this signal hits that threshold, then we initiate
a response.
* the time between the stimulus onssetting
and our response to that stimulus is going to be the latency or the reaction time.
* Now, a key thing in our, um, later model is
this. Whilst the rate of rise on any given trial, uh
is linear between trials, that rate varies randomly and that that rate varies as a Gaussian normal distribution.
* So it follows that if the rate of rise of
a line the gradient of a line varies as a
normal distribution, where that line intercepts the Y axis.
That intercept will be distributed as the reciprocal of the,
uh of the normal distribution.

Question 9

What is u (mew)

Answer 9

average rate of information arrival.

Question 10

When is S0 closer to threshold line?

Answer 10

when expectation of making a particular decision increases. Takes us less time to get to threshold.

Question 11

If incoming information does not give more supporting evidence to a particular hypothesis what will rate of rise look like across the two hypotheses?

Answer 11

The rate of rise for the hypotheses will be the same if there is no ADDITIONAL supporting eveidence of one being truly the more likely one. Example hearing glass shatter. Could either be cat or burglar no other supporting eveidence.

Question 12

When will the more unlikely hypotheses reach ST?

Answer 12

rates of rise are randomised and they’re randomised
independently for each hypothesis.
So even though we would expect most times, the cap
hypothesis is going to win If we ran this experiment
many, many times there was a sound downstairs. Occasionally, we might expect you to decide in favour of
the burgle hypothesis simply because the rate of rise of
the burglar one happened to randomise. although generally it is less likely to occur.

Question 13

What happens if we change evidence coming in with stimulus?

Answer 13

For example evidence that overrides the more likely hypotheses will result in a much faster rise of signal to S threshold meaning that even though hypotheses is unlikely it will override other more likely hypotheses as there is now more evidence. Hence decision signal will reach threshold much faster.

Question 14

what will happen if urgency is increased?

Answer 14

Remember, urgency changes If we increase urgency, that decreases that
threshold.
But that decision need signal needs to rise to
before we make that decision that’s gonna speed our reaction
times.
And so that’s gonna leave us less time to accumulate
evidence.
And so that means that our decisions are gonna be
more driven by these starting expectations because, um, they’re gonna
be more heavily biassed to what our pre preconceived ideas were.

Question 15

What happens when we increase the probability of a hypothesis?

Answer 15

Latency will decrease. This is shown as a function of logarithmic probability.

Question 16

Why gratitious randomisation?

Answer 16

Even if input isi known output is not. In some cases can allow survival, learning from one’s mistakes. outwitting opponenets. So all our our decisions have some chance of reaching threshold.
But these are all going to have evidence coming
in in various support for these.These are all gonna have randomised rates of rise.
Whichever one wins the race is the thing that I
decide to do.Note that based on these biases, if we have input coming into these decision signals, that is in support of
all of these equally, we would expect A to win
the race on average. But it won’t all the time because of that randomization.
Sometimes B will win very occasionally.
F will win.
F will win more often if we have evidence coming in. That’s highly in support of F and less in support of others. So it seems that LATER can explain variability in our action times via this gratuitous randomization.

Question 17

What other things may cause latency variability?

Answer 17

maybe this variability we’re getting in reaction times is because sensory signals are noisy it might take longer for the accumualtion of this noisy signal to diverge from background noise. So perhaps might be even more so the case when we are dealing with targets that are very hard to see.

Question 18

What are models that explain the variability in latency?

Answer 18

random walk and diffusion models

Question 19

Random walk. How is it different from LATER?

Answer 19

Later has this just straight linear rise to threshold. Here we’ve got this random walk.
Sometimes it’s going in one direction.
Sometimes it’s going the other on average.
Over time, it’s heading in a particular direction.
But at any given time it can be going in,
uh, various different directions. Ultimately signal will acumualte and reach threshold. Whichever signal reaches threshold is the one the decison that will prevail.

Question 20

If we have two hypotheses, one, due to evidence is more positvely skewed and the other is not what will happen?

Answer 20

the first one will reach threshold first as there will be more signals in support of it due to bias whereas the other will not. This is because of the variability will cause more fluctuations in in signal.

Question 21

So what happens when we decrease contrast of target?

Answer 21

Reaction times increase. Because we have a weaker
signal relative to the noise in our neural
system, it’s gonna take a lot longer to accumulate that
evidence until it reaches a particular, uh, evidentiary threshold.

Question 22

What is two stage model?

Answer 22

comprised of two sequential stages, whose latencies sum: - first stage = random walk
- second stage = LATER

Question 23

what do the two stage represent?

Answer 23

stages can be thought of representing
- detection (collection of evidence), &
- decision (“verdict” based on evidence), respectively

The first stage simply detecting Is there some in the
case of vision, some visual element there.
Is there a target there?
The second stage is the decision making one that says
given there’s a target there, what am I gonna do
with it?

Question 24

How do they work together?

Answer 24

once first stage (random walk) complete, unit “raises its hand” (effectively saying “something is here!”) & provides a constant input signal to second (LATER) stage to decide what to do with this information
* when stimulus difficult to detect: first-stage takes most time, & so dominates overall behaviour
* when stimulus easy to detect: second-stage takes most time

First Stage: response to low-level features (colour, contrast, local configuration, etc) via random-walk
**Second Stage: **integrates constant afferent signal from first stage, generating linear rise-to-threshold (LATER).

Question 25

Why 2 stages?

Answer 25

• when contrast & prior probability simultaneously manipulated their effects appear to add linearly & independently
  [Carpenter (2004). Current Biology 14: 1576-1580] [Taylor, Carpenter & Anderson (2006). J Physiol 573: 741-751]
  “Philosophical” Evidence
• low level visual features may constitute evidence for competing hypotheses:
  – inappropriate for global decision parameters (e.g. expectation, urgency, etc) to act at low-level detection stage, therefore
  – global decision parameters (e.g. expectation, urgency, etc) influence second stage only

Question 26

What research showed that we need two stages?

Answer 26

Experiment showing model if we had t wo stages. This was compared with observed data. If we just had a later or random walk we wouldn’t be able to
model what’s going on here.
So that provides some empirical evidence that, um, that this
two stage model, um, is required.

Question 27

What phase domianates when exposed to high contrast targets? How are two phases each contributing to variability of reaction times?

Answer 27

What we find is that initial visual detection phase
actually does the time it takes for these, **
the visual discrimination to take does correlate with sacadic reaction
time a little bit like what we were talking about
in that two stage model.
When we have low contrast targets, we’d expect the first
just, uh, visual detection phase to dominate behaviour compared to
the later stage and the opposite in when we have
a very easily detectable task that’s gonna be done very
quickly.
The thing that’s gonna dominate the variability of our reaction
times is going to be that subsequent decision making or
later phase.