Recency vs Primacy-- an ongoing project

Nov 17th 2009

Juan Gao

People

Question

• What is the mechanism underlying perceptual decision making in time-controlled paradigm?

I1 I2

X1 X2

Leaky competing accumulators

Usher and McClelland 2001

Two successful models

Accumulation to the bound

Ratcliff 1978, 1999, Kiani et.al.2008,

t

x

How are they different

• In ATB– earlier > later --primacy.

• In LCA– earlier > later if inhibition> leak --primacy; – later > earlier if leak>inhibition --recency.

See also a theoretical study by Zhou, Wong-Lin and Holmes 2009

Usher McClelland 2001

H S H H H S H S H H S H S S S S

A sequence of 16 H and S letters flashing one by one. Are their more Hs or Ss?

leak > inhibition inhibition > leak

Kiani, Tanks and Shadlen 2008

Random dots. Time controlled.

Stimulus duration = exponential distribution.

‘go’ cue followed by 300ms response window.

Earlier pulse matters more

Earlier pulse matters more

Two monkeys? Earlier > Later for all subjects?

Earlier > Later in all moving dots experiments?

If no, what determines it?If Yes, ATB

Ongoing Experiment

• Random dot motion stimuli, following the procedure in Kiani et.al.

• Multiple coherences, [6.4, 12.8, 25,6, 51.2]. But for figures in this talk, we collapse data across coherence levels.

• Three participants per experiment, each run for up to 25 sessions

• Ongoing recruitment, Ongoing analysis…

The experiments

0. Repeat Kiani 2008

1. Same question, different experiment setup.

2. Release the time pressure.

Experiment 1Stimulus Duration

1) Early

2) Late

3) Constant

4) Switch

Results in Exp.1

0.2 0.4 0.6 0.8 1 1.2 1.4 1.60.4

0.5

0.6

0.7

0.8

0.9

1

Time (s)

Acc

urac

y

switchearlyconstantlate

CS

Results in Exp.1

0.2 0.4 0.6 0.8 1 1.2 1.4 1.60.4

0.5

0.6

0.7

0.8

0.9

1

Time (s)

Acc

urac

y

switchearlyconstantlate

CS

Results in Exp.1

0.2 0.4 0.6 0.8 1 1.2 1.4 1.60.4

0.5

0.6

0.7

0.8

0.9

1

Time (s)

Acc

urac

y

switch

earlyconstant

lateMT

Results in Exp.1

0.2 0.4 0.6 0.8 1 1.2 1.4 1.60.4

0.5

0.6

0.7

0.8

0.9

1

Time (s)

Acc

urac

y

switch

earlyconstant

lateMT

Results in Exp.1

0.2 0.4 0.6 0.8 1 1.2 1.40.4

0.5

0.6

0.7

0.8

0.9

1

Time (s)

Acc

urac

y

switchearlyconstantlate

SC

Results in Exp.1

0.2 0.4 0.6 0.8 1 1.2 1.40.4

0.5

0.6

0.7

0.8

0.9

1

Time (s)

Acc

urac

y

switchearlyconstantlate

SC

Take home message

• Yes, it seems earlier > later in all three subjects with this time pressure.

The experiments

0. Repeat Kiani 2008

1. Same question, different experiment setup.

2. Release the time pressure.– Stimulus duration: exponential uniform;– Response Window: 300ms 1 s.

Results in Exp.2, without time pressure

0.2 0.4 0.6 0.8 1 1.2 1.4 1.60.4

0.5

0.6

0.7

0.8

0.9

1

Time (s)

Acc

urac

y

switch

earlyconstant

lateMM

Results in Exp.2, without time pressure

0.2 0.4 0.6 0.8 1 1.2 1.4 1.60.4

0.5

0.6

0.7

0.8

0.9

1

Time (s)

Acc

urac

y

switch

earlyconstant

lateMM

Results in Exp.2, without time pressure

0.2 0.4 0.6 0.8 1 1.2 1.4 1.60.4

0.5

0.6

0.7

0.8

0.9

1

Time (s)

Acc

urac

y

switch

earlyconstant

lateWW

Results in Exp.2, without time pressure

0.2 0.4 0.6 0.8 1 1.2 1.4 1.60.4

0.5

0.6

0.7

0.8

0.9

1

Time (s)

Acc

urac

y

switch

earlyconstant

lateWW

Results in Exp.2, without time pressure

0.2 0.4 0.6 0.8 1 1.2 1.40.4

0.5

0.6

0.7

0.8

0.9

1

Time (s)

Acc

urac

y

switchearlyconstantlate

DG

Results in Exp.2, without time pressure

0.2 0.4 0.6 0.8 1 1.2 1.40.4

0.5

0.6

0.7

0.8

0.9

1

Time (s)

Acc

urac

y

switchearlyconstantlate

DG

Take home message

• Yes, it seems earlier > later in all three subjects with this time pressure.

• As time pressure gets released, earlier = later.

Take home message

• Yes, it seems earlier > later in all subjects with this time pressure.

• As time pressure gets released, earlier = later.

• Uniform distribution only long stimulus condition: later > earlier.

possible future direction

Take home message

• Yes, it seems earlier > later in all subjects with this time pressure.

• As time pressure gets released, earlier = later.

• Uniform distribution only long stimulus condition: later > earlier.

possible future direction

What this means to the models

• So far– LCA can account for the observations by

decreasing the inhibition.– ATB can do the same by raising the bound.

• When future is now, If later> earlier – LCA is more general.

• Is decision making a fixed process or does it depends on experiment setup?

Back up slides

0.2 0.4 0.6 0.8 1 1.2 1.40.4

0.5

0.6

0.7

0.8

0.9

1

Time (s)

Acc

urac

y

500ms-kg.mat

switchearlyconstantlate

0.2 0.4 0.6 0.8 1 1.2 1.40.4

0.5

0.6

0.7

0.8

0.9

1

Time (s)

Acc

urac

y500ms-jl.mat

switchearlyconstantlate

A theoretical study

Zhou, Wong-Lin and Holmes 2009

Usher McClelland 2001

H S H H H S H S H H S H S S S S

A sequence of 16 H and S letters flashing one by one. Are their more Hs or Ss?

Literature 1Drift Diffusion model: dx = A dt + noise. A is a constant

Zhou, Wong-Lin and Holmes 2009

Literature 1OU process: dx = (bx+A) dt + noise. Stable when b<0, unstable when b>0.

Zhou, Wong-Lin and Holmes 2009

Results in Exp 1. The pulse study

-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.60

0.2

0.4

0.6

0.8

1

-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.60

0.2

0.4

0.6

0.8

1

-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.60

0.2

0.4

0.6

0.8

1

-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.60

0.2

0.4

0.6

0.8

1

SC

-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.60

0.2

0.4

0.6

0.8

1

-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.60

0.2

0.4

0.6

0.8

1

-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.60

0.2

0.4

0.6

0.8

1

-0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.60

0.2

0.4

0.6

0.8

1

mt

Both successful models

Time (ms)

Usher and McClelland 2001