Human
s can learn a
ssociation
s between vi
sual
stimuli and motor re
spon
se
s from ju
st a
single in
struction. Thi
s i
s known to be a fa
st proce
ss, but how fa
st i
s it? To an
swer thi
s que
stion, we a
sked participant
s to learn a briefly pre
sented (200 m
s)
stimulu
s-re
spon
se rule, which they then had to rapidly apply after a variable delay of between 50 and 1300 m
s. Participant
s showed a longer re
spon
se time with increa
sed variability for
short delay
s. The error rate wa
s low and did not vary with the delay,
showing that participant
s were able to encode the rule correctly in le
ss than 250 m
s. Thi
s time i
s clo
se to the fa
ste
st
synaptic learning
speed deemed po
ssible by diffu
sive influx of AMPA receptor
s. Learning continued at a
slower pace in the delay period and wa
s fully completed in average 900 m
s after rule pre
sentation on
set, when re
spon
se latencie
s dropped to level
s con
si
stent with ba
sic reaction time
s. A neural model wa
s propo
sed that explain
s the reduction of re
spon
se time
s and of their variability with the delay by (i) a random
synaptic learning proce
ss that generate
s weight
s of average value
s increa
sing with the learning time, followed by (ii) random cro
ssing of the firing thre
shold by a leaky integrate-and-fire neuron model, and (iii) a
ssuming that the behavioural re
spon
se i
s initiated when all neuron
s in a pool of
m neuron
s have fired their fir
st
spike after input on
set. Value
s of
m=2 or 3 were con
si
stent with the experimental data. The propo
sed model i
s the
simple
st
solution con
si
stent with neurophy
siological knowledge. Additional experiment
s are
sugge
sted to te
st the hypothe
si
s underlying the model and al
so to explore forgetting effect
s for which there were indication
s for the longer delay condition
s.
This article is part of a Special Issue entitled Neural Coding 2012.