Prev: Short Term Synaptic Dynamics (RSE) Contents Next: Synaptic waveform (PSG)

Chapter 9: Long Term Synaptic Dynamics

Long Term synaptic dynamics or Hebbian learning also determines the effectiveness of a synapse. Depending on when a synapse's pre-synaptic spike occurs, there can be positive learning (the USE value is increased) or negative learning (the USE value is decreased). A time window is placed around the resulting post-synaptic spike of a cell. If the pre-synaptic spike occured before the post-synaptic spike, it was in the positive learning portion. If the pre-synaptic spike occurs after the post-synaptic spike, it was in the negative learning window. Pre-synaptic spikes occuring at the same time as post-synaptic spikes or occuring outside either window result in no change to the USE value.
KeywordValueDescription
SYN_LEARNING
N/A
Indicates the start of a Learning definition section
TYPE
name(string)
The name which will be used to refer to this object
SEED
value(integer)
The random number generator will use value as the seed
LEARNING
type(string)
The type of learning to be used. Valid learning keyword types are NONE, +HEBBIAN, -HEBBIAN and BOTH
LEARNING_SHAPE
type(string)
The shape of modification curve for learning. Valid keyword types are TRIANGLE(by default), EXPONENT. Exponential function is given as exp(-t/tau)
NEG_HEB_WINDOW
mean(real)
stdev(real)
The length of time (in seconds) used in calculating the negative learning window
POS_HEB_WINDOW
mean(real)
stdev(real)
The length of time (in seconds) used in calculating the positive learning window
POS_HEB_PEAK_DELTA_USE
mean(real)
stdev(real)
The maximum change in USE due to firing within the positive learning window
NEG_HEB_PEAK_DELTA_USE
mean(real)
stdev(real)
The maximum change in USE due to firing within the negative learning window
POS_HEB_PEAK_TIME
mean(real)
stdev(real)
Units in second. For TRIANGLE shape, it's the time of peak positive learning; for EXPONENT shape, it's the decay constant (tau) of exponential function. Used with POS_HEB_PEAK_DELTA to compute the amount of learning at other times
NEG_HEB_PEAK_TIME
mean(real)
stdev(real)
Units in second. For TRIANGLE shape, it's the time of peak negative learning; for EXPONENT shape, it's the decay constant (tau) of exponential function. Used with NEG_HEB_PEAK_DELTA to compute the amount of learning at other times
END_SYN_LEARNING
N/A
Indicates the end of a Learn definition section

Example

SYN_LEARNING
        TYPE                    0HEBB
        SEED                    999999
        LEARNING                BOTH
        LEARNING_SHAPE          EXPONENT
        NEG_HEB_WINDOW          0.1             0.0
        POS_HEB_WINDOW          0.1             0.0
        POS_HEB_PEAK_DELTA_USE  0.005          	0.0
        NEG_HEB_PEAK_DELTA_USE  0.0055         	0.0
        POS_HEB_PEAK_TIME       0.02           	0.0
        NEG_HEB_PEAK_TIME       0.02           	0.0
END_SYN_LEARNING

This example gives a double exponential STDP modification function. Maximum amounts of synaptic modification are 0.005 for +HEBBIAN and 0.0055 for -HEBBIAN.
Both curves have the same decay constant as 0.02 second, but tails are cut off to zero at 0.1 second time window. 

SYN_LEARNING
        TYPE                    0HEBB
        SEED                    999999
        LEARNING                +HEBBIAN
        LEARNING_SHAPE          TRIANGLE
        NEG_HEB_WINDOW          0.05            0.0
        POS_HEB_WINDOW          0.05            0.0
        POS_HEB_PEAK_DELTA_USE  0.005          	0.0
        NEG_HEB_PEAK_DELTA_USE  0.005         	0.0
        POS_HEB_PEAK_TIME       0.01           	0.0
        NEG_HEB_PEAK_TIME       0.01           	0.0
END_SYN_LEARNING

This example gives a triangular STDP modification function. Maximum amounts of synaptic modification are 0.005 for +HEBBIAN. +HEBBIAN triangle peaks at time 
0.01 second time window. -HEBBIAN is not used in this case.
Prev: Short Term Synaptic Dynamics (RSE) Contents Next: Synaptic waveform (PSG)
Last updated Friday 11/2/2007