Abstract
Spike-Timing Dependent Plasticity (STDP) is characterized
by a wide range of temporal learning patterns, depending on
the studied network and the experimental conditions. These
characteristic patterns represent the change in the measured
post-synaptic potentials or currents as a function of the forced
time interval (Δt) between pre- and postsynaptic spikes during
the experiment. Long Term Potentiation (LTP) is marked by a
positive sign of the characteristic function and Long Term
Depression (LTD) by a negative sign. It is a common practice
to define this function in segments of the time interval –
typically in two segments, one for positive Δt (the causal
branch) and the other for negative Δt (the acausal branch).
Here we suggest a model in which this pattern is constructed
from a superposition of two separate processes one for the LTP
and the other for the LTD.We approximate these two functional
branches using a continuous non-segmented “probability like”
function that captures the essential features of the STDP. We
demonstrate how the various experimentally observed STDP
temporal structures can be obtained by a gradual change of a
single continuous parameter in our model. Analysis of the
STDP dynamics reveals a critical point. Below this critical point
the STDP dynamics is governed by a negative feedback and the
synaptic weights are characterized by a unimodal distribution.
Above this point, the stability of the STDP dynamics is governed
by the synaptic weight dependence of the STDP rule. In
the latter case there is a different parameter with a critical value,
above which, a bimodal synaptic weight distribution exists.We
show that the location of these critical points depends on
general properties of the temporal structure of the STDP rule
and not on its fine details. These results hold for both excitatory
and inhibitory synapses. The symmetry in the learning dynamics
of excitatory and inhibitory synapses is discussed.
by a wide range of temporal learning patterns, depending on
the studied network and the experimental conditions. These
characteristic patterns represent the change in the measured
post-synaptic potentials or currents as a function of the forced
time interval (Δt) between pre- and postsynaptic spikes during
the experiment. Long Term Potentiation (LTP) is marked by a
positive sign of the characteristic function and Long Term
Depression (LTD) by a negative sign. It is a common practice
to define this function in segments of the time interval –
typically in two segments, one for positive Δt (the causal
branch) and the other for negative Δt (the acausal branch).
Here we suggest a model in which this pattern is constructed
from a superposition of two separate processes one for the LTP
and the other for the LTD.We approximate these two functional
branches using a continuous non-segmented “probability like”
function that captures the essential features of the STDP. We
demonstrate how the various experimentally observed STDP
temporal structures can be obtained by a gradual change of a
single continuous parameter in our model. Analysis of the
STDP dynamics reveals a critical point. Below this critical point
the STDP dynamics is governed by a negative feedback and the
synaptic weights are characterized by a unimodal distribution.
Above this point, the stability of the STDP dynamics is governed
by the synaptic weight dependence of the STDP rule. In
the latter case there is a different parameter with a critical value,
above which, a bimodal synaptic weight distribution exists.We
show that the location of these critical points depends on
general properties of the temporal structure of the STDP rule
and not on its fine details. These results hold for both excitatory
and inhibitory synapses. The symmetry in the learning dynamics
of excitatory and inhibitory synapses is discussed.
| Original language | American English |
|---|---|
| Pages (from-to) | s76-s76 |
| Number of pages | 1 |
| Journal | Journal of Molecular Neuroscience |
| Volume | 51 |
| DOIs | |
| State | Published - Nov 2013 |