Biomedical Engineering Reference
In-Depth Information
11.6
STDP in a network
In this section, we will consider several network models using STDP. However, be-
fore we dive into the models, let us consider some general characteristics of STDP in
the network setting. We have already noted in the previous section that STDP prefers
inputs with shorter latencies, so if there are multiple pathways conveying similar in-
formation feeding into oa cell in the network, the pathway with the shortest latency
will eventually dominate. Secondly, temporally asymmetric STDP discourages the
formation of mutually excitatory loops. If neuron A is predictive of the firing of neu-
ron B, neuron B cannot be predictive of the firing of neuron A and must lag behind.
Therefore, if the synapses from neuron A to neuron B are strengthened, the synapses
from neuron B and neuron A will be weakened, making a mutually excitatory con-
figuration unstable. However, under some circumstances, a temporally asymmetric
STDP could switch to a temporally symmetric STDP if the firing rates are suffi-
ciently high, which could explain why many mutually excitatory loops nonetheless
exist where temporally asymmetric STDP has been observed [78]. Experimental
studies of STDP in a network have been carried out by Bi and Poo [7], highlighting
the sensitive of patterns of firing in networks to the specific timing of the inputs.
However, also apparent from these studies, studies of STDP in a network will es-
pecially difficult because of the intricate interplay of dynamics of network activity
and the sensitivity to timing of STDP itself. The computational meaning of the net-
work activities also needs to be investigated. It seems that a coherent computational
framework is especially lacking here and will be a fruitful area of investigation.
11.6.1
Hebbian models of map development and plasticity
Since Hubel and Wiesel's pioneering studies of monocular deprivation [39], forma-
tion and alteration of columns and maps have been one of the favorite models for the
study of activity-dependent changes in cortical circuits. It is widely recognized that
activity is critical for refining synaptic connections to give adult patterns of connec-
tivity and function (for reviews, see [17, 41, 100]). In development, spontaneous
correlated activity is present from an early stage. Examples include waves of activity
that propagate in the retina and LGN [27, 31, 56, 63, 65, 96, 99]. Later on, sen-
sory inputs further refine the synaptic connections for cortical circuits [41, 81, 100].
Alterations of spontaneous and sensory induced activity can change both the degree
of formation and the specific shape of cortical maps [39, 76, 94]. Mechanisms of
synaptic modification include both changes in synaptic strengths and sprouting of
new synapses. Changes in synaptic strengths have been linked to activity through
mechanisms like LTP and LTD [2, 3, 9, 10, 50]. More recently, studies of spike-
timing dependent plasticity have provided a more direct link between activity and
modification of synaptic strength [5, 21, 23, 26, 36, 46, 49, 51, 78, 101]. On the
other hand, the local release of neurotrophins and other molecules has been pro-
posed to translate patterns of activity into patterns of synaptogenesis and neuronal
Search WWH ::
Custom Search