Computational Models for Generic Cortical Microcircuits - Computational Neuroscience: A Comprehensive Approach - page 578

"one", speaker 5 "one", speaker 3 "five", speaker 1 "eight", speaker 4

40

20

0

135

90

45

0

0 0.2 0.4 0 0.2 0.4 0 0.2

time [s]

0 0.2

time [s]

time [s] time [s]

Figure 18.4
Application of our generic neural microcircuit model to the speech recognition from [10]. Top
row: input spike patterns. Second row: spiking response of the 135 I&F neurons in the neural
microcircuit model. Third row: output of an I&F neuron that was trained to fire as soon as
possible when the word one was spoken, and as little as possible else. Although the liquid
state presented to this readout neuron changes continuously, the readout neuron has learnt to
view most of them as equivalent if they arise while the word one is spoken (see [16] for more
material on such equivalence classes defined by readout neurons).

for training, the other 200 for testing. The generic neural microcircuit model was
drawn from the distribution described in Section 18.3, hence from the same distribu-
tion as the circuit drawn for the completely different tasks discussed in Figure 18.2,
with randomly connected I&F neurons located on the integer points of a 15

3
column. The synaptic weights of 10 readout neurons f w which received inputs from
the 135 I&F neurons in the circuit were optimized (like for SVMs with linear ker-
nels) to fire whenever the input encoded the spoken word w . Hence the whole circuit
consisted of 145 I&F neurons, less than 1

× 3 ×

30 th of the size of the network constructed
in [10] for the same task. ∗∗ Nevertheless the average error achieved after training
by these randomly generated generic microcircuit models was 0.14 (measured in the
same way, for the same word one ), hence slightly better than that of the 30 times
larger network custom designed for this task. The score given is the average for
50 randomly drawn generic microcircuit models. It is about the same as the error

/

∗∗ If one assumes that each of the 800 large pools of neurons in that network would consist of just 5
neurons, it contains together with the b and c-neurons 5600 neurons.