Page 132 - Artificial Intelligence in the Age of Neural Networks and Brain Computing

3. Spiking Neural Networks (SNN) as Brain-Inspired ANN 121

• An evolving fuzzy neural network based on the mapping of similarities [87];
• Incremental learning by heterogeneous bagging ensemble [88];
• Fuzzy associative conjuncted maps network [89];
• EFuNN ensembles construction using CONE with multiobjective GA [90];
• Risk analysis and discovery of evolving economic clusters in Europe [91];
• Adaptive time-series prediction for financial applications [92];
• Adaptive speech recognition [93];
• and others [37].

ECOS methods and systems presented above use predominantly the McCulloch
and Pitts model of a neuron (Fig. 6.2A). They have been efficiently used for wide
range of applications as some of them listed above.

3. SPIKING NEURAL NETWORKS (SNN) AS BRAIN-INSPIRED
ANN

3.1 MAIN PRINCIPLES, METHODS, AND EXAMPLES OF SNN AND
EVOLVING SNN (ESNN)
The traditional ANN and ECOS discussed in the previous sections are the theoretical
bases on which the third generation of ANN was developed - spiking neural networks
(SNN). Spiking neural network (SNN) architectures use a spiking neuron model and
spike information representation. Spike information representation accounts for time
in the data and for changes in the data over time. This is where SNN can be chosen as
preferred methods and used efficiently.
A spiking neuron model receives input information represented as trains of spikes
over time from many inputs. When sufficient input information is accumulated in the
membrane of the neuron, the neuron’s postsynaptic potential exceeds a threshold and
the neuron emits a spike at its axon (Fig. 6.6).
Some of the state-of-the-art models of a spiking neuron include: early models by
Hodgkin and Huxley [94]; Spike Response Models (SRM); Integrate-and-Fire
Models (IFM) (Fig. 6.6); Izhikevich models; adaptive IFM; and probabilistic IFM
[95,96].

FIGURE 6.6
The structure of the LIFM of a spiking neuron.