ABSTRACT
The performance of the MuSpiNN model and the Multi-SpikeProp learning
algorithm is evaluated using three increasingly difficult pattern recognition
problems: XOR, Fisher iris plant classification (Fisher, 1936; Newman et al.,
1998), and EEG epilepsy and seizure detection (Andrzejak et al., 2001; Adeli
et al., 2007; Ghosh-Dastidar et al., 2007, 2008; Ghosh-Dastidar and Adeli,
2007). The classification accuracy is computed only for the iris and EEG
datasets because they are large enough to be divided into training and testing
datasets. The transformation of real-valued inputs and outputs to discrete
spike times (output encoding) is different for the three problems and therefore
is addressed for each problem separately.