Improved Izhikevich neurons for spiking neural networks

Abstract

Spiking neural networks constitute a modern neural network paradigm that overlaps machine learning and computational neurosciences. Spiking neural networks use neuron models that possess a great degree of biological realism. The most realistic model of the neuron is the one created by Alan Lloyd Hodgkin and Andrew Huxley. However, the Hodgkin-Huxley model, while accurate, is computationally very inefficient. Eugene Izhikevich created a simplified neuron model based on the Hodgkin-Huxley equations. This model has better computational efficiency than the original proposed by Hodgkin and Huxley, and yet it can successfully reproduce all known firing patterns. However, there are not many articles dealing with implementations of this model for a functional neural network. This study presents a spiking neural network architecture that utilizes improved Izhikevich neurons with the purpose of evaluating its speed and efficiency. Since the field of spiking neural networks has reinvigorated the interest in biological plausibility, biological realism was an additional goal. The network is tested on the correct classification of logic gates (including XOR) and on the iris dataset. Results and possible improvements are also discussed.

Are Non-Standard Neural Behaviors Computationally Relevant?

“…For the second experiment the weights are optimized through the use of a genetic algorithm. The genetic algorithm used for optimizing the weights had the exact same configuration as the one in (Kampakis, 2011): two populations that ranged from 50-100 members each, with crossover ratio 0.6 and 1 elite. The algorithm terminated after 150 generations had passed.…”

“…Each instance contains four attributes: sepal length, sepal width, petal length and petal width. Only sepal lengh and sepal width were used, like in (Kampakis, 2011) because the rest of the attributes are noisy.…”

“…The optimization stopped as soon as the genetic algorithm reached the upper limit of generations. The parameters of the neurons and the architecture were the same as the ones used by Kampakis (2011). Table 1 presents the results of the first experiment.…”

“…However, it could be the case that the flexibility of the model could help it adapt better to the task at hand. In fact, this model was used by Kampakis (2011) in the context of spiking neural network for a supervised learning task. That research provided some evidence that different neural behaviors, other than the ones assumed by the integrate-and-fire neuron, could be useful.…”

“…Architecture of the network used for the supervised learning task as it was presented in(Kampakis, 2011).…”