Abstract
Spiking Neural Networks are recognized as the third-generation neural networks. They are biologically inspired and energy efficient, making them suitable for underwater data analysis tasks. A common implementation of SNNs is the leaky integrate-and-fire, or LIF, neuron, which integrates input over time. Another implementation, proposed by Eugene M. Izhikevich, uses intrinsic parameters to control the neuron and its reset. Both models can have limitations in regression tasks, however, limiting accuracy. In this paper, we proposed a Hybrid model, combining aspects of the LIF and Izhikevich neuron. Experimental results on oceanographic data found that our proposed method achieves competitive accuracy and improves CPU time compared to the baseline models. The Hybrid model has lower L1 loss compared to existing models, which means that the hybrid model is good at predicting sparse oceanographic data. Future work on this model could include introducing adaptivity to the tuning parameter.