SNNs are distinguished from traditional artificial neural networks by their use of temporal dynamics to simulate the activity of biological neurons. Each neuron in an SNN communicates by sending discrete, time-dependent signals (spikes) to other neurons, which only respond when their input exceeds a certain threshold, much like biological neurons. This allows SNNs to process information in a more biologically realistic manner, potentially leading to more power-efficient computations than conventional neural networks. SNNs are particularly suited for tasks involving temporal pattern recognition, such as speech and gesture recognition, and are also seen as a step towards creating more efficient hardware implementations of neural networks, such as neuromorphic computing.
The concept of spiking neural networks was first developed in the late 1990s, with significant attention from the computational neuroscience and artificial intelligence communities starting around the early 2000s. This interest was driven by the desire to better understand brain functions and to develop neural models that more closely mimic biological processes.
One of the key figures in the development of SNNs is Eugene Izhikevich, who formulated the Izhikevich model of neuronal dynamics that is widely used in these networks. Another significant contributor is Wulfram Gerstner, who has advanced theories related to the dynamics and computational functions of spiking neurons. Their work, among others, has been crucial in the theoretical and practical advances in SNN technology.