In the field of computational neuroscience, spiking neural network models are generally preferred over rate-based models due to their ability to model biological dynamics. Within AI, rate-based artificial neural networks have seen success in a wide variety of applications. In simplistic spiking models, information between neurons is transferred through discrete spikes, while rate-based neurons transfer information through continuous firing-rates. Here, we argue that while the spiking neuron model, when viewed in isolation, may be more biophysically accurate than rate-based models, the roles reverse when we also consider information transfer between neurons. In particular we consider the biological importance of continuous synaptic signals. We show how synaptic conductance relates to the common rate-based model, and how this relation elevates these models in terms of their biological soundness. We shall see how this is a logical relation by investigating mechanisms known to be present in biological synapses. We coin the term ‘conductance-outputting neurons’ to differentiate this alternative view from the standard firing-rate perspective. Finally, we discuss how this fresh view of rate-based models can open for further neuro-AI collaboration.
CITATION STYLE
Knudsen, M., Hendseth, S., Tufte, G., & Sandvig, A. (2019). Viewing Rate-Based Neurons as Biophysical Conductance Outputting Models. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 11493 LNCS, pp. 164–177). Springer Verlag. https://doi.org/10.1007/978-3-030-19311-9_14
Mendeley helps you to discover research relevant for your work.