Topic: non-linearly separable functions

Model neurons are often considered as simple linear threshold units. However, the diversity and cell-type specificity of morphoelectric properties of dendritic trees, as well as observations of active integration of inputs to dendrites, suggest that individual cell types might have non-linear processing leading to more complex computational properties. The limited number and distribution of such non-linear processes put neurons in an intermediate category between a simple linear threshold unit and a large deep network with millions of units. In this work, we examine bounds on the complexit