ion alre… is the cell body where inputs from all the dendrites come together, and based on all these signals a decision is made whether to fire an output (a “spike”). This is a bit of a generalisation, as some of the calculation already happens before the Soma, and …
I appreciate that you are pointing out the many obvious differences between neurons which are the components of biological neural networks (a theoretical structural/functional unit of the brain) and artificial neural networks. However, I do not believe you go nearly far enough and are using terminology which gives neurons and neural networks capabilities they do not and cannot possess in an attempt to draw “key similarities” where there are none.
First, there is no decision for a neuron is not capable of making decisions, and neither is a brain. Only of a (mostly) intact human person (and some non human animals) with a (mostly) fully functioning nervous system including a brain can we say they are capable of making decisions. To suggest otherwise is to commit the mereological fallacy. Second their is no calculation for a neuron cannot and does not calculate (non-linear or otherwise). Third to suggest something is “encoded” in the dendritic structure of the cell is also incorrect for it implies that these cells are capable of coding are being coded, which they are not. While it is interesting to draw comparisons between biological and artificial things it is not helpful when these comparisons ascribe capabilities to either system which they either do not or cannot possess.
Finally, plasticity or what you call the “key feature of the brain that enables memory and learning” is still theoretical, and far from proven. There are many alternate theories of memory and learning some of which have at least an equal probability of being ultimately correct. Even though these alternate theories are not featured in scads of books, or seen on Oprah, they are still legitimate science. Moreover, iteratively modifying the weights of the networks parameters based on batches of inputs as is done in many ANNs is an interesting mathematical approach to selecting a desired output. It is not learning, it is calculation and that is why ANN’s use this approach, they are computers, and thus are quite good at calculating. Biological neural systems are terrible at calculating because they do not calculate, and cannot calculate. Yet the later can learn while the former cannot.