Dendritic computation

Dendritic computation is the realization that neurons are not simple point processors that sum synaptic inputs and fire when a threshold is crossed. Instead, the dendritic tree — the branching input structure of a neuron — performs complex, nonlinear computations that transform synaptic input patterns before they reach the soma. This discovery has overturned the classical integrate-and-fire model and replaced it with a view of neurons as multi-compartmental computing devices.

The key operations performed by dendrites include:

Sublinear summation: For widely separated synapses, inputs sum less than linearly due to cable properties — the dendrite acts as a passive attenuator.

Supralinear summation: For clustered, synchronous inputs, NMDA receptor activation produces a regenerative depolarization that amplifies the input beyond what linear summation would predict. This is the mechanism underlying coincidence detection in pyramidal neurons.

Branch-specific output: Individual dendritic branches can generate local spikes that may or may not propagate to the soma. A neuron with multiple basal dendrites and an apical tuft can therefore respond differently to the same total synaptic input depending on how that input is distributed across its dendritic arbor.

The computational consequence is that a single neuron can implement functions that would require multi-layer networks in standard artificial neural networks. The dendritic tree provides a spatial dimension to neural computation that is absent from most machine learning architectures, suggesting that the gap between biological and artificial intelligence may be wider than commonly assumed.