Neuron Models Flashcards
McCulloch-Pitts model - Analog Neurons
Have real-valued output that can be interpreted as the neuron’s firing rate. Simple linear layer: A(wT*x +b)- Activation functions
Spiking Neurons
Emit impulses that correspond to the
spikes emitted by biological neurons.
The Leaky Integrate-and-Fire Model - Spiking Neurons
Information processing in biological
neurons is largely driven by ionic currents
that are accumulated in the soma
(Zellkörper)
- The flow of these currents can be modeled
with functionally equivalent electrical
circuits (ionic currents are represented by
electrical currents) - The first spiking neuron model that is still
of high relevance today is the leaky
integrate and fire (LIF) model - LIF neurons are comprised of a simple RC
circuit that is driven by an external current
Levels of Modeling
Detailed compartmental models
Reduced compartmental models
Single-compartment models
Cascade models
Black-box models
Detailed compartmental models
capture the detailed neuron
morphology based on anatomical reconstructions. The continuous
geometrical structure of the neuron is discretized into compartments, each of which can be modeled as a point neuron
Reduced compartmental models
are comprised of only a few
dendritic compartments to decrease computational complexity
and to increase analytical tractability
Single-compartment models
completely ignore the morphology of
the neuron. Detailed spiking point neuron models capture the
flow of ion currents and spike generation.
Cascade models
are purely functional models that only represent abstract computations but no biological detail.
Black-box models
do not model any computational function at all but consider the neuron as an input/output system with defined
statistic behavior