cse 153modeling neurons chapter 2: neurons a “typical” neuron… how does such a thing support...

CSE 153 Modeling Neurons

Chapter 2: Neurons

• A “typical” neuron…

• How does such a thing support cognition???

QuickTime™ and a decompressor

are needed to see this picture.


Neurons: detectors?• The simplest idea is that a neuron

is a “detector” for a pattern.

• When it is stimulated by its pattern, it fires.

• The pattern could be thought of as in the “input” (but what is “input” to a neuron?)

• Or as the output of other neurons - patterns in the patterns…




Neurons as detectors




Neurons: Pandemonium?

Oliver G. Selfridge

1958 idea: The mind is made up of “demons”, who shout based on other demons shouting to them.

The loudest is heard…


Neurons: Pandemonium?

What model does this remind you of???


Neurons: Pandemonium?• Each neuron has a simple job, but

together...

• Layers of more and more complicated detectors.

• “B” is simple! What kind of detectors would be needed for language, face recognition, driving, algebra, problem solving, etc.?


How to simulate this?

• Neural activity (and learning) can be characterized by mathematical equations.

• We use these equations to specify the behavior of artificial neurons.

• The artificial neurons can then be put together to explore behaviors of networks of neurons.

• Simulation.


Basic Properties of a Neuron• It’s a cell: has a cell body, membrane, nucleus,

DNA, RNA, proteins, etc.• The cell membrane has channels (“holes”),

passing ions (salt water).• The cell has electrical potential (voltage),

integrated in cell body, which activates an action potential output in axon, which releases neurotransmitter.

• The neurotransmitter activates potential via dendritic synaptic input channels.

• Excitation and inhibition are transmitted by different neurons!


The Synapse


The Synapse

• Synaptic efficacy = the amount of activity of the presynaptic (sending) neuron communicated to the postsynaptic (receiving) neuron.

• Presynaptic: # of vesicles released, NT per vesicle, efficacy of reuptake mechanism.

• Postsynaptic: # of receptors, alignment & proximity of release site & receptors, efficacy of channels, geometry of dendrite/spine.


Abstract Model neurons

1. Compute weighted, summed net input:

i = wij*xj + biasi

2. Pass this through a sigmoidal function to get output:

yj=1/(1+e-i)


“Biologically Plausible” Neural Nets

1. Compute weighted, summed net input (actually averages): ηj ≈ aiwij ≈ ge

2. Compute Vm:

3. Compute output as: Spikes

Or, rate code via a sigmoidal function:

aj =1/[1 + (γ[Vm(t) − Θ]+)−1]


Neurophysiology

The neuron is a miniature electro-chemical system:

1. Balance of electric and diffusion forces.

2. Principal ions.

3. Putting it all together.


Switch to pdf slides

cse 153modeling neurons chapter 2: neurons a “typical” neuron… how does such a thing support...

Documents