Overview of Brain Structure

- Telencephalon – neocortex- Diencephalon – limbic system- Mesencephalon – midbrain- Metencephelon – brainstem, cerebellum- Myelencephalon – spinal cord

Neocortex

- Cerebrum – 2 hemisphereso Corpus callosum

- Lobeso Occipitalo Temporalo Parietalo Frontal

- Mattero White – axons of neurons, tractso Gray – cell bodies

Lobes

- Occipitalo Visual cortex

- Parietalo Somatosensoryo Spatial orientation

- Temporalo Auditory cortexo Speech understandingo Visual, object/face recognitiono Autobiographical memory

- Frontal, prefrontalo Working memoryo Planningo Cognitive assessment of emotiono Personalityo Judgment and decision makingo Motor, pre-motor etc…o Most recently evolved part of the brain

Limbic system – Mammalian Brain

- Hippocampus – memory- Cingulate gyrus – brain wrapped around the corpus colosum - Entorhinal cortex

- Thalamus – sensory relay- Amygdala – emotion detection/response (esp. fear)

Midbrain and Brain Stem – Reptilian Brain

- Oldest part of the braino Reptilian > 500 million years old

- Of note for this courseo See a number of subcortical sensory systems

Superior colliculus, optic tectum Inferior colliculus Midbrain structures in the auditory system

Cerebellum – the “little brain”

- Function?? o Balance and rhythm (and a lot others, see Reading 5 for more information)

Generic Neuron

- Everything is done in approximately 10^11 in the human brain- Over 10,000 different kinds of neurons

Parts of a generic neuron

- Soma – cell body- Axon – up to a meter long- Dendrites – many (receivers of the electrical current)- Transmit information from one point in the brain to another, can be close or far.

Action Potentials

- Signal must be continuously amplified to travel without losing signal.

Nerve impulse – Action Potential

- Resting potential – 70 millivoltso Sodium Potassium outside of the neurono Na+ K+ protein on the insideo Na+ constantly being pumped out to maintain -70mv

Sodium pump

Synapse

- End of axono Synaptic vesicles filled with neurotransmittero Fuse with the side of the membraneo Spill contents into synaptic cleft

Time Course of Action Potential

- Refractory periodo 1 millisecondo Can fire 800-900 times per second

Properties of Soma and Dendrites

- Graded potential- Decremental conduction(unlike axons)- Temporal and spatial summation

Properties of axons

- Non-graded petoential (fires all or none)o Threshold

- Non-decremental conduction(gets to the end in the same form that it started)- More intense firing -> more spikes per second

Summary of the Generic Neuron

- Partso Soma, dendrties, axon

- Statisticso 10 to the eleventh of themo 10,000 different kinds

- Physiologyo Conduction of nerve impulses, action potential, fire

- Functiono Transmits a signalo Intensity of signal related to firing rate

Receptive field

- Definition – part or aspect of the world that can cause a change in the firing rate of a neuron

Levels of the Receptive Fields in the Brain

- Most peripheralo Receptor connected to a neuron

- Higher upo Neuron in a brain region that receives a reasonably direct projection from the “outside”o Takes only one firing of neuron. Meets up with spinal cord.

- Higher still…o No direct link to sensory world…

Face cells..? Hand cells….? (some neurons in monkeys only respond to hands) Invariance to retinal input?

Modalities of Receptive Fields in the Brain(neurons firing at certain things)

- Vision

o Hueso Special quality, line orientationso Motion – direction, speed

- Auditiono Frequencyo Spatial location…?

- Toucho Temperatureo Paino pressure

- Chemical senses

3 Practical Questions

- What can a neuron do? (Fire signals)- What can a neuron know?- How can a neuron learn?

What can a neuron do?

- Compute a threshold function of a spatial and temporal integration(add together information spatially and temporally, compares integration to a certain number(threshold) and fires depending on outcome)

- Spatialo Which dendriteso Extent and direction of dentrites

- Temporal time window of refractory periodo Time window of refractory period(has to have enough firing to happen within a time

period)

How can this be useful?

- Theoretical model of a neuron (McCulloch & Pitts, 1943)o A logical calculus of the ideas immanent in nervous activity

- 1. Neuron = binary device with (binary) input (not exactly true)o Excitatory inputs add linearly o Inhibitory inputs prevent the neuron from firing

- 2. Neuron has fixed threshold- 3. Neuron has binary output (somewhat true, it either fires or doesn’t, but has to have a certain

amount of activation to fire.)

Results and Implications

- By combining logical propositions into networkso Any finite logical expression can be realized

- Paper had little effect in neuroscience literature- Paper had enormous effect in computer science

o Binary operations, logic gates, computation…etc

What can a neuron know?

- How is one neuron different than another?- Anatomically – hardware

o Number and extent of dendriteso Spatial outlay of the dendrites (receptive field)

- Physiologically – softwareo Physically identical --- how can they differ?

Lateral inhibition

- Hartline, Wagner, & Ratliff (1956- Horseshoe crab- Physiology- Did a lot of research about them because their axons are so big

o Light on cell A(in eye), recorded cell, lots of activityo Same with cell Bo Light on both, record from A, A’s firing rate goes down a lot.

What can a neuron learn?

- Hebb (1949) – Organization of Behavior- Modification of the synaptic efficacy of dendrites are the neurons ability to learn- Changes in the values of the w1 values of the model

Simple Operation Rules

- Set all weights to random values- A set of input patterns to classify (train one to detect A, but not other letters)- A set of targets “goals” or categories of output- Apply first input pattern and compute output - O = Ewx- If it exceeds threshold, it fires, if it doesn’t, it doesn’t fire- If correct – do nothing- If incorrect

o If too high w <= w-cx (change weight)o If too low w <= w+ cx (boost weight)(if you put letter up and it did not fire)

(Neurons will fire for certain things, and if they fire for something they aren’t supposed to fire for, they turn the weights down on those certain neurons. This repeats until the neurons that are supposed to fire, fire.)

Results

- Neural network/perceptrono Categorizes inputo i.e., detects instances of a category

o limits linearly separable

Summary

- parallel computationo time steps

- distributed and non-rule-governedo emergent order

- analysis of the weightso statistics