LeDoux – Chapt 3

• All mammalian brains share same organization• Neocortex and particularly telencephalon is larger

and more developed in primates and humans• The basic computational unit of all parts of the

brain is the neuron• Neurons receive inputs through dendrites and

communicate and send signals through axons• The junction between cells, the synapse, is the

site of neural plasticity

LeDoux – Chapt 3- continued

• Neurons connect in circuits– Circuits, as they traverse through the brain, have a

hierarchical organization (e.g., retina, LGN, visual cortex)– Local circuits are points of lateral communication that

function to inhibit (reduce) or excite (intensify) signals• Systems are functional networks of circuits– Sensory systems (vision, hearing, touch)– Emotional circuits (fear)– Systems link many areas of the brain and generally have

cortical and subcortical components

LeDoux – Chapt 3- continued

• At synapses, local communication is handled by glutamate (excitatory) and GABA (inhibitory)– Glutamate receptors synapse on synaptic spines– GABA synapse on cell bodies

• Drugs such as Valium enhance GABA’s ability to regulate glutamate

• Neuromodulators are chemicals such as peptides, amines and hormones– Alter a cells responsiveness (ex. opiates, monamines in arousal)– Prozac alters availability of neuromodulator serotonin, by

preventing removal from synapse– Endocrine glands modulate synaptic activity through hormones

LeDoux – Chapt 3- continued

• The fear system involves circuits that course through the amygdala

• Innate or learned threat stimuli are routed through the amygdala

• An inhibitory GABA gate is normally closed preventing sensory signals from activating fear responses

• Threat signals open the gate activating fear responses• Anxiety disorders arise when gate opens to signals

excessively• Valium and Prozac enhance GABA activity permitting the

gate to close

Churchland, Chapt 2

• The inputs of neural circuits are composed of input vectors

• An identifiable stimuli has a unique activation pattern

• Vectors and be graphed in vector space (exs., taste space, color space)

• Psychological properties emerge in vector space (conceptual categories, opposites)

Churchland, Chapt 2

• Species differences in sensitivity occur because of the number of values that can be encoded on the dimensions of the input vector, creating a much larger representational space

• Vector coding occurs at all hierarchical levels• Facial coding emerges from higher order

coding in the visual system

Emergence of Concepts

• The prototypical face– Average is mid-point on dimensions that compose

the face vector• All faces tend to lie at some distance from the

mid-point of these dimensions– A hyperbolic representation is built by altering a

face on the various dimensions of face space– Morphing between two faces is movement along a

straight line between two points in face space

Psychological qualities of taste emerge in taste space

Wenner reading (R02)

• What is a taste modulator?• Where are taste receptors for different

modalities found?• What may be the basis of individual

differences in taste sensitivity?• What is the robot taste tester?• How does taste tester technique permit the

development of chemical “flavors”.

Neural Signals Travel from the Retina to Several Brain Regions

Note how light space codes psychological properties of similarities and opposites. Higher order concepts are emergent from this organization.

Prototype Research

Hawkins, Chapt 3

• On Intelligence focuses exclusively on telencephalic circuits – primarily the neocortex

• Goal – not to duplicate the mammalian brain / rather to understand the architecture of intelligence

• Central architectural properties (what gives rise to your conscious experience and distinctly human mental attributes)– 2 mm thick, 6 layers– Contains about 30 billion neurons

Neocortex is divided into lobes

View illustrates neocortex relative to subcortical processing areas – note position of

diencephalon (thalamus and hypothalamus)

Brodman numbers identify different sub-regions of 6 layer neocortex

The 6-layers of the neocortex- Hawkins will have a lot to say about this!

Labeled cortex illustrating networks of interconnectivity and also columnar organization

Hawkins, Chapt. 3

• Mountcastle’s organizing principle– Differences in region are due to connections, NOT function

• All sensory systems work in the same way• Accounts for plasticity of the brain- ex. Wiring different

modalities into a cortical region• Spatial patterns• Temporal patterns• Spatial and temporal patterns in vision, hearing and

touch• World coded in patterns, which become expectations

Mountcastle’s Body Map

Somatotopic Organization – The Hommunculus

Retinotopic Organization

The Multiplicity of Visual Maps- Hierarchical Networks/Parallel Processing

Tonotopic Organization

Multiple Maps in the Motor System

Facial Recognition Cells in Inferotemporal Cortex

Circuity for Object Detection

Principles Applied to Facial Recognition

Facial Recogniton – using a middle layer (related to XOR)

Inputs [Training set]

Principle Underlying Network Organizaton

Error Correction via Backpropagation

Middle Layer Representation of an Occluded Face – see page 15!

Holons:Stimuli that get largest response from middle layer cells- the dimensions of the facial vector

Categories (concepts) spontaneously developing in Face space

Churchland- Properties of a Trained Network

• Network is not “memorizing”• Capable of discriminations (male vs. female)• Recognizes familiar faces – though obscured• Conceptual categories are derived from

experience– Cultural differences in facial recognition

• Inductive inference through vector completion• Partitioning of vector space into “basins of

attraction” – available conceptual categories