Transcript Darwin VII after - Ohio University

Chapter 3.
Elsevier web materials.
To accompany Baars & Gage Chapter 3
1
Chapter 3.
Neurons.
• Teaching
materials.
• Powerpoints
with movies,
figures, and
major chapter
points.
• Study Guide
• Quiz items
Yun Bo Yi & Sastri, U Mich
Single neuron model - NSF
To accompany Baars & Gage Chapter 3
2
Basic parts of a neuron.
To accompany Baars & Gage Chapter 3
3
It's a jungle in there -Real neurons come in a rich variety of sizes, shapes and functions.
To accompany Baars & Gage Chapter 3
4
Recording two patterns of spikes from a thalamic
neuron.
QuickTime™ and a
YUV420 codec decompressor
are needed to see this picture.
McCormick Lab, Yale University
(Audio will click when neuron fires)
To accompany Baars & Gage Chapter 3
5
Recording axonal spikes from a face-sensitive
neuron.
Points to Notice:
-- This is the right hemisphere,
"looking right."
-- close-up zoom
-- can you name the major
lobes?
-- rotation to the medial (inside)
view
To accompany Baars & Gage Chapter 3
6
Neurons that control eye movements.
Notice the
flattened
sheet of the
cortex
To accompany Baars & Gage Chapter 3
7
"Mirror neurons" are involved in imitation.
Mirror neurons fire
when a monkey
observes an action it
may imitate.
Note that this figure
shows a monkey
brain, looking to the
left.
Notice that the graphs record "spikes per second" for
each neuron. The upper traces show the actual firing
patterns.
•
Source: Pulvermueller, (?)
To accompany Baars & Gage Chapter 3
8
Neurons work together
--- in pathways, circuits, networks and maps.
A layer of pyramid-shaped
(pyramidal) neurons in
hippocampus. (Using a fluorescent
dye.)
QuickTime™ and a
TIFF (LZW) decompressor
are needed to see this picture.
A common artificial neural
network, called a
backpropagation net, feeds
back to its own input,
adjusting connection weights
to reduce errors. (Abraham,
TINS 2005)
To accompany Baars & Gage Chapter 3
9
Neurons that fire in synchrony with each other.
Two separate electrodes are
needed to record synchrony
between neurons.
When two competing images into the two
eyes are interpreted as a single image in
the cat brain, visual neurons recording the
"seeing" eye show higher, synchronized
voltages near 40 Hz. (Fries et al, TICS)
To accompany Baars & Gage Chapter 3
10
Single neurons fire spikes; groups of neurons are usually
recorded as "field potentials," like the EEG.
Sleep:
Group
averages
show field
waves.
Single
neurons
fire single
spikes
Waking:
Group
averages
show field
waves.
The electro-magnetic field
generated by tens of
billions of neurons extends
to the outside of the scalp.
Scalp EEG picks up
massive neuronal activity.
Single
neurons
fire single
spikes
To accompany Baars & Gage Chapter 3
11
Neural net models can simulate some human
functions.
This brain-based robot rides balanced
on two wheels, using a Segway
platform. It can play soccer --- slowly.
Its neural net brain simulates the
human cerebellum, using simplified
neurons. Other versions of the Darwin
robot series simulate other brain
regions. All are run by brain-like
neuronal nets with thousands of
"neurons." Such models help to test
out our detailed models of the brain at
the level of cell assemblies.
(With thanks to Dr. Jeff Krichmar, The
Neurosciences Institute, San Diego,
www.nsi.edu).
To accompany Baars & Gage Chapter 3
12
Brain models can simulate simple human functions.
Darwin VII
Retina
Auditory
Object Map
Motor Map
Notice that the "Retina" is picking up the visual markings
on the blocks. Like the real retina, it only picks up colors
and "pixel" locations.
A brain-based robot, with video cameras for eyes, a microphone for ears, a grasper
hand, and the ability to detect "tastes" from the electrical conductiviey of objects.
With kind permission from the authors: J.L. Krichmar & G.M. Edelman, (2002) Machine
Psychology: Autonomous Behavior, Perceptual Categorization and Conditioning in a BrainBased Device, Cerebral Cortex 12:818-830. The Neurosciences Institute, www.nsi.edu
To accompany Baars & Gage Chapter 3
13
Darwin VII before learning to associate a "good" or "bad" taste with a visually
distinctive block (parallel marks vs. round dots).
Retina
Object Map
Darwin VII's brain consists
of neural maps, connected
in the way they are in the
real brarin. You can see
four squares --- matrices of
neurons with connections
between them.
Motor Map
QuickTime™ and a
YUV420 codec decompressor
are needed to see this picture.
Auditory
Motor Map
On the maps, yellow and
red colors indicate higher
neural activity. You can see
the Retina, an Object Map
(area IT), an Auditory map,
and a Motor Map.
Notice that Darwin VII
"hears" its own sound
when it grabs the
object to "taste" it.
To accompany Baars & Gage Chapter 3
14
Darwin VII after learning to associate a "good" taste with a visually marked block.
Note: Notice the improved efficiency of Darwin's association between the visual input and the
rewarding "taste.' Darwin hasn't been told what to do, but learned by experience and association
between its neural maps, guided by "good" and "bad" tastes.
Retina
Object Map
Motor Map
QuickTime™ and a
YUV420 codec decompressor
are needed to see this picture.
Auditory
To accompany Baars & Gage Chapter 3
Darwin VII does Hebbian
learning (see Chapter 3), to
learn objects, and to
associate different inputs
with each other. It has some
20,000 neurons, with a half
million connections between
them. It also has built-in
instincts --- objectavoidance and tries to
sample the taste of
surrounding objects.
15