Transcript Results - UCSD Cognitive Science

Introduction to
Cognitive Science - 1
www.cogsci.ucsd.edu/classes/WI08/COGS1/
Adrienne Moore, 1-16-08
Office hours: Wed. 4-5,
Cognitive Science Building, Room 127
Overview of Today’s
Section
• Review Gary Cottrell’s lecture & reading
– What is a neural network
– What kinds of things can neural networks
show us (2 applications)
• Solving the Visual Expertise Mystery
• Explaining Conflicting Views on Emotion Expression
• Review Jaime Pineda’s lecture
– I’ll address whatever you’d like to discuss
• (such as making social inferences, macaque mirror
neurons, human mirror system, EEG, mu rhythm,
autism…)
What is a Neural Network?
• “a cartoon version” of a neuronal network
• Neuron – node/unit, synapse – connection
• Simple processing units connected by + & - links
spread activation and inhibition to other units
Neural Net Learning
• Present the network with training examples (a
pattern of activities for the input units plus the
desired pattern of output unit activations)
• See how well the actual output matches the
desired output (calculate the error in each output
neuron)
• Change the weight of each connection so that
the network produces a better approximation of
the desired output
1.The Visual Expertise Mystery
• Fusiform face area (FFA) –
• face specific? or
• fine-level discrimination
The Mystery: Why is an area that begins as a
face processing area recruited for these other
types of stimuli?
How do Neural Nets address this?
• “Expertise”: performing as fast at identifying
members of a category as individuals (e.g. indigo
bunting) – “subordinate level” -- as at verifying
category membership (e.g. bird) – “basic level”.
• Method: compare basic NNs to expert NNs
• Pretraining: all NNs learned basic level
categorization of faces, books, cans, & cups;
experts also learned expertise at one of the 4
• Phase 2: Compared classification ability of basic
NNs to expert NNs
Results, Conclusions
• Computational models trained to make
fine-level discriminations learn Greeble
expertise faster than models that have
never been trained to become experts
• So, if there is a competition between
cortical areas to solve tasks, the FFA
would be primed to win other fine-level
discrimination tasks after learning faces
2.Conflicting Views on Emotional
Expression
• Is face perception an
example of categorical
perception? Paul Ekman –
6 discrete categories, the
basic emotion
• Or are emotion
expressions continuous,
not discrete?
James Russell –
“circumplex model” of
emotion space
What does GURU’s neural net
model say about this problem?
• Their model of facial expression recognition:
•
•
•
•
Performs the same task people do,
On the same stimuli,
With about the same accuracy.
It organizes the faces in the same order around the
circumplex,
• And ranks the difficulty of classifying the emotions about
the same as people do.
• (& w/out “feeling” anything, w/out access to human
culture, etc)
• The GURU model simultaneously fits data
supporting both categorical and
continuous theories
– & also suggests the improvement at
boundaries is in the representation of the
data, not of the categories
• Conclusion -- Discrete categories of facial
expression (Ekman’s idea) are not
required to explain the data
Pineda lecture: How do we
comprehend the actions of others?
• We’re really good at it
(even w/ sparse
information)
• Theory theory vs
Simulation theory
• Mirroring
Macaque Mirror Neurons
• Activated by goal-directed hand actions,
• And by observation of same actions.
• Do not respond to target alone, to
intransitive gesture, or to mechanical
movement.
• Discovered by single-unit recording of
neuron “spiking” or firing
Human Mirroring System
• Central sulcus, motor
cortex, premotor
cortex
• Mirroring system:
Sensorimotor cortex,
to IFG and inferior
parietal lobule, to STS
• Mirror neurons are
found in premotor
cortex
• Congruent cells (fire to reaching for peanut and
observing reaching for peanut)
• Logically related cells (fire to reaching for peanut and
to observation of eating)
• that is, they respond to the intention behind the action, or
a generalization made from the action
• Evidence in monkeys: cell fires even when the target is
occluded iff monkey knows there is a target
• Evidence in humans: hand reaching when intention is to
clean up -- ; hand reaching when intention is to eat –
“these cells know the difference”
Using the Mu Rhythm to study
Mirroring
• The mu rhythm is an EEG signal said to index mirror
neuron activity
• Hypothesis: Mu suppression (desynchronized) = mirror
neuron activation, mirroring
• Mu normal (synchronized) = mirror neurons at rest
What is EEG?
• Electroencephalography:
noninvasive method for
measuring the brain at work
with very good temporal
precision
• EEG Spectral analysis:
• Asks where in the signal (at
what frequency band) is
there a lot of activity?
Pineda’s experiments in normal
population
• 1. People tossing balls -- 3 conditions: little
social interaction vs 3rd person social interacting
vs 1st person social interaction – Results: the
greater the social interaction, the greater mu
suppression
• 2. Emotional faces -- 2 conditions: name the
emotion vs name the gender – Results: right
hemisphere, no difference, both tasks
suppressed mu; left hemisphere, greater
suppression for emotion than gender:
Interpretation: right automatic, left task
dependent
Pineda’s Experiments in Autistic
Population
•
•
•
•
•
Some Symptoms of Autism:
Social impairments
Language development delayed
Repetitive behavior, restricted interests
Hypothesis: autism involves impaired mirroring
system
• Experiment: 3 conditions: move hand, observe
hand movement, observe ball movement
• Results: in autism, mu rhythm is not suppressed
when observing hand movement in others
Creating a Temporary “Autistic”
Brain?
• Transcranial magnetic stimulation (TMS) used to
“turn off” IFG area
• Social ability measured with Baron-Cohen’s
“reading the mind in the eyes” task (ASD
population doesn’t perform well)
• Preliminary Results: RT increases after IFG
TMS in emotion task and not gender task;
accuracy decreases after IFG TMS in emotion
recognition task, not in gender task
Reversing Deficits in Autism?
• EEG neurofeedback training (NFT) allows you to
gain control over your mu rhythm
• If mu rhythm indexes mirror system activity, NFT
may increase mirror system activity
• Cognitive, behavioral, and anatomical
assessments are taken pre and post NFT
• Preliminary data: mu rhythm suppression looks
more normal post NFT; parents report positive
behavioral changes post NFT