Cognitive Science Exam 2

three practical questions

what can a neuron do?

what can a neuron know?

what can a neuron learn?

what can a neuron do?

compute a threshold function of a spatial and temporal integration

spatial

dendrites, extent and direction of dendrites

temporal

time window of refractory period (can only fire a certain amount of times per second)

a logical calculus of the ideas immanent in nervous activity — paper by mcculloch & pitts 1943 || theoretical model of a neuron

1) binary device with (binary) inputs (excitatory inputs add linearly, inhibitory inputs prevent neuron from firing, but different neurotransmitters will inhibit OR excite — false, it’s not a black and white scenario)

2) neuron has a fixed threshold (not true)

3) neuron has binary output (true)

inclusive or

mcculloh-pitts neuron — binary neural network based off neuron in brain

you either have action potential or you dont

input of 1, fires

input of less than 1, doesn’t fire

enigma machine

computer during WW2 that the germans used to encrypt messages so americans couldn’t decipher it

results and implications of mcculloch & pitts paper + inclusive or

1) by combining any logical propositions into networks, any finite logical expression can be realized

2) paper had very little effect in neuroscience literature

2) paper had enormous effect in computer science: binary operations, logic gates, computation, etc.

how can neurons be different from each other?

neurons can be different from another either anatomically or physiologically

anatomical differences between neurons

“hardware” of a computer

number of dendrites, spatial reach of dendrites (receptive field), length

physiological differences between neurons

“software” of a computer

physically identical — what can we make analogous to software in a neuron? — CHANGING THE WEIGHTS

synaptic efficacy

how easy is it for input from a dendrite to excite a neuron

w < 1

w > 1

inhibition vs. excitation

mach bands

an visual illusion with perception, physiology, and neural computation

black and white ombre — gets lighter before it gets darker — increases edge detection to better see transitions

horseshoe crab

photoreceptors in the eye that capture light — convolution of all the light from the different photoreceptors in the neurons

convolution

computes weighted sum of inputs

(ex: photoreceptors — take their input of light and add them together)

receptive fields

larger receptive field: all orientations blurred

smaller receptive field: sharper image and better edge detection

hermann grid illusion

extra foveal representation in the corners

how can a neuron learn?

modification of the synaptic efficacy of dendrites, change in the values of Wi of the model

neurons can change with experience — learning

hebb 1949

take the same route 100 times, a path is created in the grass

a is reliable to firing b, creating a strong connections between the two neurons — a’s efficiency as one of the cells that fires b is increased

perceptron — rosenblatt 1958

first neural network

weighted sum of input — fires or doesn’t fire

neural model: input from retina (light?), output neuron, 
“pattern detector”

artificial intelligence

the theory and development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.

neural network

a computer system modeled on the human brain and nervous system

turing machine

-set of instructions — “computer program”

-infinite tape (environment)

-labeled states (where you are on the tape)

-binary code (0,1) - language used to specify everything — programs, data, states, etc.

algorithm

a process or set of rules to be followed in calculations or other problem-solving operations

first computers

world war two — enigma machine

when did technological advances start

1969

“turing-like” computers

-programs (python, c, java)

-input (output) from (to) environment (disks, paper, etc)

-labeled states (CPU)

-binary code (on-off electrical pulses)

what was computation limited to?

tasks that can be carried out with step-by-step goal-directed instructions — “computable”

what is uncomputable?

many things humans can do that can’t be translated into computers

human insight (aha moment)

spontaneously gained knowledge

halting problem

determining from a program and input if a program will end or run forever

turing proved that an algorithm can not solve this in the general case

meta-knowledge

knowledge about what you know
(ex: what is the largest shopping mall in russia? — you immediately search your brain without thinking, and can answer immediately that you have no idea)

turing test

can machines think?

3 person test — can a computer successfully imitate a human?

algorithmic-based computation is a subset of human mental computation — limited

ai winter

schank and minsky warned that overenthusiam for AI would lead to disappointment when progress wasn’t made as fast as wanted — projects and funding are abandoned

neural networks inspired by two things:

1) limited knowledge humans have on neurons

2) psychology of learning

associative memory

associate things with good/bad outcomes, effective/ineffective actions

neurocognition / neocognition

inspired by mammalian visual system by fukushima (1979)

focused on: pattern recognition

first convolutional neural network

back-propagation

rumelhard, hinton & williams 1986

multiple layers of neurons - “hidden layers”

capable of non-linear classification

small scale successes 1980-90s

cognitive science applications

-parsing

-grammatical

-story understanding

-(limited) success with real world problems

era of the non-intelligent computer

1990-2012

deep learning revolution

2012-present

long-standing difficult problems in AI solved to a level of accuracy unimaginable just a few years before

deep convolutional neural networks

ability for face recognition, multiple layers of stimulated neurons, “in-the-wild” images

right hemisphere (simplified)

emotional, holistic, intuitive

left hemisphere (simplified)

logical, calculating, linguistic

lateralized brain functions

one hemisphere “controls” task processing — doesn’t necessarily mean the other hemisphere doesn’t do anything

localized brain functions

contiguous portion of the brain controls task

amnesia

selective memory loss

apperceptive agnosia

good vision, no ability to recognize shapes or forms

associative agnosia

good vision, good shape and object recognition

man who mistook wife for a hat

dyscalculia

cannot do simple calculation — left angular gyrus

aphasia

broca — cannot speak (tan) — left frontal

wernickie’s aphasia — can’t understand — left temporal

alexia

cannot read

agraphia

cannot write

prosopagnosia 

cannot recognize faces

capgras syndrome

think someone you love has been replaced by an imposter, no other signs of dementiace

cerebral hemispheres

corpus callosum

vision

optic chiasm

audition

subcortical

sensory motor

spinal chord

emotion

anterior commissure

question 1 - split brain patients

visual, verbal, tactile coordination

cup in left hemisphere — cup

cup in right hemisphere — nothing — no perception of visual signalque

question 2 — split brain patient

emotion processing

shown word “devil”

left hemisphere: was word good or bad — says bad — says the word was devil

right hemisphere: was the word good or bad — says bad — doesn’t know what the word was

how do emotional stimuli make it across the hemispheres without corpus callosum

anterior commissure

question 3 — split brain patient

systematic coordination of motor activity

alien hand phenomenon: middle aged woman would try to strangle herself with left hand, right hand would try to fight off left

right hemisphere acts on suicidal tendencies with left hand — right hand (left hemisphere) is not aware of intentions and tries to fight off left hand

question 4 — split brain patient

independent existence

young man — asked what he wants to do with his life. left hemisphere said architect, right hemisphere said race car driver

left hemisphere interpreter

makes sense of data

picture matching task — left hem chicken foot; right hem snow storm

left hand picks shovel for snow storm — matches right hemisphere and vice versa

why did your left hand pick up the shovel? — scoops up chicken foot — makes up answer to make sense of choice

right hemisphere statistician

experiment; light 20% of left, 80% on right

predict location of next light (right/left) — light locations are random but with the 80/20 left to right probability

RESULTS: human distribute responses to match probabilities (push right button 80% of time) — 68% correct — overthinking pattern recgnition

rats maximize odds and push right 100% of the time — 80% correct (goes where food is more probably, doesn’t decide to go somewhere else — takes information at face value)

perception

processes by which physical energy from the environment impinging on senses is converted into electrochemical energy by the sense and processed by the brain for the purpose of effective and adaptive interaction with the world.

steps of percpetions

sense, energy, organ, transduction, brain

intensity of vision (light)

photons / unit area / unit time

wavelength composition

color, hue, spectra

saturation — percentage of white light

retinal image

pattern of light on the retina

-illuminant (light source)

-reflective properties of the object

-viewer position

philosophy of vision

create a 3D map of the world int he brain

vision is for action — navigation, manipulation, exploration

focus in an image

cornea and lens

contrast in an image

pupil — not too much or too little light

light sensitivity in an image

rods and cones

where does light enter in the retina?

ganglion cell layer

dorsal stream / ventral stream

where / what

seeing

surface perception: segment into “figure” and “ground” — a complete 3d representation or map of scene

segment / parse into objects, recognize and identify objects, scene perception (conglomerations of objects, layout)

occlusion

filling in the blanks of an object — sometimes might not even be true

stereopsis

each eye gets a slightly different image — depth perception and disparity

binocular vision

perceiving objects — what are the problems with the direct analysis of shapes?

viewing angle, photometric problems (illumination, shadows, highlights) object setting (isolation, occlusion) rigid or non-rigid

bottom-up (marr 1981) perception to object analysis

starts with raw data and builds around that to create a perception of the object

bayesian models

hidden faces in the rocks — top-down processing (pre-existing knowledge and expectations to interpret data)

convolutional neural network

sound

pressure variation in t around the mean atmospheric pressure

how do you make a sound?

vibrations and resonance

frequency

dimension of sound — number of sound waves to pass any point in a second

measured in hertz (Hz)

psychological correlation of frequency

pitch

amplitude

magnitudes of the movements produces

measured in decibels

psychological correlation to amplitude

loudness

timbre

whatever is left after equating pitch and loudness

psychological correlation of timbre

sound quality

fourier spectrum

representation of sound

x axis: frequency

y axis: amplitude

spectogram

representation of sound

y axis: frequency

x axis: time

language comprehension

speech perception

lexical access

sentence processing

discourse processing

psychoacousitcs

auditory processing

speech perception

factors that affect speech processing

linguistics

study of language structure — structure of language human cognition universal traits (what do all languages have?)