notes from cognitive psychology

disciplines in cognitive psych

• psychology

• linguistisc

• computer science (AI)

• neuroscience

• philosophy

• cognitive anthropology

3 dimensions of variation

variation according to:

1. the acpect fo cognition being studied

2. the level of organisation at which that aspect is studied

   1. the degree of resolution of the techniques that are being used

strategies for repsonding to the integration challenge

1. global strategies: look for overarching models explaining how cognitive science fits together

2. local strategies: briding levels of explanation and levels of organisation

modularity of the mind

the mind if made up of innate, functionally independent modules

basic assumptions of cognitive pscyhology

1. firmly acknowledge existence of mental processes, focus on investigating them

2. see the mind as active → selecting info from envrionment, relating it to prior knowledge, acting on the processing results

3 reasons for rapid growth of cognitive psychology

1. behaviourism couldnt account for findings in areas such as language acquisition

2. new measuring devices to examine mental activity

3. rise of computer and mind-as-computer metaphor → most significant

history of cognitive science

philosophers (rationalism/associatism) — introspectionists (empericism) — behaviourism — information processing theorists (connectionism) — neuroscientific approach to understanding cognition

how to upgrade the mind

improving individual parts of the CRUM model

1. data: chunking

2. algorithm: add meta-cognitive routines

   1. slow down for high stakes

   2. force alternative decisions

uses of models

• describe

• predict

• explain

• control (treatments)

• prescribe (how it should be done)

• emulate (AI)

2 computational models

1. symbolic models

   1. mechanisms for manipulating symbolic representations. believing a cat is lazy, need both symbols

2. connectionist models

   1. models of cognition preserving neural information processing properties

property of aboutness

to say they involve representations

mental processes have 2 things

mental states and mental content

• the act of believing and the content that the cat is lazy

explanandum

thing to be explained

• why does the fire have different colours?

explanans

the explanation itself (model)

• it is because…

secondary explananda

things we dont want to explain

3 levels of marr

1. computational level

2. algorithmic level

3. implementation level

computational level

what is the goal of the computation, why is it appropriate, and what is the logic of the strategy by whcih it can be carried out?

• what is the model supposed to do? the goal

  • the explanandum

algorithmic level

how can this computational theory be implemented? what is the representation for the input and output, and what is the algrithm for the transformation?

• how is it computed? set of instructions

• explanans

implementation level

how can the representation and algorithm be realised physically?

CRUM

• representational power

• computational power

• psychological plausibility

• biological plausibility

• practical plausibility

CRUM explanation schema

• explanation target

— why do people have a particular kind of behaviour?

• explanatory pattern

— mental representations and algortithmic processes which produce the behaviour

5 assumptions about compuations in the brain on which connectionist models are based

1. neurons integration information

   1. receive

   2. if exceeds threshold, neuron fires

   3. signal passed down

2. neurons pass information about the level of their input

   1. activity level

3. brain structure is layered

   1. hidden layers

4. the influence of one neuron on another depends on the strength of the connection between them

   1. weight determines the effect

5. learning is achieved by changing the strength of connections between neurons

   1. change the weights

human characteristics of connectionist models

• relatively immune to damaged components or to noise input

• allow retreival by content

  • are likely to retreive typical instances from categories

nodes

processing units of the model

typicality effect

the answers which are more likely to be given are typical examples of the category. some info is more easily retreivable

constraint satisfaction

each unit influences the states of all units its attached to. system can find a state in which any change in activity levels would reduce the overall number of satisfied constraints

• allows for consensus, not exact evidence - fault tolerance

supervised learning

given training data, comes up with a model explaining this data

input_ij = a_jw_ij

the input for a cetain cell is the activity times the weight of a previous cell

netinput_i = E_ja_jw_ij

the total input fo a cell is the cum of all activity times all weights of all previous cells

types of inputs

• linear

• threshold linear

• sigmoid

• binary threshold

DELTAw_ij = [a_i(desired) - a_i(obtained)]a_je

the change in weight is determined by the desired output minus the obtained output. it put it back through the network, so i also consider the activity of the cell. and then i change the weight based on this

[a_i(desired) - a_i(obtained)]

the difference between the desired and actual activity level of unit i

• ensures that the size of any weight change is proportional to the size of the error

bias unit

receive no input.

weights can change from learning

• negative: base firing

• positive: threshold

supervised learning

calculate error, this info gets sent back

unsupervised learning

HEBBS LAW: cells that fire together wise together. dont calculate the error

deep

auto-associator

hidden - layer

simple associator

basics of Bayesian statistics

perception, prediction, and decision making all come down to reasoning with probabilities

cue combination

incorporate both visual and acoustic observations as likelihood functions

Bayesian statistics

shifts in x-axis

prior shift

shift in y-axis

likelihood shifts

posterior probability distribution

brain’s beliefs in each possible world state based on prior/likelihood

mondegreens

high prior, flat likelihood (ambiguous)

prediction error minimising

track own systems reliability

• presicion weighting: estimation of reliability

• error units: carry all unexplained sensory info

predictive processing

brain is proactively generating perceptions by continuing to recreate from top-down the world of sensory signals

• constantly minimising prediction errors

  • change your predictions (prior)

  • change your actions (likelihood)

Gregory’s constructivist theory (PERCEPTION)

• indirect, top-down

• construction based on physical sources of energy

• perception = best guess

constructivist because he is saying that retinal images are ambiguous and we construct the perception i out mind.

• meaningless sensory cues + cognition = meaning

Gibson’s theory of direct PERCEPTION

• less computation

• the part thats not moving is telling you where you’re going

  • optic array: pattern of light with all the visual information

The world provides us all the info we need

optic array

pattern of light with all the visual information

• optic flow patterns

• texture gradients

• affordances

Neisser’s analysis-by-synthesis model of PERCEPTION

good viewing conditons - bottom up

ambiguous stimuli - top-down

• most times it is both

Marr’s computational theory of vision (PERCEPTION)

4 stages of deriving representation fo the shape of the object from retinal info:

1. image (gray-scale description)

2. primal sketch

3. 2,5D sketch

4. 3D model representaion

each stage makes info more complex

BOTTOM UP

gestalt psychology

elements of visual input are linked, the identity of the elements depends on the link

gestalt principles

1. similarity

2. proximity

3. good continuation

4. closure

5. simplicity

6. figure/ground

   1. common faith

Barslou’s simulation view of concepts (PERCEPTION)

recognising is top-down. seeing thre word chair evokes the chair related experiences

• re-enactments reported as imagery

imagery debate

when you perceive something in your mind, is it the same mechanisms at play when you are perceiving it in real life?

• degree to which imagery/visuo-spatial processing overlap

critical views:

• demand characteristics

• images can only be transformed

• expert/beginner will see different things

  • epiphenomenon: byproduct

symbolic distance effect

difference judgements easier when things are widely different

• cat, whale — cat, plane

feature detectors

features we rely on depends on how we interpret the data. we choose what matters for recognition

high vs. low imagery skills

high: actually visualise in their heads

low: self-referential strategies. you match features and try to match it

• prone to error

overt attention

head shift

covert attention

no looking difference, attention somewhere else

characteristics of attention

• steerability

  • selecting

3 categories of attention

1. alerting: alters the cognitive system something needs to be processed more (BAYESIAN)

2. orienting: reorienting after being alerted, makes attention more coherent

3. execution: balancing repsonsiveness to the flow of sensory information + memory with adherence to own goals

   1. stroop interference: when activation of one concept interferes with the activation of another

inattentional blindness

gorilla. miss obvious changes because you paid attention to something else

attention

the prioirised processing of some inputs from a larger set of selectable items

cocktail party effect

top down attention / endogenous attention

voluntary

lunch-line effect

bottom-up attention / exogenous attention

involuntary

Broadbent’s filter model of ATTENTION

bottlebeck: selective filter

attenuation filter: filtered stimuli are dampened not blocked, salient can pass through → explains lunchline effect

feature integration theory (ATTENTION)

1. feautre segregation

   1. feature maps

2. features come together on the master map

3. top-down processes of recognising and attention

explains how we recognise

resource theory (ATTENTION)

1. single resource: attention = limited

   1. degradation of multitasking

2. multiple resources

   1. some tasks combine well

Posner task

Cue validity → reaction time

• focus on cross. given cues about where attention is supposed to go. valid and invalid cues.

• see how fast you steer back to right side after invalid cue

attention improves attention

• change blindness

• sistaind inattentional blindness

• ambiguous images

  • selective attention: posner task

MBCT

• being mode insted of doing mode

  • more weight to likelihood, more weight to prediction error, large belief updates

  • enhancing sensory precision at lower level of cortical hierarchies

• decentring

  • being able to experience thoughts and precepts simply as events in the mind that arise and pass

  • prediction errors carry high weight → carried up to higher levels where abstract beliefs are represented → constant belief updates at higher levels may lead to recognising that “fixed beliefs” can actually change

• reactivity

  • plattening priors: any deviation from the mean is less meanignful and thus decreases the impulse to reponse to incoming sensations

  • as a result of being mode and decentred perspective → constant belief updates → higher prediction errors are acceptable → less tendency to defend beliefs against anxiety

Garret’s model of speech production (LANGUAGE)

top-down

speech errors

meaning → words → order → sound → prepare for speech

doesnt account for errors across levels

Levelt’s model of speech production (LANGUAGE)

top down

production of single words

concept of word prepared → selected → meaning concept selected → syllabes selected → sounds → speech preparation

monitoring: recognising an error (may be broken in speech difficulties)

feedback between levels limited

modular theories of speech production

series of non-interacting stages. each stage different kind of processing

interactive/parallel theories of speech production

less constrained, multiple sources of info operating

Dell’s spreading activation model (LANGUAGE)

connectionist

predict speech errors

4 levels

1. semantic level

   1. conceptualise what we want to say

2. syntactic level

   1. structure of the sentence

3. morphological level

   1. morphemes selected

4. phonological level

   1. sounds activated

think of meaning → serveral possible words active → the most active word wins → the sounds of that word activate → the sounds are place in order and spoken

invarience problem

speech not always the same

the segmentation problem

difficutlt to distinguish sinlge words in continuous speech

the cohort model (LANGUAGE)

expectations once we hear initial phonemes

• uniqueness point: point where other optino have reduced so much that one word remains

context plays early role.

evidence: lexical decision task: string of words, decide if it a word

sometimes delay in word recognition…

TRACE model (LANGUAGE)

top down

spoken input

constantly updates

3 layers of detector:

1. feature level: basic sound properties

2. phoneme level: recognise individual speech sounds

3. word level: phonemes activate possible words

why TRACE: you hear, your brains leaves a trace of activation across the network

the dual route model of reading (LANGUAGE)

step 1: see word

• route 1: grapheme-to-phoneme coversion

  • letters → sounds

    • sound out letters

    • for new words

• route 2: lexical route

  • word recognition → meaning → pronounciation

    • recognise word visually

    • find it in stored known words

    • assess meaning

    • retreive correct pronounciation → word recognition

• route 3: direct lexical route

  • word recognition → pronounciation

    • you can pronounce it, even if you dont know its meaning

dual route:

• sounding out rules

• recognising stored words

4 characters of emotions

• cognitive: appraising significance of emotion

• motivational-behavioural: acions in reponse to emotion

• somatic: physiological reponses to emotion

  • subjective-experimental: actual experience

Paul Ekman: Basic EMOTION theories

1. there are seperate discrete emotions with distinct facial expressions

2. they have distinct evolutionary functions

Ekman study

• non-literate cultures

  • short stories → what emotion is it?

  • facial expression → what emotion is it?

    • westerners also understood their expressions

not always consistent

some cultures have shame universal

dimensional-affective space models (EMOTION)

valence: (un)pleasantness

arousal: engagement/reponsiveness

multidimensional models (EMOTION)

1. scalability

2. calence

3. persistence

4. generalisation

5. global cordination

6. automaticity

7. social cordination

James- Lange theory (EMOTION)

body → interpretation = emotion

facial feeback hypothesis: 🙂 → happy feelings

x paralysation

x people differ in introception

x experiencing emotion first: blushing?

x no cognition

Cannond-Bard theory (EMOTION)

physiological + emotional state at the same time

• can have different emotions in response to the same physiological symptoms

• can have no emotions in reponse to physiological symptoms

characteristic, but do not give rise to emotion

x no cognition

two factor theory, Schachter-singer theory (EMOTION)

situation dependent

• physiological arousal + interpretation

test: adrenaline

• placed in anger/euphoria

• if told about side effects, they dont attribute it to emotion

  • if not told about side effects, they attribute it to emotion

Affective primacy Zajonc’s theory (EMOTION)

cognition independent from emotion. can be effective judgement without emotion

• mere exposure effect

cognition is conscious

• 🙂 / ☹ experiment

  • unconscious (4 m/s): 🙂 did have effect

  • conscious (1s): 🙂 had no effect

Cogntive-primacy, Lazarus appraisal theory (EMOTION)

emotion involves cognitive appraisals

• emotions are goal-directed mechanisms

3 types of cogntitive appraisals:

1. primary: initial assessment

2. secondary: assessment with resources

3. reappraisal: continual monitoring

emotion as a result from interpretations of our reactions

multilevel theories (EMOTION)

pre-attentive and conscious processes in emotion

• fast/slow route

basic emotions

anger, disgust, fear, happiness, sadness, surprise

predictive coding

through accurately inferring causes, the brain can anticipate needs of body + prep to meet needs before they arise

• prior: predict what is needed/whats most likely

  • prediction is updates by prediction error

    • emotion

  • likelihood + current brain state → inluence probability of future brain states

concept

mental representations

• context dependent

• conepts are predictions that the brain uses to categorise sensory inputs and motor actions

  • = predictive coding

• there is variation is emotion categories

  • meaningful: so we can generalise

• challenge: needs regularities to make predictions

experience depends on (EMOTION)

• introception of bodily signals

  • higher-level cognitive apprialsals

recognition or emoiton first

Connan-bord/ holman barret: same time

James-lange / affective primacy: emotion first - then recognition

two-factor / cognitive primacy: recognition first - then emotion