Lecture 001

How to Read Behavior Paper

1. What is the research question? (be specific)
2. What is the research design? What is being compared? What is the control condition.
3. What is the behavioral (dependent) measure? Is there speed-accuracy tradeoff?
4. Interpret data yourself.

Experimental Manipulation: do with stimuli, check change in behavior

• Experimental Condition

• Control Condition: apply "Subtract Logic" to change only the manipulation we are interested in.

Error Types:

• semantic error (accuracy): binary

• response time (response latency, reaction time): continuous

Speed-Accuracy Tradeoff: Two extreme strategy. You cannot draw clear conclusion with graphs that show Speed-Accuracy Tradeoff.

As human optimize for accuracy, binary accuracy might be similar while response time vary drastically different

Flanker Effect: interference in visual processing (identifying the dirrection of middle arrow in a sea of arrows)

• >>>>><>>>>>

• >>>>>>>>>>>

Simon Effect: spacial interference in visual decision (press red bottom when there is a red square on the right of the screen)

Stroop Effect: word color vs. word meaning

Incongruence always increase reaction time

Eye-tracking: tracking quick, online performance

• how do people understand sentences

• we find if we say "the boy wants to eat _____"

• the participant will look at the eatable item on picture

Reading the graph

• Fixation(x): whether the eye is looking at target object x

• Restriction: restrict to the word "eat"

• Non-restriction: control group

Participants Types: this is so that cognition change by age and diseases; this is so that we can observe learning;

• Neurotypical yong adults

• Healthy old adults

• Healthy children, adolescents (infants to highschool)

• People with specific disorder

  • tell us about disorder
  • tell us about brain structure

Testing 2-year old: This is a MIP, there are two MIPs

Neural Network: behavior can be explained with neural network model

Neural Tools

Why use neural tools: interested in specific part of the brain

• Non-invasive

• Invasive

CT Scans

• quick easy

• good for seeing bones

• X-ray unhealthy

• cannot see soft tissues

MRI

• expensive

• no X-ray, healthier (but magnetic field)

• see soft tissues well

PET: not really used

fMRI: functional MRI

• detect changes in blood flow (oxygen in the blood) in the brain as a function of neural activity.

• BOLD fMRI: Blood-oxygen-level-dependent imaging (usually used)

• great for special resolution millimeter precition

• bad for: temporal resolution; brain changes a lot more faster than resolution.

Networks: multiple regions of brain on specific or general things (e.g. hard tasks)

• Anatomical connectivity: looking at white matter (mostly axons), may imply functional connectivity; invariant in development of the brain

• Functional connectivity: which region work together when the brain is carry out cognative activity (statistic correlation, no causal relation); task dependence

• Effective connectivity: temporal analyses (can draw causal relation as experiment)

Neural Psychology: study lesion using "voxel-based leision symptom mapping" and make statistic correlation with functional impair

How to Read Neural Imaging Paper

1. Understand question (Introduction)
2. What is being compared to what (Methods)
3. Behavioral Data and Graphs (Results)
4. Neural Data (Results)

You don't need to understand charts stuff. They are for comparing with other studies and non-important details.

Non-Invasive Neural Methods

EEG (Electroencephalography): non-invasive electors measure of electrical signals projected to surface of the skull.

• non-invasive

• good for time resolution, not good for spacial resolution due to the nature of electrical signals

• ERP (event-related potential): condensed form of EEG

  • event: make event associate with patterns

• source location methods: large dataset, many electors with computer reconstruction

MEG (Magnetoencephalography): high spacial and temporal resolution

• non-invasive

• extremely expensive

tDCS: Transcranial Dirrect Current Stimulation

• non-invasive

• cheap, current goes everywhere, low spacial resolution

• Classic tDCS: directly sending week electrical signals to brain in one region

• HD-tDCS: many focused electro

• alternating current vs. dirrect current

• measure change in activity as a result of injecting electrical durrent

TMS:

• non-invasive

• very precise, paried with structural MRI, good spacial resolution

• temporally interrupt neurons, just like tDCS

Invasive Neural Methods

ECoG (electrocorticography): record passive data from the brain

• before surgery, measure brain data with invasive electors

• essentially EEG, and conduct cognitive study to identify location of seizure

Awake Surgery: playing guitar while doing surgery

Visual Processing

Visual Processing: happens in occipital lobe

Inattentional Bias: can't perceive gorilla in video

Change Blindness: can't spot change of environment in video

Visual Sensing

Neuropsychology: the study of cognition in individuals with neurological deficits.

Visual agnosia: can't identify and classify object. (The person can see objects, pick them up and manipulate them without difficulty, but cannot tell what they are or what they are for by looking at them.)

Optic ataxia: can't interact with object, don't know its location. (The person can see objects and knows very well what they are and what they are for, but has difficulty interacting with them, like reaching and grabbing them.)

Dorsal pathway: control spacial location and action. Damage would cause Optic ataxia. Generate Ebbinghaus illusion.

Ventral pathway: object classification. Damage would cause Visual agnosia.

// QUESTION: what is identity priming, and non-identity

priming: Facilitation in the processing of an item as a function of prior exposure

• Identity priming: repetition priming make participants quicker response time (e.g. apple image and apple image)

• Non-identity priming: emantic, phonological (word "DOCTOR" and "NURSE")

subliminal priming: priming in the surface of conscious

• Subliminal identity priming works

• Subliminal priming is visual (only seen for the same picture, not even for the same object with a different picture)

• Subliminal priming is sensitive to position, but doesn’t go away. More higher level in ventral visual pathway.

Visual Attention

Foveal (central) vision: what you fixate on; highest visual acuity Peripheral vision: outside of your fixation point; lower acuity

Disorder of Visual Attention

Hemi-spatial neglect: cannot notice strange in left vision, no awareness.

• situration

  • cannot copy left half of star
  • draw only right side of image
  • cannot process left side

• damage: right side of occipital lobe and parietal lobe

Early selection theories: This view posits that people have limited capacity for visual processing, and only ever see what they attend to.

• select region of pixel and process all the way through

• evidence: inattentional blindness

Late selection theories: This view posits that perception is automatic and has unlimited capacity. Brain processes everything, regardless of whether people want it to or not.

• process all pixels in some degree

• evidence: subliminal priming (see something, all pixels, not in conscious)

load theory of attention:

• Perception is automatic (late selection view)

• But it IS capacity-limited (early selection view)

Concept of Attention:

• Targets: what is relevant to our goals and we wish to process.

• Distractors: what is in our visual field, but you’re not intentionally processing. Sometimes distractors are good (in traffics, targets are roads, distractors are unexpected people on roads)

Perceptual load: load in external world

• High: Attention consumed by focusing on the task and the target. No capacity left for processing distractors. This would look like an early selection view. hard to distracted

• Low: Target processing is easy. Lots of capacity left for processing other information. This would look like a late selection view. easily distracted

Cognitive load: load in your head

• High: lower ability to focus on the target. easily distracted

Brain Areas:

• Frontal Lobe: cognitive control

• Parietal Lobe: spatial attention (mostly right brain)

• Occipital Lobe: sensory processing of visual stimuli

Cognitive Modules

Advantage of Having Modules:

• increase processing speed

• modular development, specialization

• supports

  • Behavior: Face Inversion Effect - you can't easily recognize absurdity in inverted faces
  • Neuropsychological: Prosopagnosia - neurological disorder in which the affected individual can no longer recognize familiar faces
  • Neural: FFA Region

proactive interference: old info interfere with new info (what we want) retroactive interference: new info interfere with old info (what we want)

Disadvantage of Having Modules:

• brain plasticity

• supports

  • expertise for cars and birds recruits brain area involved in face recognition (therefore FFA not exclusive for faces)

• debate: nativism vs. empiricism (nature vs. nurture)

  • result: monkey growing up not looking at faces don't have FFA region recognizing faces, instead, recognizing hands.

Reproductive Memory: reproduce what was remembered Refill Memory: fill gaps of missing memory

Memory experiments:

• Recall: free or cued recall

  • high correctness in high frequency items

• Recognition: present with the material, ask whether the material is new (seen before), or old (unseen)

  • high correctness in low frequency items (because notice of weird item)

Primacy Effect: remember first few words better Recency Effect: Remember close in time

The more words gets recalled (more similar words), the exponentially (not linear) more words is false memory.

know vs. remember:

• know: only know it appear

• remember: being able to re-live the experience

• recall is not the cause of false memory (subjective feeling of remember/know is the same)

Result: remember, confidence, memory detail doesn't indicate actual happen

False memory created by network of knowledge tree because they are related.

Depth of Processing: knowledge tree

transfer appropriate processing: remember things you attended to

chunking: breaking up knowledge into chunks to remember better.

state(context)-dependence of memory: you remember the best if you are in the same room / context

• episodic memory: memory of events with spatial, temporal context

• semantic memory: long term general knowledge in large scale.

Serial (feed-forward) system: long term memory cannot affect working memory (bad, old model)

Interactive feedback System: long term memory can affect working memory (good model)

• supported by expert / novice remembering chess pattern in short time

Multimodal Processing: if impaired, cannot associate visual with auditory information.

phonological loop: vocal or subvocal rehearsal

• word length effect: only remember around 2 seconds

• disrupt rehearsal process: by repeating the word "the" outload while trying to remember phone number

Recourse in Working memory

• discrete: if you have 4 bits of memory, and need to remember 6 bits, you can remember 4 bits perfectly while can't remember other 2 bits

• continuos: you remember all 6 bits but less precisely

• set size effect: ??

Cognitive Map

Place Cells: orientation-invariant and location-specific cell to fire. Boarder Cells: activated when the mouse is near a boarder of an environment Head Cells: location-invariant but orientation-specific cell to fire. (not only in Hippocampus) Grid Cells: Combine input from head direction and (in Entorhinal Cortex)

Two systems for navigation

• Coordinate System: a coordinate system, produced by grid cell, forming equalaterial triangle

• Landmark: assign landmark to perceived coordinate

Matter for Spacial Navigation:

• Perspective Taking: the ability to imagine self in different location and orientation.

• Working Memory

• Anxiety

• Motivation

• ...

Nature vs. Nurture

Statistical Learning vs. Rule Learning

Encoded Latent Space constraints Statistical Pattern

• Temporal constraints

• Visual constraints

• Learned Latent Meaning constraints

• Gestalt Principle

Language

Phonemes: smallest units of sound.

• Letter: symbols (two letters can correspond to 1 phoneme. SHEEP has 3 phonemes and 5 letters)

• Phonemes are language specific

How baby segments words in new language: statistical learning

• separate phonene for a lanuage is difficult

phonene restoration effect: brain fills phoneme

Mcgurk Effect: "ba-ba-ba" vs "fa-fa-fa" (multi-modal learning)

Phantom Words: hearing words from non-sense sound.

The cohort Effect:

1. we process word even before the whole word finish.
2. IMPORTANTLY: people do tend to look at phonologically similar words even before hearing the similar parts (look at "speaker" more after hearing "bea" in "beaker" than unrelated item.)
3. semantic related, but not sound-related will be fixated better than unrelated item.

Neuronetwork Model

• feedforward: explain acoustic related activation

  • from phonemes to words to semantic knowledge layer

• feedback: explain semantic related activation, explains phonene restoration effect

  • from semantic layer to words layer
  • from words layer to phonemes layer

Some Review

Sketchpad, Episotic buffer, Phonological Loop

• sketchpad: short term visual memory

• phonological loop: language verbal short term memory

• episodic buffer: cannot link verbal and visual memory

biased effect: you always need to compare data with a baseline. People are not unbiased (response 50% yes and 50% no to a question with no information). two mechanisms: decay, interference

crime identification

• test effect

• freezing effect

• post-event misinformation

• retrival induced forgetting

Language Production

Activation: multiply by weight in neural network Selection: activation function, needs "monitoring"

Monitoring

• Comprehension-based Monitoring

• Production-based Monitoring

Domain General vs. Domain Specific

Language Error

Lexical Errors: language error by swapping places of word in sentences.

• noun always swap with noun

• can swap accross long time

• when speak, slight activation of false activation produce error

Phonological Errors: language error by swapping sound.

• don't care as much as noun or verb

• can only swap when together

• when speak, the feedback from phonome to meaning back to phonome produce error

Aphasia is a extreme of daily language error

Global noise: network suffer similarly Noise in specific layer: more error down the stream

Wernick's Aphasia: confidently put stream of speech that is grammatical correct but make no sense. Can't quiet understand instruction. Can make up word.

Brocka's Aphasia: kinda don't know how to link concept to word and therefore can't put sentence together. Can only find nouns, hard time finding verbs.

Speech complementary models

• forward (and inverse) model

Comprehension-based monitor: high level semantic correction Forward model: still have backprop, deaf child can't correct motor sound based on what the child hear.

• generate semantic error in word production is in (meaning to sound): feedforward

• generate phonological errors in word comprehension (sound to meaning): feedforward

• phonene restoration effect: feedback

phonological: words semantic: high level idea

L1 different than L2 for different language. Language is not the same in multilingual than in monolingual??

Bilingual benifit: switch language, inhibition gives daily practice

Meta-analysis: combine data

Task Switching

Task Switching:

• can be conditioned

• not global, it is task dependent

• context can pre-load task specific memory (my interpretation)

• stimuli with non-uniform probability can facilitate task switching

• the classification of whether I need to switch to specific task in the next timestamp is reward-driven (loss calculated from ground truth, is there real task switching?) and can be conditioned.

Traditional view: cognitive control (including task switching) is high level regulatory system that governs low level system

Cognitive Flexibility

cognitive control: override habits (automatic behavior)

• habit: bottom-up processing (driven by stimulus)

• control: top-down processing (driven by conscious)

Associative Learning between stimulus and control:

• stimuli -> control -> behavior

traditional view control: control supervise everything learning view: control is baked in forward model and used throughout (learning view does not predict far transfer)

Two Biases in Studies:

• passive control group result placebo effect (compared to active)

• publication bias

Meta-analysis: get rid of publication bias (null result hardly published and accumulative error) by analysis on multiple studies

Effect Size: significance times the mean difference

Intelligence Tests

IQ: operationalized measure $g$: general intelligence by construct (underlying variable, can't be directly measured)

• g is reflective by classical view

Positive manifold: positive correlation among different tests of intelligence. The score in subtests are correlated.

• ability differentiation: correlation is lower in people with higher mean ability

• worst pedormance rule: the worst performance of all subtests is a better predictor of $g$ than better performance of subtests

Latent variable: can't directly measured

• principle component analysis

• factor analysis

• structural equation modeling

reflective: variables that affect other variables (lower level latent) formative: variables that is affected by other variables (high level)

Manifest variables: can be directly measured

Brain is Malleable

Evidence

• people solve task by reuse other partis of brain: blink people use visual region of brain to establish acoustic, tactile representation

• life-long adaptation: brain is malleable to evolve for new tasks

• tools evolve to maximize brian process efficiency (e.g. words, math formula)

  • people use perception in solving math equations
  • our focus is align with order of operation
  • older people use more perception cue: so learning math involves learning common patterns of mathematical symbols, and we use spacial transformation to solve algebratic equations
  • Notation adaptation: we write $x \cdot y$ as $xy$ (we write notation with different space in between unconsciously)