Cog Sci 1B Midterm

Listening/Processing Info

- 7+-2 rule: we can only process about 7 items at same time Filter Model of Attention: Everything is picked up on some level but our focus/awareness limits what we experience Experiments: People listen to two different audios in either ear and are instructed to ignore one. People are very good at this and won't even realize that the audio in the ignored ear is in a different language or the same word over and over again. However, they are sometimes able to pick out specific words like their name or fire. Implications: People consciously attend to one source of info and ignore other sources but cary in degree to which they focus/ignore distractions

Human Neurons

- 86 billion in brain -one neuron may receive input from 10,000 others (as high as 50,00 in HC and 200,000 for Purkinje cells in cerebellum -many different neurons connect to dendrites of each neuron -produce varying degrees of excitatory and inhibitory effects -activation of neuron results from sum of excitatory and inhibitory synapses -- when it reaches a certain minimum threshold the neuron will fire.

Default Mode Network

- Brain areas that are active when we are not focused on a task (resting state) - Can also be active in certain goal=oriented tasks (self-referencing, recognition of emotions in others, remembering the past, imagining the future) - Associated with rumination about past, worrying about future and self-consciousness - Meditators show reduced activity in DMN → increase happiness - Posterior Cingulate Cortex, Precuneus, medial frontal cortex, temporoparietal junction Cognitive Disorders Associated with DMN - Alzheimer's, Autistic spectrum disorder, schizophrenia, bipolar, depression, PTSD

Convolutional NN

- CNN used in facial recognition, self-driving cars, Natural language processing - First revealed in machine vision competition in 2012 - Creates a type of filter that functions as a localized feature detector - Successively examines different parts of the image to filter out everything except the particular feature that it has been trained to detect, an edge of a particular size and orientation.

ML - Expert Systems

- Computer programs designed to replicate and improve on performance of human experts in specialized domains - Start with human experts → write a program that codifies their collective knowledge - Create a decision tree by analyzing large databases of examples and deriving rules that can be used to classify new examples ex. MYCIN for diagnosing infectious diseases (600+ heuristics rules) which accuracy of 69% ex. Expert Systems for diagnosing Psychological Disorders. Can also sometimes predict the trajectory of the disease Not completely refined/accurate yet Need to collect data from multiple kinds of scanes (fMRI, MRI, PET) to effectively classify disorders ex. Online Mortgage Wizards

Reinforcement Learning

- Depends on feedback signal - Feedback signal doesn't tell network exactly what it has done wrong; instead network is driven by a reward signal - The job of the network is to maximize the reward, but it is not told how to do that - It has to work out for itself which outputs are most profitable ex. having computer play 1 million rounds of pac-man until it masters how to win

Computer Based Personality Judgements

- Machine learning can make a lot of predictions based on Facebook likes - Had higher validity when predicting life outcomes such as substance use, political attitudes, and physical health than friends had - Even outperformed self-rated personality scores for Facebook activities, substance us, field of study and network size; and was comparable in predicting political attitudes and social network characteristics

Biological Connection to Small-World Networks

- Mammalian Brains are also small-world networks - Most nodes make only a short-range connections to one another and tend to be clustered into densely connected modules - In 2001, Latora and Marchiori analyzed data on cortical connections in cats and monkeys and found that there were only between two and three steps separating neurons - This allows an animal is to respond rapidly to a stimulus - say, to the sight of a predator or prey since each synaptic link would add additional response time - Small-world networks also allow for neuroplasticity - researchers found that the brain of a person with schizophrenia tends to be measurably less of a small- world network - modules aren't as densely connected

Biological Plausibility of Neural Networks

- Neural network units are all homogenous, vs. there are many types of neurons - Brains are nowhere near as massively parallel as a typical NN since each cortical neuron is connected to ~3% of neurons in surrounding sq mm - The cortex is organized in cortical columns that consist of populations of highly interconnected neurons with similar response properties (as many as 200,000 neurons) vs. Even the most complicated ANNs rarely have more than 5,000 units - There is no evidence that anything like back-propagation takes place in the brain - Most neural networks are supervised networks that require detailed information about the extent of the error at each output unit, vs. very little biological learning seems to involve this sort of detailed feedback

ID3

- Statistical measure of informativeness of each attribute (info gain) - Algorithm works out how well each attribute organizes examples - Assigns attribute with highest info gain to first node on tree - Process is repeated - Have to feed with with data that has already been interpreted - can't use raw data such as photographs

Learning in Single-Layer Networks

- The more that you activate a neural network the more likely it will randomly fire Ex. the more you think about a thought the more and more it will appear in your head "Neurons that fire together wire together"

Monism

- There is only one kind of substance in the universe Idealism: Everything including the material world is actually the mind Materialist: everything that exists including mind is physical, everything that happens in mind is just the brain Aristotle: diff shapes of clay of your mind takes on are the different mental states

Small World Networks and Neuroscience

- any network where one can get from any single point to any other point in only a small number of steps even though the total number of elements may be huge - Called the "six degrees of separation" idea - for instance, that all people are six or fewer social connections away from each other

Captcha

- based on assumption that its hard for machines to work with raw data ex. Humans can read distorted text but bots can't

Turing & the Turing Test

- developed computational model of mind - mind is algorithmic or rule-based calculation process for info processing Turing Test Hypothesis: if an observer is communicated with a machine and cannot tell the difference between it and a human than the computer is genuinely intelligent

Cerebral Cortex

- forebrain - best distinguisher from animals - outermost layer of grey matter - convolutions of cerebal cortex increase surface area of brain --> more info can be stored Four Lobes: A: Frontal lobe - judgement, speaking, muscle movement, personality/social behavior B: Parietal - somatosensory, touch, temp, spatial awareness C: Occipital - visual processing, shapes, colors, patterns D: Temporal - auditory processing, language, memory formation, HC for long-term memory

Fuzzy Logic

- idea of probablistic representation Traditional Logic: something can be represented by having a value of 1 or 0 Fuzzy Logic: you can have a value that's anywhere between 0 and 1 (partially true and mostly false values) ex. used for automatic braking in cars - fuzzy logic can be combined with fuzzy rules to make decisions

Lashley Planning and Organizing Complex Behavior

- larger actions are broken in simpler plans - as opposed to behaviorism (linked responses), activities require top-down or bottom-up processing Task analysis: breaking larger tasks into hierarchy of smaller tasks Subconscious info processing: unknowingly place tasks within hierarchy even though we consciously are aware of high-level plans

Multilayer Neural Networks

- network's knowledge is distributed across the relative strengths of the connections between different units and is constantly changing, rather than being encoded in discrete symbol structures - processing is parallel : flow of info through network is determined by what happens in all of the units in a given layer -- but none of those units are connected to each other GOFAI - highly structured and sharply defined - ex. playing checkers, decision trees ANN - perceptual and involve recognizing patterns - ex. patterns in past tense of english

Artificial Neural Networks (ANN)

- nodes are organized into layers in same way that human neural networks are - weights are attached to connections between pairs of units in adjacent laters that determine overall behavior of network (similar to way excitatory/inhibitory neural connections of various strengths connect to particular neuron) - positive weight = excitatory, negative = inhibitory - multiply each input by its weight gives the strength of signal at each synapse. If the activation levels are large enough to read bias term --> will fire. - bias term = a number between -1 and 1 and indicates what the weight sum need to be before the node activates (similar to firing threshold of neuron)

Correlates in Reinforcement Learning

- reinforcement learning is most common type of learning in animal kingdom - many parallels between reinforcement learning algorithm and deep neural networks that use reward processing in human brain involving dopamine

Tolman Implications on Learning

1. Learning isn't just stimulus-response associations but has mental processes 2. Info is somehow stored/processed about environment rather than chained sequences of conditioned reflexes

Application of Deep Learning with Cats/Dogs

1. feed computer pictures labeled as dogs or cats without extra data 2. data is sen through layers of deep learning network corresponding to different layers of abstraction (ex does this part of pic have a brown spot --> is there a nose in this part of the picture?)

Two Types of Consciousness Theory

Access Consciousness (A Consciousness) : - accessibility to info, prefrontal + parietal, - "easy problem" since explaining in computational or neural terms how an organisms access + deploys info Phenomenal Consciousness (P Consciousness): - how and why we experience the world as we do, posterior hot zone likely involved - "hard problem" since its difficult to uncover why/how sentient organisms have qualia or phenomenal experiences - Frank Jackson's Mary room thought experiment

Overview - Subcortical Structures of Forebrain

A: Basal Ganglia - Sensation B: Motor Cortex - movement C: Prefrontal Cortex - Judgement D: Thalamus - Pain, All incoming and out-going info gets routed through here E: Nucleus Accumbens - Reward F: Hippocampus - Memory G: Cerebellum - Coordination H: Visual Cortex - Vision

Limbic System

A: Limbic Cortex - older part of Brain/Evolutionary remnant B: Hippocampus - Important for Learning and Memory C: Amygdala - Emotions, especially fear and aggression

Alpha Go/Alpha Go Zero

AlphaGo: - Created by Google's deep mind research group - Beat the world's leading human experts at game of Go - Go has 10 ^ 360 different possible moves - Uses a mix of supervised learning and reinforcement learning - Trained on database of 30 million moves from an online server using supervised learning - received explicit feedback on how successful it was - Once reached high level of playing strength → reinforcement learning AlphaGo Zero - Learned very quickly to beat previous interaction of AlphaGo - No supervised learning Within 40 days able to beat all existing versions of AlphaGo and was computationally much more efficient **takes more risks comparatively to humans

Neuroanatomical Orientations

Anterior: toward the front Posterior: toward the back Dorsal: top head or back Ventral: toward bottom of skull or front surface Lateral: toward side of the body away from side Medial: toward the middle of body, away from side Ipsilateral: located on same side of body Contralateral: located on opposite side of body Coronal: slice through brain parallel to forehead Sagittal: perpendicular to ground and parallel to temporal lobes Transverse: parallel to the ground

Other Section of Forebrain

Basal Ganglia: - OCD = increase activity in caudate nucleus Thalamus: Caudal to basal ganglia Relay station for neural messages: directs messages to sensory receiving areas in the cortex and transmits replies to the cerebellum and medulla All incoming and out-going info gets routed through the thalamus Amygdala Beneath thalamus Controls autonomic nervous system Controls anterior and posterior pituitary glands Organizes behavior related to survival: fighting, feeding, fleeing, mating

Learning without reinforcement Tolman Experiments

Behaviorist thought that learning was purely due to stimulus-response associations (requires reward/punishment) Tolman #1: rats in maze, 1 group guided through, other left to run around in it. Both were equally successful when released later —> Latent Learning: Rats can learn without bodily responses/reinforcement Tolman #2: rats trained in maze, faced with change in starting place/blockades —> behaved as if they had a map to work out best route —> rats had a map-like knowledge not just memorized sequence of bodily responses of left/right turns.

Dualism

Belief in existence of both mental and physical substances, mind and brain are two distinct things

HIndbrain/Brain Stem

Brain system: Oldest part and central core of brain Starts where spinal cord swells as it enters the skull Reticular formation Large network of neural tissue in central part of brain stem Sleep, arousal, attention, various vital reflexes Pons: Bulge in brain stem important in sleep, arousal, sensory analysis and movement Medulla: Base of the brainstem. Controls vital functions: heart rate, breathing, BP Spinal Chord: Little brain near read of brain stem Helps coordinate voluntary movement and balance

Neuro Imaging Techniques

CT /CAT scan: x-ray photos taken from different angles combined by computer to represent slice of brain/body (structure) Pet Scan: Radioactive sugar is injected and scan reveals where in brain radioactive substance winds up when person is engaged in a particular task. Very expensive and not very safe. Can do subtraction paradigms: subtract two images to see what is left over (reading aloud vs. silently) (function) fMRI: Determines level of blood flow in different parts of the brain by measuring relative levels of oxygenated blood. (function) Knife-edge scanning microscope (KSEM): Photos are taken of very thin slices of brain tissue then compiled into a composite picture. Produces 3D reconstruction down to cell level Not in vivo. (function)

Other Effects of Dreams

Cognitive Development Theory of Dreams: - neural activation is associated with dreaming aids cognitive development - peak REM occurs in 30 wk features who spends almost 24 hrs in this state and is when they develop the most Memory Consolidation view: - participants trained on visual search tsk or word learning and deprived of REM sleep performed worse that control participants - REM sleep increases following stressful experiences or periods of intense learning Unconscious problem solving: - stories of people discovering solutions to thought problems while asleep

Learning Without Reinforcement

Cognitive Maps (Tolman Experiments) Over justification Effect: Children who have bee offered reward for playing with specific markers were less likely to play with them in free play because associated them as work rather than as play. Rewards can undermine motivation

Marr's Model of Visual Processing

Complex pattern of unstructured stimuli in visual field and interpret them into representations that can then serve as input to more complex cognitive functions First stage: Image projected onto retina is analyzed in terms of intensity of areas of light and dark adjacent regions of sharp contrast indicate the presence of edges and contours (Raw primal sketch) Second stage: Processed again to produce a representation of the object that includes is surfaces and layout (2.5 D sketch) Third stage: Transformed into 3-D sketch in which the axes of symmetry and elongation link the object parts

Blindsight

Condition: people who are blind in one or both visual fields due to damage to their visual cortex can nonetheless 'guess' significantly above average the change in identity or location of particular objects. Or Particular emotions expressed by face in a photo in front of them Proposed explanation: - Second pathway of visual perception that doesn't go through the visual cortex - Simply makes a very short loop through limbic system from the superior colliculus directly to the emotional/instinctual centers of the brain - Proposed mechanism for "intuition"

Global Neuronal Workspace Theory of Consciousness

Consciousness emerges when incoming sensory info inscribed on a blackboard is broadcast globally to multiple cognitive systems various types of unconscious processing are associated with deficits in these areas (Hypnosis, lucid dreams, vision neglect) Prefrontal + Parietal Lobes are most important for consciousness

Linguistics + Language Analysis

Deep Structure: meaning of a sentence due to rules of sentence Surface Structure: actual organization of words in a sentence/superficial appearance Implications: Language involves stored bodies of info which can be analyzed/manipulated algorithmically

Two Pathway Model of Visual Processing

Dorsal Stream: system of interconnected regions of visual cortex incolved in perception of spatial location (WHERE pathway) Ventral stream: system of interconnected regions of visual cortex involved in perception of form (WHAT pathway)

Measuring Activity of Populations of Neurons

EEG: Electrodes plastered on the scalp measure activity on the surface of the Brain. Cheap, widespread, used for epilepsy and tumors. (function) MEG: Sensors placed on scalp measure magnetic field generated by the brain's electrical activity. Good spatial resolution. (function)

Brodmann Areas

Early studies in structure/function of brain cerebral cortex = divided into segregated areas with distinct neuronal populations. Each part is distinguished is distinguished by types of cells and density of those cells 52 different cortical regions

Evidence for GWT

Ex. Repressed memories Dissociative Identity Disorder (DID): - two personalities have diff brain waves, vital signs, and hormonal levels -some personalities are aware of others (hierarchy of personalities) - at risk: trauma + women, can't remember before 12 well Explanation: repression/dissociation in prefrontal cortex disengages processing in hippocampus Experimental Studies: brain is more active when avoiding recalling memory than during recall

Lesion Studies

Examination of destruction of brain tissue through stroke, accident or surgery Earliest Methods of studying functional specialization ex. Phineas Gage, railroad worker, frontal lobe damage, violent, aggressive, loss of self control

Motivation and Learning

Expert Tutors: best tutors will do anything to avoid telling students they are wrong. Learning is 90% motivational. Growth vs. Fixed Mindset: Study gave control group a presentation on study skills while another group got a special module on how intelligence can be improved. Those with special module had dramatic improvement in study skills and grades. Pushed themselves harder. Tried new things

Subdivisions of Brain

Forebrain, midbrains hindbrain subdivisions correspond to different stages of evolutionary development

Dreams

Freud's Wish Fulfillment Theory: dreams are a safety valve that discharges otherwise unacceptable feelings/desires Modern Psychodynamic View: - dreams = what is going on in our lives, or hidden impulses, desires and underlying conflicts - dreams give clues to solution of these conflicts/problems Hobson + McCarley's activation-synthesis hypothesis: - dreams serve no purpose — just side effect of random firing of neurons that serve to develop and preserve neural pathways through stimulation

Types of Brain Waves

Gamma: high frequency, exchange of info, conscious or REM Beta: irregular, evident in frontal lobe, arousal/alertness Alpha: associated with relaxed awake state Delta: low frequency, deep sleep and loss of consciousness Theta: transition between sleep/wakefulness, associated with deep meditation, hypnagogic state, creativity and memory retrieval

Selectivity/Invariance

Highly selective representation: Ex. difference between white samoyed dog and white wolf Invariant: - ignores conspicuous differences Ex. baby samoyed is same as an older samoyed or a samoyed laying down Deep learning can accomplish selectivity/invariance challenge because each layer codes the input in terms of increasingly complex features

Multiple Intelligence Model

Howard Gardener Types: Linguistic, Logical-Mathematical, Spatial, Musical, Bodily-kinesthetic, interpersonal, intrapersonal, existentialist, naturalist Based on neurological data on brain function researched savants to create categories dance might be only activity that develops all kinds of intelligences at once

Intelligence

Intelligence relates to consciousness and ability to understand meaning (not IQ) Prior to 21st: intelligence = memory, learning, problem solving and IQ tests Problems with IQ: - cultural and economic bias (children in higher income families hear 30 million more words by age 3 than those less privileged) - assumes IQ is innate and fixed which undermines motivation - ineffective at predicting success in life, high IQ people are not more likely to have better marriages, raise children, or have overall better mental/physical health

Introverts vs. Extraverts and Stimulation

Introverts are more sensitive to stimulation: - salivate more to lemon juice - retain elevated HR for longer period after exposure to bad odors - more sensitive to pain and learn more quickly from punishment

Newell and Simon General Problem Solver

Means ends analysis: Evaluate difference between current and solution state. Identify transformation that reduces the difference between the two states. May need to divide problem into sub problems and move backwards ex. hobbits/orcs problem

Tri-Level Hypothesis

Mental/Artificial info processing events can be evaluated on three different levels Computation- Highest level - what does problem what is output of system, what is purpose/reason for process Algorithmic- Programming Level - what info are being used to solve problem. - need formal procedure that specifies how data is to be transformed Implementation - Lowest Level - where is hardware that is being used? How can representations/algorithms be realized physically - ex. computer hardware or human brain/neurons

Music Therapy

Music is the only sensory experience that can activate all areas of the brain simultaneously Alzheimer's --> improve memory Children --> emotional self expression, social skills, motor functioning Cancer patients --> reduction in anxiety, pain and tiredness

What is Cognitive Science?

Neuroscience - Biological machinery, case studies/imaging Psychologist - Mental processes/perception, scientific method CS - modeling processes in computers, coding models Philosophers - nature of mind, deduction Evol Bio/Anthro - howthe mind evolved

Neural Correlated of Consciousness (NCC)

Neuroscience Def of Consciousness: consciousness results from coordinated activity of a population of neurons Minimal set of neural events --> specific conscious experience

How many Hidden layers Are needed? Do more Hidden Layers produce better AI?

Not necessarily. Most of time only 1 is necessary.

Searle Chinese Room Thought Experiment

Overview: PSS is False Experiment: suppose a person who doesn't speak Chinese is given pieces of paper with Chinese symbols. They also have a manual that tells the person which symbol they should return depending on what they receive. Technically person is responding in Chinese but doesn't truly understand Chinese

Bayesian Probability

P(A) = probability of event A. (probability of disease in general population) P(B) = probability of event B. (Probability of a positive test result) P(A|B) = probability of observing event A if B is true (probability of having disease if test is positive) P(B|A) = probability of observing event B if A is true (probability of testing positive if one has disease) P(A|B) = P(B|A)P(A) = 0.99 x 0.01 = 50% P(B) 198/10,000

Physical Symbol System Hypothesis (PSS)

Simon and Newell All intelligent behavior involves transforming physical symbols according to rules Hypothesis: any PSS has a necessary means for general intelligence Implications: anything capable of intelligent action is a PSS, thus machines with PSS can be intelligent

Posterior Hypothesis for Consciousness

Posterior part of brain is most influential for consciousness Evidence: In brain surgery you can remove large parts of the prefrontal cortex and the conscious experience will remain largely in tact (aside from some emotional/motor problems). However, if you remove even a small section of posterior you can cause loss of entire class of conscious content (ex. facial recognition, motion, color)

History of Cog Sci

Previously: Studied the mind without any regard for brain -- changes in the 70s with neural activity tech Ancient Greeks: method of loci/memory palace, human thinking = mechanical manipulation of symbols 1879: Wundy's psychology lab, used introspection Behaviorism: Psychology should study relationship between stimuli and behavioral responses (mind banished completely) 1950: Cog Psych, Research on memory, development, linguistics + CS

Combining Neuroimaging Techniques

Problems: - can be a lot of noise in system Voxel = 3d version of pixel Smaller voxel → higher spatial resolution, lower signal strength Larger voxel → increase range of differ types of brain tissue occurring in each voxel **No technique for studying large pop of neurons that is high temporal and spatial resolution (inverses) Applications of brain imaging - Diagnosing disorders - Identify which patients are more likely to respond to which treatments Removing tumors without invasive surgery, but instead zapping all the connective pathways to that area of the brain to remove it.

Hierarchical Organization of Human Visual System

Process: - input from retina is conveyed via optic nerve ---> superior colliculus of brainstem --> lateral geniculate nucleus - LGN projects to area V1 (primary visual cortex) --> striate cortex - in area V1 info processing begins. Neurons in V1 (feature detectors) are sensitive to low-level features such as orientation and direction of movement. Also looks for edged/contours - V1 --> V2 which processes more complex features such as complex edges, shape, and depth. - Depth = determined by retinal disparity that objects fall on slightly diff location on two retinas. This is called stereopsis (thumb demo) - Extrastriate cortex processes additional feature of info such as movement, frequency, disparity and color

Consciousness

Psychology Def: awareness of our environment and our perceptions, images and feelings Philosophers: Emergent property of physical brain that may not fully be explained by understanding its component parts

Communication Within a Neuron

Resting Membrane Potential: occurs bc various ions are located in different concentrations in fluid inside/outside the cell. Extra cellular fluids = rich in Na and Cl. Intracellular fluid = rich in Ka and anions Action Potential = brief electrical impulse that provides the basis for conduction of info along an axon. All or none phenomenon. Once action potential is triggered in an axon —> fires. Neurons represent intensity by their rate of firing Single Cell Recording: Micro-electrode is placed close to a neuron to record the discharge of action potentials in the cell Used to identify neurons that respond to particular stimuli Ex Mirror neurons: fire both when a person performs a specific action as well as when they observe that action.

Spatial vs. Temporal Resolution

Spatial = scale on which image/test give precise measurement Temporal = time intervals to which they are sensitive

AI Formulations

Strong AI: Machines posses artificial general intelligence (AGI). Just as Smart as humans Applied AI: Teaching machines to learn how to master complex tasks. (face ID, playing Go, spotting tumors etc. )

Supervised vs. Unsupervised Learning

Supervised: networks receives explicit feedback on how successful it is Unsupervised: network doesn't receive feedback instead it learns to detect patterns in the data - uses convolutional neural networks

Autonomic Nervous System

Sympathetic: Arousal/expenditure of energy, adrenal medulla (epinephrine/norepinephrine), increased heart rate, rise in blood sugar, goosebumps Parasympathetic: relaxed state, acetylcholine, stored energy, salivation, gastric intentional motility

Zip Zap Technique

Technique: - Developed by Gulio Tononi - zapped patients' scalp with intense magnetic puls - causes brief electric current in neurons with reverberate across the cortex exiting + inhibiting other neurons - network of EEG sensory record the signals —> movie - data from movie are zipped into a computer file - zipping gives an estimate of complexity of brain's response Significance: can help determine difference between locked in vs. comatose people Effects: our tech today prevents us from getting to the necessary level of II that is necessary for consciousness (computer's zip file is minuscule to brain) Criticism: methods are crude proxy of brain's II

Functionalism

We need to emphasize function to understand how things produce their effects Identity of mental state is determined by its causal relation to sensory stimulation, other mental states and behavior Significance: if you can understand what causes pain you can recreate it in AI. Also suggest each kind of mental state is due to a type of neural state

Coding of Color

Trichromatic Theory: retina has 3 types of color receptors each especially sensitive to red, green, and blue Color blindness: - lack function red or green sensitive cones - see world in shades of yellow/blue - red/green looks yellowish - visual acuity is normal

Building Decision Trees

Two Processes: 1. Interview experts --> develop tree based on answers 2. program produces tress on its own by recognizing patterns from huge databases

Visual vs. Verbal Learners

Visualizers: more activation in fusiform gyrus area associated with processing visual representations. Verbalizers: more activation in supramarginal gyrus associated with linguistic processing Math without words: - Most students are not verbal learners -Completely visual (non verbal) software for math vastly improve math proficiency for students in Orange County - Language free approach can also improve language skills

Perceptrons and Perceptron Convergence Rule

Whenever the network produces the wrong output for a given input, it is told that it has made an error It then changes the weights and/or the threshold in response to the error

Analog vs. Propositional Processing

analog = pictorial representation of info propositions = descriptive representation of info

Central Nervous System

brain + spinal cord

Neuroplasticity

brain's capacity for modification, as evident in brain reorganization following damage (esp in children) and in experiments on the effects of experience on brain development - People born deaf from birth have enhanced visual processing - auditory cortex is responsive to touch/visual input bc brain repurposes these areas - Children given hemispherectomies show to have remarkable recoveries - Constraint-induced therapy: rewire brain by restraining a fully function limb, forcing patients to use the 'bad' hand or leg → rerouting

Brain and Electrical Mapping

cognitive functioning involves coordinated activity of different brain areas electrical activity is a good index of activity in neurons

Opponent-Process Theory

color sensitive receptor cells respond in opposing center-surround fashion to pairs of primary color theta are excited by light in the center and inhibited by light that surrounds also excited by light of a particular cone but inhibited by light of the opposing color sincerity = white, yellow excitement = red, orange competence = blue sophistication = black, punk purple Ruggedness = brown.

Integrated Information Theory (IIT)

consciousness arises from neural integration and complexity of how neurons are set up

Somatic System

controls voluntary movements of skeletal muscles

Lateralization of Brain

corpus callosum: bundle of axons that interconnects corresponding regions of association cortex on each side of brain Left Hemisphere: - information analysist, recognition -logical, linear, language Right Hemisphere: -synthesis of information -patterns, perceive emotions, non-verbal holistic *LH projects on right side *RH projects on left side

Aphasia

difficulty in producing/comprehending speech caused by brain damage Brocas: - expressive aphasia, speech is meaningful but halting, labored and ungrammatical - Associated with frontal lobe damage Wernicke's - Fluent aphasia, loss of ability to understand speech to produce meaningful words - word salad - Do not appear to recognize they cannot produce meaningful speech/understand others → paranoid - Damage to left posterior superior temporal lobe

Tact Tracing

early form of exploring anatomical connectivity Process: marker chemical such as radioactive amino acids/horseradish to particular brain region and look to see where marker ends up allowing us to identify where cell project to (through microscope) Limitations: - cannot be done in vivo - don't carry info about the direction of the info flow - studied almost completely independently of cognitive functioning such as how different brain regions might form circuits to preform particular info-processing tasks

Evidence for Analog Processing

ex. Imagery and Rotation Studies - asked people to mentally flip images to see if they were the same -took humans varying amounts of time even though quantity of info remained the same - implies that humans use analog processing rather than propositional (like computers) Ex. Showing people picture of tiny rabbit vs. large rabbit and asking if it had two paws - took people longer to identify two paws in smaller rabbit bc distance to info was larger

Evidence for Propositional Code

ex. Looking at a shape, asked to create a mental image of it and then being asked about it later. Many people can't identify the relevant part unless it was highlighted earlier ex. Rabbit/Duck image. Some ambiguous figures are difficult to reinterpret in a mental image

Sleep and Sensory Stimuli

may effect learning and memory Ex. participants learned 2 simple piano melodies then took 90 min nap While slept one of melodies was quietly played on repeat Able to play cued melody more accurately than the other one when they awoke Ex While participants slept were exposed to cigarette smoke and rotten fish During following week, people who had smelled mix of both odors hat 30% less cigarettes than those in control groups Conditioning particularly effective during stage 2 REM

Subliminal Perception

messages presented below one's absolute threshold for conscious awareness may influence behavior. ex. emotionally positive scene subliminally flashed before subjects when viewing a slid of faces led to more positive rating of those faces later on. ex. advertisers trying to use this to sell products. However didn't make ppl buy more beef just more hungry. Drawbacks: - hard to ensure exact response - stimuli act as an emotional priming and not direct link to thoughts/decisions

Auditory Perception: Effects of Music Cognition

music can influence memory, decision making, and other cog processes Classical, Jazz, Blues, Folk Music lovers = more verbally intelligent, open to experience Country, Pop, Religious music lovers = more cheerful, outgoing, conscientious

Optogenetic

process: inserting opsin genes into neurons. Causes neurons to manufacture light-sensitive opsin proteins and incorporate them into the membrane. Neuron can then be activated by particular wavelength of light. ex. Mice + seeing vertical stripes. Tagged neurons associated with certain memories. Induced those neurons artificially through light waves to make new associations -- implants false memories.

Brain Stimulation Techniques

rTranscranial Magnetic Stimulation: Pulse of magnetic energy sent through coil on surface of skull --> electrical firing of neurons beneath scalp. Can enhance/disrupt neural activity. Applied to Left Pre-Frontal Cortex may reduce depression symptoms. Penfield Studies: Electrical stimulation of areas of brain associated during open brain surgery while patient is fully conscious. ECT: Anesthetic paralyzes muscles. Jolt of electricity applied 1/25th of a second triggering brieg seizure. May increase norepinephrine by calming neural centers overactive in depression. While more rapid improvement then SSRIs, excessive use causes damage, memory damage, relapse tDCS: weak current applied to scalp on left dorsolateral prefrontal cortex. Safer the ECT Deep brain Stimulation: Surgery that places neurostimulator which sends impulses through implanted electrodes to specific targets. Parkinsons severe depression.

Peripheral nervous System

sensory/motor neurons that connection CNS to rest of body Split into Autonomic and Somatic systems

Synesthesia

stimulation of one modality --> perception experience in another Color Grapheme: seeing specific letters/number in specific colors London Animation Study: People overwhelmingly chose the synesthetes animation as matching the music better. We all unconsciously link together music and vision but synesthetes are consciously aware of these things Theories: 1. Brain architecture of synesthetes is equipped with more connections between neurons, causing the usual modularity to break down 2. Feed-backward connections that carry info from high level multi sensory areas of the brain back to single sense area are not properly inhibited. Inhibition is disrupted allowing different senses to be jumbled

Machine Learning

subset of AI that allows computers to "learn" - progressively improve its performance by creating new algorithms to produce a desired output based on structure data "Feature engineering" ex. deciding between if pic is a cat or a dog Problems: - humans need to label info first -hard to determining best way to categorize/label data - data can be labeled in many different ways

Deep Learning/Representation learning

subset of machine learning involving numerous layers of algorithm Computer doesn't need to be provided with structured data programs are written to do their own feature engineering

Artificial Intelligence

tries to design computer models that accomplish the same cognitive tasks as humans do

Priming

unconscious activation of particular associations in memory