Class Note for PSYCH 100 at UMass(1)
Class Note for PSYCH 100 at UMass(1)
Popular in Course
Popular in Department
This 29 page Class Notes was uploaded by an elite notetaker on Friday February 6, 2015. The Class Notes belongs to a course at University of Massachusetts taught by a professor in Fall. Since its upload, it has received 13 views.
Reviews for Class Note for PSYCH 100 at UMass(1)
Report this Material
What is Karma?
Karma is the currency of StudySoup.
You can buy or earn more Karma at anytime and redeem it for class notes, study guides, flashcards, and more!
Date Created: 02/06/15
Concept formation Categories and concepts F Category set of entities things actions properties In the world 3 Concept our mental representation of a category Are categories arbitrary Or based on What is in the world Borge s Taxonomy of Fauna Taken from Lakoff Women F ire and Dangerous Things The taxonomy of the animal kingdom from The Celestial Emporium ofBenevolenZKnowledge On those remote pages it is written that animals are divided into a those that belong to the Emperor b embalmed ones c those that are trained d suckling pigs e mermaids f fabulous ones g stray dogs h those that are included in this classi cation i those that tremble as if they were mad j innumerable ones k those drawn with a very fine camel s hair brush 1 others m those that have just broken a ower vase 11 those that resemble ies from a distance J L Borges 1966 Otherlnquisitions What are the functions of concepts P Permit inferences I Allow reasoning about new things P Support communication gt Add new information about concepts v Permit economy of communication Classical approach to concepts P Concepts are arbitrary set by culture PThey have necessary and sufficient features defining features PHypothesis testing to learn defining features gt Example color categories gt Demonstrations classical concept formation experiments 4c m a Figure 83 p 269 Different abject aH Chairs Eleanor Rosch s position The Roschian revolution lit Observation 0 Wittgenstein you usually can t nd de ning features for categories lit Observation 1 an entity usually belongs to several different categories igt But one is BASIC Observation 2 categories have better and worse members 9 Categories don t have xed boundaries Basic Level Terms And superordinate and subordinate terms PTaxonomies of categories gt Animalmammaldogcollie r Artifactfurniturechaireasychair P Rosch s observations gt Basic level the preferred name gt Fastest to verify nameobject match I Learned rst gt Shortest v ASL single sign mm Supamydmmn Mm l x Table Ham V Yumk k I 39 x y J mam um mm mm W We summm Fm aquot W van Rm mm mm m in 12 Figure 810 p 279 Level uf categories furfurmture and mm Ruseh pruvmeu evidence furthe idea thatthe bash level s psyhuugmaHy mugged Basic level principles PBasic level the highest most inclusive level at which all instances share a large number of features PFeature listing Superordinate uniture few features gt Basic level chair many features gt Subordinate easy chair only a few more features PHemenway members of basic level categories have parts in common not so for superordinates Underlying principle P Jointly maximize informativeness and discriminability of categories gt lnformativeness what do you know given that you know that X is in category Y gt Discriminability how hard is it to tell that X is in category Y rather than category Y P A tradeoff gt thing easy to discriminate but not very informative gt son of Citation very informative but not easy to discriminate gt horse basic level jointly maximizes both 39 391 H H H H H H l39 391 H H l39 1 H H H H H H Last points about basic level P Expertise matters gt Dog experts as fast on subordinate term as basic term Their basic level may really be our subordinate term Distinctiveness and informativeness increased for them i i D i i i D Li Ci 4 I D D i i i i C i i Typicality and category membership W itliincategory structure PRelations between categories Basic level analysis PWithincategory relations Typicality and graded category membership Rate the typicality Fruits 7 Birds Apple T Robin 625 689 m 1 338 153 Olive 1 Flamingo 225 337 Grape 1 Seagull 513 l 626 Typicality effects Time to identify name as category member 33quot Time to identify picture as category member Ease of learning category memberShiP Name members of category frequency of naming Similarity ratings asymmetry less typical members are more similar to more typical members than VV Prototype models Alternative ideas of prototype 7quotIdealized abstract form Some kinds of instances Mental representation of ideal form or instances Similarity Feature overlap features produced by students Bird Features Robin Swallow Vulture Flamingo ies sings lays eggs is small nests in trees eats insects SIM TO BIRD 6 6 2 6 An application of typicality A Psych 100 trick P Or is it really thought transmission Thought transmission P I ll pick number between 10 and 50 Pit has to be odd Peach of its digits has to be odd Pits two digits have to be different gt 11 is not good D 15 is OK P I ll write it down and think about it P Then YOU write down what I m thinking 37 Exemplar theories of concepts These theories contrast with prototype theories PPrototype theory claim You lean some kind of abstraction some single mental representation of a category PExemplar theory claim You remember individual instances exemplars of concepts and compare new things to these remembered instances to categorize these new things Exemplar theories of concepts Their virtues 7 Explain typicality effects very well Permit mathematical models Supports quantitative predictions of choice learning Accounts for feature variation effects pizza VS quarter quot7 Accounts for feature covariation effects little singing birds vs big nonsinging birds Medin et al 1982 Medical diagnosis P EMWeight gain puffy eyes stiff muscles splotchy skin gt Terrigitis P RL Weight gain sunken eyes stiff muscles skin rash gt Mdosis P AM Weight loss sunken eyes stiff muscles skin rash gt Mdosis P L F Weight gain sunken eyes muscle spasms skin rash gt Terrigitis P J J Weight loss sunken eyes muscle spasms splotchy skin gt Midosis P S T Weight loss puffy eyes stiff muscles splotchy skin gt Tenigitis P plus 2 more Medin et al cont d Test phase P Weight gain sunken eyes stiff muscles splotchy skin v Terrigitis 67 P Weight loss puffy eyes muscle spasms skin rash v Teirigitis 75 P Weight gain sunken eyes muscle spasms slotchy skin v Midosis 72 P THE POINT Terrigitis is characterized by both stiff muscles and splotchy skin or by both muscle spasms and skin rash Midosis is characterized by both stiff muscles and skin rash or by muscle spasms and splotchy skin People pick this up pretty well But what they learn is COVARIATION between features not just features of a prototype Strengths and weaknesses of exemplar models P Strengths gt Accounts for typicality effects PLUS variability and covariability effects gt Permits development of explicit accurate mathematical models P Weaknesses gt Supporting data come from odd experiments that encourage instance memory gt People CAN form abstractions gt Model doesn t address main points of why we have concepts economical prediction gt Exemplar models depend on how you compute similarity 1MHHHHJ WV 1HHHHquot Concepts as theories Murphy amp Medin P Theory as glue that holds concept together P Theory as Whatever principles tell you Which properties should be important Which unimportant gt And WHY gt That is they tell you how to compute similarity Evidence for concepts as theories PFacilitation in learning concepts eg thing that can be used as a hammer prey vs predator PAbility to form ad hoc categories PBelief in essences essentialism PKeil transformation studies children s concepts More evidence for concepts as theories P Illusory correlation Chapman and Chapman PClinicans diagnoses of mental disorders Kim and Ahn gt DSM IV checklists of symptoms presumably theory free gt But diagnosis is categorization gt Individual clinician s diagnosis affected by hisher theory of the disorder Associative Theories of LTM 95 Network models 95 Connectionist models Collins and Qu lian expt Reaction Time to Verify P Propertyquesuons An oak has acoms7 r dxsmnce o gt Asprucehasbmnches 7 dstance gt Abxrchhas seeds 7 dxstanceZ P Categoryquesuons gt Amapwe samaple7 r dxstanceO gt Acedar 1 atree r dxstancel gt An elm ls aplanm r distanceZ gmuwu won Hanqu w uwmmu nmmmm Wu game 29 muumsm memm mum w w wwvwm WWW lvualawq mm mm g CollinsMamp Q illian xpts 222 44 my Ha mm m r mm F 3 vu m on My He mm W 1 manta w Wm Em C um mm m mu H Merver A mnl39y unn F WWW mum Aclmvy 393 mu mun Y r Am I J i quot r E HUM 3quot mm g mm mm m J a quotan E sun mm a mam Spreading activation PA metaphor a concept can be active to a greater or lesser degree PIS activation spreads to related concepts P Priming PDistance effects P Fan experimenm Fan experiments John Anderson Smith Adams amp Schorr 1978 P Learning materials gt Marty broke the bottle Marty did not delay the trip gt Herb produced sour notes Herb realized the seam was split gt six more pairs P Test materials gt TF Marty broke the bottle Herb did not delay the trip MEASURE RT accuracy Fan 2 links Broke the bottle Marty Did not delay the trip Fan Experiments phase 2 PAugmented learning materials gt Marty broke the bottle Marty did not delay the trip Marty was chosen to address the crowd gt Herb produced sour notes Herb realized the seam was split Herb painted an old barn I six more pairs P Test phase gt Same as before Marty broke the bottle etc Fan three links Broke the bottle Marty Did not delay the trip Was chosen to address the crowd Fan experiments P The time to verify the original facts was longer when the third fact about each individual was added PInterpretation The activation spreading from the Marty node was divided among more other nodes 3 not 2 Therefore these other nodes were activated more slowly The paradox of the expert Pthe start of the paradox learning some material slows you down in remembering other material and even causes you to forget the other material retroactive inhibition gt So who should have the hardest time remembering material I well the person who has learned the most other material an expert gt but experts seem to have the BEST memory for a topic not the worst The Paradox of the Expert cont Pthe start of the solution schema theory gt experts don t have random facts they have integrated schemas about their topic gt perhaps random facts interfere with one another but an integrated schema helps integrate and hold new facts Fan Experiments Phase 2 version 2 PLearning materials augmented with thematicfacts gt Marty broke the bottle Marty did not delay the trip Marty was chosen to Christen the ship gt Herb produced sour notes Herb realized the seam was split Herb played a damaged bagpipe P Adding this thematically third fact did not slow and may even have speeded veri cation RT Connectionist neural net models Simple models complex results P Cognitive tradition symbolic models gt Symbolic representations gt Information processing rules for operating on representatlons PNonsymbolic models gt Simplify theoretical assumptions gt Still get complex results Connectionist models Components of a connectionist model PNodes P Connections gt Strength of connections P Activation Axow AXONS FROM A OTHER NEURONS v DENDRITES gm 11 A chemati neumn A node 1 o A connection with a weight quotn Localist Connectionist Models PA node represents an entity stimulus response individual idea concept PActivation is transmitted between nodes 3 One node can either excite or inhibit another node P System provides a computational foundation for associative networks Input nodei Output node j 12 3 12 3 W33 Connection weights wij Delta rule perceptron convergence procedure etc Rule for adjusting weights based on instruction feedback Previous examples of localist models PWord recognition models gt Hierarchy of feature letter bigraIrL word detectors McClelland and Rumelhart connectionist model 7 With bottomup excitation from input opdown excitation from words and lateral inhibition mm 4 union nun1 min Human mm m mu sum mutt mom 5 Hamid llnu ly int nu in is on mquot tn 4mm 1mg m mmpi um nnly Mr Wm nm 04 me o n mom and not mi 039 Ivy in min m nll mm mm no Meow Ami om Alan mule Wino noimmnmngibottonnnti mdudmgl Q andS Thuwillmnlmltmd wuk mnm w m apprapnile bigmm dmvdon nn CDdn lumnhuwuu i wrl pnniou um mulvkulymrmpvmd m mm itimwnwn 2 wk mi weather oigmin dmm lo nun CQUXCSMRlumel pnmnna 0 win nit Mm m m wk inpin memequot om it 1 minim gimme dining leumfumm ntmvmoimv moo hy monk Sign warm and mm mm ram Wampum ma mm an alarm a are warm MM leuv Mam amt amt Wm rm Ma mama mam llama by mm an alarm mm M mer manna hem Coma m Wm 5 rpm W are ala mam M w maple mm Marmara Mullah Mama mmaarr rm Learning in a Localist Model P Hebbian learning gt Learn covariation in node activiation gt Rule strengthen connection between nodes that are simultaneously active Weaken other connections gt Result activation of one node predicts activation of othe nodes P Guided instructed learning gt Example set Weights in semantic net gt Delta rule strengthen Weights that result in activation of targ et output node weaken other weights 7 Barkpmpagallor algorithm allow strengtheningweakening to apply to earlier connections in net Distributed Models PParallel Distributed Processing PDP models gt Represented an object event 7 by pattern of activation across nodes 7 Not by anode as in alocalist model Memory set of activation Weights ofconnections betweennodes 7 Pemlits aparticular pattern of activation to develop given an input gt Underlying metaphor bmin shuct39ure 7 Node corresponds to nemon links correspond to synapses Cognmv Adivuy in Arti ml Neunl Network 203 OOOOW HIDDEN UNIYS mm ommm xnm cu39HP INPUT UNITS n5 1 A mph mwmk Connectionist models cont PA hybrid model the minerock detector P Hidden layers gt Increase compuhtional power PLeaming models gt Guided instructed learning 7 Delta rule r Backpropagazion i i O S s mwsn FREnuEch Hgm 15 Ptrnpluzl magmaquot wnl i hrge nulwnlk Advantages Disadvantages P Attractions gt Neural plausib ty gt Parallel processing gt Graceful degradation under damage gt Emphasize leaming P Disadvantages Ovel looks the stmetnre that the hiain DOES have Don t give insight into rules the mind follows Don t do as much abstraction as we seem to P Resolution gt Marr s analysis day 1 ofthe course symbolic model the algon39thmie level eonneetionist model the implementational level
Are you sure you want to buy this material for
You're already Subscribed!
Looks like you've already subscribed to StudySoup, you won't need to purchase another subscription to get this material. To access this material simply click 'View Full Document'