Use LEFT and RIGHT arrow keys to navigate between flashcards;
Use UP and DOWN arrow keys to flip the card;
H to show hint;
A reads text to speech;
62 Cards in this Set
- Front
- Back
paradigmatic approaches |
symbolic (rule based) sub-symbolic (connectionist) |
|
what is ACT-R |
a production-system cognitive architecture designed to redict human behavior by processng info (cognition, visual attention, movement etc. ) and generating behavior models many cognitive phenomenon |
|
types of knowledge |
declarative - conscious - facts procedural - unconscious |
|
chunks of declarative knowledge (2) |
type - category (bird) slots - attributes (color, size) |
|
basic program structure for a production |
if - goal then - subcoal |
|
productions in general comprised of (2) |
conditions actions |
|
conditions |
- dp on declarative knowledge (chunks) - &/or sensory input - specify the goal, and number of chunks - often tests contents of buffers |
|
actions |
- can alter declarative knowlege - initiate actions - produce changes in buffers |
|
Architecture of ACT-R modules --> buffers what are these relationships (4) |
intentional (n/k) - goal buffer DLPFC declarative (temp/hippo) - retrival VLPFC visual (occ) - visual Par manual (motor/cerebellum) - manual motor |
|
Architecture of ACT-R productions |
matching (striatum) selection (pallidum) execution (thalamus
|
|
modules devoted to
|
idetifying objects controlling hands retrieving declr info keeping track of goals
|
|
role of central production system |
respods to info which is deposited as chunks in buffers then fed forward for central processing |
|
visual buffers (2) |
dorsal 'where' path - object locations ventral 'what' path - object identities |
|
info processed in parallel or serial? |
mixed |
|
parallel processing: |
visual system - whole viusal feild declarative system - retreival of mems |
|
what are the 2 serial bottle necks? |
content buffer limited to 1 declarative chunk so: - one mem retr. at a time/ - one object encoded from visual feild
one production is selected to fire ea. cycle |
|
hybrid cognitive archetecture consists of (2) |
symbolic production sub-symbolic production |
|
sub-symb. parallel processign implemented by _____ controls many of the _____ processes |
equations symbolic |
|
productions and chunks have ______ parameters which reflect _____ |
subsymbolic past content |
|
Activatsion of delclarative memorie to what degree does ACT-R makes chunks active? |
to degree tht they will be useful/are relevent in particular moment |
|
conflict resolution for when multiple productions may match (but only 1 may fire) dp on the _________ utility f(x) |
subsymbolic utility fx
|
|
sub-symbolic utility function estimates (3) to select proudction with highest ___ |
for a given production fx estimates: - probability that current goal will be acheived if fires - relative cost (time to acheive goal) - benefit (value of goal) highest utility |
|
learning in ACT-R - symbolic (2) - sub-symbolic |
symbolic - declarative : new chunks - procedural : production compilations sub-s - rational analysis (optimized enviro stats, bayes rule) |
|
in the stick-building problem, what are the strategies |
- undershoot - overshoot - hill-climbing |
|
production compilation |
combining 2 existing rules if A --> B, if B --> C compiles if A --> C |
|
result of production compilation is this is refered to as the _______ of practice |
more efficience, speed power law of practice |
|
compilation with past-tense model predicted: |
faster prfrm on irregulars (regular = defalt) - contradictory to neural network findings |
|
what does W represent |
the attentional weighting of elements that are part of current goal |
|
why vary W? |
to represent individual differences |
|
hybrid cognitive architecture chuncks and productions are _____ controling _____ |
symbolic components overall info flow |
|
hybrid cognitive architecture chuncks have sub-s __________ while productions have sub-s __________ |
activations utilities |
|
learning can involve
|
- acquiring new hchuncks/ prodct. - find -tuning sub-s parameters |
|
limitations |
- learing rule and parameters is difficult - cant learn from scratch - extensive engineering rq - autoomus dvlp seems out - many wrong predictions (past-tense latencies) |
|
connectionism |
network of units and weights - unit computs weighted sum of inputs |
|
modify weight to ++____________ |
reduce error |
|
why must brain be a parallel processor? |
rapid computation performed by sluggish units |
|
computational properties of brain |
robust flexible approximate parallel compact and efficient |
|
tranlate into neural net: neuron activity synapse synaptic reception threshold |
unit activation weight sum of products (activation * weight) S-shaped function |
|
translate into psych equivalents - pattern of activation across net - connection weights - adjustment of weights |
= active/working memory = LT memory = learning |
|
back-propogation network structure |
output units ^ hidden units ^ input units |
|
activation function is ________--- |
sigmoid 1/(1+e^-x) |
|
3 Boolean problems |
AND - both true - linearly seperable OR - 1/2 true - lineraly seperable XOR- only 1 true - lineraly non-seperable |
|
how are inputs calculated? |
unit type (0/1) * weight) |
|
prospect theory differs from EV theory |
ps replaed with subjective decision weights |
|
gains vs losses in prospect theory |
gains - concave - risk averse losses - convex - risk seeking |
|
reference point |
customary wealth, at origin x = 0 |
|
loss aversion |
steeper for gains than losses |
|
problems with prospect theory |
- predicts preferences that are never observed - limited to binary - poor predictions when more possible outcomes |
|
desicion making biases |
allais stochastic dominance preference reversals similarity attraction compromise |
|
allais paradox ex: |
:: a) 100% win of 1m b) 89% win of 1m and 10% win of $5m |
|
stochastic domminance |
:: a) 85% $98m, 5% 90m, 10% 12m b) 90% 98m, 5% 14m, 5% 12m preference for A>B hw, remove items with same value and you see preference reversal |
|
preference reversals: choice and price |
- would choose lower risk gamble option but, - would assign higher price to higher risk gamble option |
|
preference reversals: min selling price vs max buying price |
would choose lower risk gamble but would assign higher price to higher risk gamble framing effects |
|
violatiosn of independence from irrelevance |
- similarity - attraction - compromise |
|
similarity |
A: IQ = 60 motivation = 90 B: IQ = 78, motivatio = 24 C: IQ = 75, motivation = 29
B when [A.B] A when [A.B.C]
|
|
attraction effect |
3rd item that is similar to another item but slightly inferior will ^^ attractivness of that item |
|
compromise effect |
3rd item is an intermediate, will compromise and pick that item |
|
computer models capturing paradoxes abound, currently the most effective one is ______ |
decision field theory shows preference emergence, as a function of time |
|
self feedback in DFT |
explains primacy/recency |
|
lateral inhibition in DFT |
explains context effects |
|
weigts |
depend on perceived prob that prospect I delivers outcome j |
|
limitations to DFT |
hand designed and highly engineered no learning no autonomous neowrk crct arbitrary specifications lots of randomness but still the champ! |