Generalization: treat more than one object as basically the same thingExperiment: train pigeons to pick a yellow key and variate them and see if they respond to it correctly or notImportant skill to have- don’t have to learn absolutely every situations you run into- you can apply similar bxs to similar situations; don’t have to learn every single situation—adaptive response!Discrimination is NOT a failure to discriminateSOR: cognitive aspect- some connections in btwn stimulus and response that accounts for certain phenomenon on how generalization worksDiscrete-component representationID input and outputTrain with yellow light, you strengthen that response, none of the other connections are strengthen for every stimulus, there must be a unique receptor to it that’s connected to an output (monosynaptic connection)- strengthened by repeated exposureConnectionist modelsDistributed representationsIntroduce intermediate layer (internal layer) btwn stimulus and response (fits with SOR model); never know what brain is doing and only look at stimulus, but presumption is that something is going on in the mind in the organism itself (receptor for yellow that has a connection through 2 synapses to an output), this same input neuron has a distribution to an internal layer with elements around it (same is true for other colors- yellow and orange) Yellow trained input is activated strongly the direct pathway, but it’s also affecting the nearby pathways as well; strength of connection to Hebbian processes increases by a value of 1 and the others is .3  different kind of curve (looks like a bishop hat -___-) closer to the generalization gradientGeneralization is default! Discrimination is different!DiscriminationDiscrimination has these characteristicsmust have sensory apparatus to distinguishsalience: know stimulus strength and intensitydeliberate training!Discrimination trainingThe peak shifts- goes away from the original trained stimulus if the two stimuli (yellow and green) are similar, the strongest response later in a generalization gradient shows a change in how you react to the stimulus AND the stimulus around itBOX pictures:Picking larger of the two items = correctSeparable features: brightness and hue- add white to a reddish color = pink and other variations, but you still recognize the red in the pink color and can distinguish btwn the brightnessIntegral features: there are things you can’t distinguish very well  brightness (intensity) and saturation, soft red = pink but it can be soft bc there’s some white in it or bc it’s a very dim color – difficult to distinguishHeight and width seem to interact with one another (don’t make judgments about how tall someone is independent of their width unless there’s a comparison)Negative patterning:Simplest example: Exclusive “or” configurationTrain a rat so that tone comes on + puff of air  learned bxal response; if you give same animal at the same time mixed up trial with a light, they’ll learn a learned response to the light (not blocking); still show response to light but not at the same time (no combination) learned how to distinguish exception of this situation is why it’s called neg patterningReal example:Right turn signal and left turn signal and emergency blinker  pattern of lights in the stimuli makes a differenceAdding additional symptoms on the negative patterning model, you have to combine all of themCorrelationsSimilarity-based generalizationEquivalence by meaningA = B, B= C then A = CSimultaneous conditioning  animals don’t learnAll the stimuli are added and then intermediate step is added (evaluation step)A1, B1, X1 || A1+B1, B1+X1 || A1- shockA2, B2, X2 || A2+B2, B2+X2 || A2- foodWhat happens to the response to X, a remote stimuli?Acts as if they are equivalent  transitive lawA, B, and X can be 6 different friends and are fun to be around, certain combinations of them are fun too, but you never have one of the friends in the other combinationsWhat might be some neural connections that are like this?Lec. 17 Generalization and Discrimination 11/08/2011Generalization: treat more than one object as basically the same thing-Experiment: train pigeons to pick a yellow key and variate them andsee if they respond to it correctly or not-Important skill to have- don’t have to learn absolutely every situations you run into- you can apply similar bxs to similar situations; don’t have to learn every single situation—adaptive response! -Discrimination is NOT a failure to discriminate-SOR: cognitive aspect- some connections in btwn stimulus and response that accounts for certain phenomenon on how generalization worksoDiscrete-component representationID input and outputTrain with yellow light, you strengthen that response, none of the other connections are strengthen for every stimulus, there must be a unique receptor to it that’s connected to an output (monosynaptic connection)- strengthened by repeated exposureoConnectionist modelsDistributed representationsIntroduce intermediate layer (internal layer) btwn stimulus and response (fits with SOR model); never know what brain is doing and only look at stimulus, but presumption is that something is going on in the mind in the organism itself (receptor for yellow that has a connection through2 synapses to an output), this same input neuron has a distribution to an internal layer with elements around it (same is true for other colors- yellow and orange)  Yellow trained input is activated strongly the direct pathway, but it’s also affecting the nearby pathways as well; strength of connection to Hebbian processes increases by a value of 1 and the others is .3  different kind of curve (looks like a bishop hat -___-) closer to the generalizationgradient-Generalization is default! Discrimination is different!Discrimination-Discrimination has these characteristicsomust have sensory apparatus to distinguishosalience: know stimulus strength and intensityodeliberate training! -Discrimination trainingoThe peak shifts- goes away from the original trained stimulus if the two stimuli (yellow and green) are similar, the strongest response later in a generalization gradient shows a change in how you react to the stimulus AND the stimulus around it-BOX pictures:oPicking larger of the two items = correct-Separable features: brightness and hue- add white to a reddish color = pink and other variations,