Presentation on theme: "Theories of Learning Chapter 4 – Theories of Conditioning"— Presentation transcript:
1Theories of Learning Chapter 4 – Theories of Conditioning PSY 402Theories of LearningChapter 4 – Theories of Conditioning
2Rescorla-Wagner Model Classical conditioning occurs only if the US (UCS) is surprising to the organism.If the UCS is already predicted by a CS, then it is not surprising – it is expected.When the CS predicts the UCS perfectly, no further learning occurs.The asymptote (lambda, l) is the point where the learning levels off (no increase in learning occurs).
34.1 Growth of associative strength (V) to a CS as a function of CS-US pairings asymptotebouton-fig jpg
4Parts of the Model DV = ab(l – V) V is the Associative Strength (amount of learning).DV is the change in learning (increase in Associative Strength.a and b are the salience of the CS and UCSl – V is the surprisingness of the US (the distance away from the asymptote).
54.2 The effect of US magnitude on learning (Part 1) Larger UCSSmaller UCSbouton-fig jpg
64.2 The effect of CS salience on learning (Part 2) Larger CSSmaller CSbouton-fig jpg
7Multiple Conditioned Stimuli (CS’s) The basic model explains changes in learning with one UCS and one CS.This doesn’t explain what happens during blocking and unblocking, with multiple CS’s.DV = ab(l – ΣV)When multiple CS’s are present, SV is the sum of the associative strengths of all of the CS’s (such as VN + VL).
8Blocking First a noise is conditioned so that VN = 1.0 Next a light is added. The formula predicts its associative strength:DVL = ab(l – ΣV)ΣV = VN + VLIf we assume that ab = .2 and VN is 0 because no learning has occurred yet, then:DVL = .2[1.0 – ( )] = 0
9Unblocking As before, a noise is conditioned so that VN = 1.0 A stronger US is presented with the new CS (VL).As before, the formula predicts its associative strength:DVL = ab(l – ΣV)ΣV = VN + VLAgain, we assume that ab = .2 and VN is 0 but now the stronger US is 2.0 instead of 1.0:DVL = .2[2.0 – ( )] = .2[1.0] = .2
10Extinction During extinction, the CS is presented without the UCS. This is the same as presenting a UCS with intensity = 0.The formula predicts the associative strength during extinction:DVN = ab(l – V) but l is now 0 (due to extinction)DVN = .2[0 – 1] = -.2The associative strength is decreasing.Use the decreased value for VN (1-.2) for the next trial.
114.3 Conditioning and extinction in the Rescorla-Wagner model bouton-fig jpg
12InhibitionDuring inhibition, a second CSL is presented that has never been associated with the UCS (V = 0).The formula predicts the associative strength for both CS’s:DVN = ab(l – SV) and DVL = ab(l – SV)DVN = .2[0 – ( )] = -.2DVL = .2[0 – ( )] = -.2V = VN + VL.
134.4 The conditioning of inhibition in the Rescorla-Wagner model bouton-fig jpg
14Protection from Extinction When extinction of an excitor takes place together with extinction of an inhibitor, the excitor is never fully extinguished.This is called protection from extinction.To fully extinguish an excitor, and to extinguish it faster, pair it with another excitor (another CS associated with the US).The model predicts both of the these results.
15Overexpectation Effect The value of a model is that it predicts new findings.If you pair two previously conditioned CS’s (excitors) on the same trial, V for each will decrease until SV equals l.This is because SV “overexpects” the UCS.Similarly, if a new CS (X) is added to the pair, it will become an inhibitor.
164.5 “Overexpectation” of the US: Kremer's two experiment designs (Part 1) bouton-fig jpg
174.5 “Overexpectation” of the US: Predictions (Part 2) bouton-fig jpg
184.5 “Overexpectation” of the US: Predictions (Part 3) bouton-fig jpg
19Contextual CuesContextual cues consist of everything in the environment in addition the CS and UCS.They cannot be ignored simply because the experimenter is not manipulating them.Whenever a CS or a UCS appears “alone,” it is still being paired with the context.When the context is considered another CS, then ideas about blocking explain learning.Zero contingency occurs because context is blocked.
20CS becomes an inhibitor 4.6 (A) Negative contingency between CS and US; (B) Zero contingency between CS and USCS becomes an inhibitorbouton-fig jpgNo learning occurs
21Comparator TheoriesAn alternative theory to Rescorla-Wagner proposes that the CS and UCS are associated and the UCS and context are associated.The two sets of associations are compared to determine the amount of responding to the CS.The comparison determines the responding, not the learning.Strengthening or weakening the context, after learning, affects the amount of responding, supporting the theory.
22Problems with Rescorla-Wagner It predicts that presenting an inhibitory CS without the UCS should lead to extinction, but it doesn’t.The model cannot account for latent inhibition (preexposure to the CS).Mackintosh demonstrated that animals learn to ignore redundant stimuli – the model doesn’t predict this learning.
23Less learning More learning 4.7 (A) Mackintosh-Turner experiment; (B) Results of exposure to LN-shock trialsLess learningMore learningbouton-fig jpg
24The Mackintosh ModelMackintosh proposed that the amount of learning depends on how much attention the animal pays to the CS.The attention to the CS is the a term in the Rescorla-Wagner model.Alpha increases when the CS is the best predictor and conditioning occurs to the best predictor of the UCS.
25Criticisms of the Mackintosh Model The model does a good job of explaining latent inhibition and its own criticisms of Rescorla-Wagner, but other problems arose.While attention is important, it doesn’t necessarily increase when a CS becomes the best predictor.Hall & Pearce showed that preexposure to a tone that was a good predictor of weak shock didn’t help learning when a stronger shock was used.
26Group 1 should have done better, but didn’t 4.8 (A) A Hall and Pearce experiment design; (B) Results of conditioning during Phase 2Less learningGroup 1 should have done better, but didn’tMore learningbouton-fig jpg
27Pearce Hall ModelAnimals don’t waste attention on stimuli whose meaning is already well understood.Instead, they devote attention to understanding new stimuli.For their model, the value of alpha depends on how surprising the UCS was on the previous trial.If the UCS is surprising, the CS is not well understood. Alpha is high when this occurs.
284.9 A rat orienting toward a light CS (Part 1) This is orienting behavior – the rat is paying attention to the lightbouton-fig jpg
294.9 Orienting to the light CS with a US pairing (Part 2) Rats paid more attention to the light when its meaning was unclear (Partial condition)bouton-fig jpg