Pavlovian Conditioning: Mechanisms & Theories
PSYC 351: Fundamentals of Learning
Chapter 6: Pavlovian Conditioning: Mechanisms & Theories
Explanations for Pavlovian Conditioning
Mechanisms of Conditioning:
Conditioning results in the activation of a "US center" in the brain, which represents the Unconditioned Stimulus (US) activated by the Conditioned Stimulus (CS) (also known as S-S learning).
Alternatively, conditioning establishes a new stimulus-response (S-R) connection between the CS and the Conditioned Response (CR).
Option 1: S-S Learning
CS substitutes for US
EX: The bell causes drooling because it elicits memories of the US
Option 2: S-R Learning
A connection is formed between the CS and the new CR
EX: The bell causes drooling because drooling has become habitual when the bell is presented (e.g., the US isn’t even important after the learning process is completed)
What is the Basis of Classical Conditioning?
Overview:
Classical conditioning can be understood through two avenues: S-S learning and S-R learning.
S-S learning implies that the CS acts as a substitute for the US, while S-R learning indicates a direct connection formed between the CS and the new CR.
Evidence in Support of S-S Learning
Key Hypothesis:
If the CS serves as a substitute for the US, then the nature of the conditioned response (CR) should be determined by the US.
Prediction:
CS’s conditioned with different US’s should elicit different types of conditioned responses
Example: Autoshaping with grain vs. water
Jenkins & Moore (1973)
However, sometimes…
the nature of the conditioned response is determine by the CS
Timberlake & Grant (1975)
Hungry rat in cage, then stimulus rat is put into the cage
Participant rat is angry, it is an intruder in its home cage and it is hungry
When the stimulus rat is put into the cage, food arrives and participant rat nom nom nom
Each time a new stimulus rat is put into the cage, food arrives and participant rat nom nom nom (e.g., intruder rat shows up, food shows up — like Santa bringing gifts, if Santa were a different rat every time)
After learning — they go over to stimulus rat, are happy to see the stimulus rat, and show affiliative/liking behaviour to that rat (e.g., grooming and stuff)… what they DON’T do is try to eat the rat
Learn that the stimulus rat becomes predictive of food… according to S-S learning, they should either engage in goal-seeking behaviour, or treat the rat as if it were a substitution for food… but rather, the rat treats the stimulus rat… as a rat!
Behaviour is specific to what the CS is
Is it S-S or S-R Learning?
Controlled experiment to test this: Devaluate the US
Hungry rats
Train them to associate a 10-sec light with food delivery
Measure conditioned orienting response to light
Rescorla (1975)
S-R learning: once pavlovian conditioning is learned, the US shouldn’t matter - US doesn’t matter, the light orentation (or drooling) should be automatic
If it’s S-S learning: CS and US are linked forever, if CS becomes extension of food, ad rats dont give a hsit about food anymore, S-S learning would say that they shouldn’t care about the light (e.g., shouldnt make a conditoning response) - if they aren’t hungey, they shouldndt drool (if they dont want food, why prepare)
RESULTS — First group, with US devaluation, don’t respond as much to light which indicates food… so the conditional response has been altered!
Models of Classical/Pavolvian Conditioning
It isn’t the only or best model, teacher thinks it does things well, but there are other models of how and why classical conditioning occurs
Timing effects, blocking, LI, and associative bias (or belongingness) show that the “rules” for CC are complex
US Modulation Approaches (Rescorla Wagner is an example of this) - as you leanr, what’s changing is your underastanding of the US
How organism changes learning to the food
CS modulational Models (Macintosh model is example) — how organism changes its learning towards the predictors
How organism changes learning to the bell
The Rescorla-Wagner Model
When you make predictions of when US will appear, surprise factor disappears — indicates that the learning is done. In other words, high surprise indicates minimal learning, and low surprise indicates maximal learning has occurred.
Assumptions
When a CS and a US are paired, an association will be formed
The learning of this association is a curvilinear function
The effectiveness of the US in conditioning a CS is determined by how different the US is from what is expected
Expect a small US and get a larger one → Excitatory conditioning
Expect a large US and get a smaller one → Inhibitory conditioning
The expectation of the US is related to the conditioned properties of all the stimuli that precede the US
What it does
It calculates surprisingness to calculate associate strength (predictability)
As associate strength increase, surprisingness decreases (as prediction increase, we are less surprised with the results)
V= associative strength (how well the CS predicts the US)
λ (called lambda) = what you receive (shock, food, etc… it is the US)
Under normal circumstances, λ equals 1
If I give you more of the US than expected, λ becomes 2
If I give you less o the US than expected, λ becomes 0.5
If there is no US, λ becomes 0
V is supposed to approach λ, until it predicts accurately what follows it. V changes in predictability until it becomes a perfect predictor of the US. In mathematical terms, if V = λ (or CS associate strength = US), then V is a perfect predictor of λ
More pairings of CS and US leads to learning — As V increases ver trials it approaches λ and the occurrence of the US becomes less surprising
Purpose is to quantify the surprise (difference between V and λ) and to use it to calculate overall amount that has been learned, but also to calculate amount of learning per trial!
The first trial is the most you’ll ever be surprised… that means the first trial is where the most learning occurs — ∆V (delta V) is change in associate strength for each trial… it is how much learning is occurring for each trial, and it decreases with each trial (because surprise decreases for each trial)
K = a catch-all variable. We aren’t expected to know how to calculate K. We gotta understand it though. It is literally everything that has to do with the experimental variables crammed into one level — age of subjects, species of subjects, gender of subjects, independent variable, dependent variable… K has to be between 0 and 1, but cannot be 0 nor 1 (e.g., if it was 0, no learning would be able to occur, and if it were 1, perfect learning would occur every time)… the higher the K value, the faster learning will occur… for instance, taste aversion K will be very high… a bell predicting food will show a lesser K value because it’ll take longer
This is the complete equation: \Delta V = k(\lambda - V) .
Predictions of the Rescorla-Wagner Model
Conditioning
Trial 1:
Trial 2:
Trial 3:
Blocking
Phase 1:
Across conditioning, the tone will become a perfect predictor of the food
At this point, VT = λ
Phase 2:
Keep in mind that VL is V of light, VT is V of tone
∆VL = k(λ - V)
However, V = VT + VL
VT = λ & VL
Therefore, V = λ + 0 = λ
∆VL = k(λ - V)
= k(λ - λ)
= k(0)
= 0
Unblocking
∆VL = k (- V)
However, V = VT + VL
λ & VT = VL = 0
λ + 0 = λ
Therefore, V =
VL = k (- V)
∆λ
λ - λ)
= k (
= k (0)
= 0
Phase 1 Tone: Food Salivation
CS1 US UR/CR
Phase 2 Tone + Light : Food Salivation
CS1 CS2 US UR/CR
Page 17 Pav Cond: Mech & Theories Chapter 6
Rescorla-Wagner Predictions and Phenomena
Blocking:
Observed that associative strength does not accrue to a new CS if paired with a predictive CS already demonstrating strong associative strength.
Unblocking:
When a new CS is introduced, if the potential US's total associative value increases, the new CS can gain strength, reflecting
\Delta V_L = k(2\lambda - V)
Overexpectation:
Noted decrease in responding to a previously effective CS when presenting two CSs predicting a single US.
Conditioning calculations yield negative associative strength, indicating no efficiency of the CS.
Prediction:
Conditioned Stimuli (CSs) paired with different USs should elicit different types of CRs.
Example: Autoshaping with grain vs. water.
Reference to research: Jenkins & Moore (1973).
Conditions Leading to S-R Learning
Counter-Argument:
The nature of the CR can sometimes be determined by the CS itself.
Referenced study: Timberlake & Grant (1975).
Testing the S-S vs. S-R Learning Theory
Experimental Design:
Devalue the US in a controlled experiment with hungry rats:
Phase 1: Pavlovian conditioning with 10-sec light paired with food delivery.
Phase 2: Devalue the US by providing varying amounts of food to assess the conditioned orienting response to the light.
Rescorla-Wagner Model
Key Components:
The Rescorla-Wagner Model serves as a mathematical representation of Pavlovian conditioning.
Key assumptions include:
Association Formation: A CS and a US form an association when paired.
Curvilinear Learning: The learning curve is a curvilinear function.
Expectation Dependence: The effectiveness of a US in conditioning a CS is determined by the discrepancy between the US received and what is expected.
Example: If a smaller US is expected but a larger one (excitatory conditioning) occurs, vice-versa for inhibitory conditioning.
Expectation of US: Related to the conditioned properties of all stimuli prior to the US.
Mathematical Representation
Change in Associative Strength:
The associative strength of a CS increases until it perfectly predicts the US.
Denote associative strength as V and maximum strength of US as λ. The relationship governed by:
\Delta V = \lambda - VWhen the CS perfectly predicts the US, V reaches its maximum.
The amount of surprise is calculated by the difference between the expected and actual US: \Delta V = \lambda - V.
Conditioning Trials and Strength Increment
Rate of Conditioning:
Initial US prediction is unexpected, leading to an increase in the associative strength of the CS as training progresses.
Conditioning trial calculations involve:
Assume \lambda = 1 and k = 0.30:
Trial 1: \Delta V1 = k(\lambda - V0) = 0.30(1 - 0) = 0.30.
Trial 2: \Delta V2 = k(\lambda - V1) = 0.30(1 - 0.30) = (0.30)(0.70) = 0.21.
Trial 3: \Delta V3 = k(\lambda - V2) = 0.30(1 - 0.51) = (0.30)(0.49) = 0.15.
Observations show greater learning occurs early in the pairing of CS and US.
Predictions of the Rescorla-Wagner Model
Example of Conditioning Performance:
Note how associative strength builds progressively during trials.
Graphical Representation: Associative strength (V) vs. Trials, indicating strengths at various steps.
Rescorla-Wagner Predictions and Phenomena
Blocking:
Observed that associative strength does not accrue to a new CS if paired with a predictive CS already demonstrating strong associative strength.
Unblocking:
When a new CS is introduced, if the potential US's total associative value increases, the new CS can gain strength, reflecting
\Delta V_L = k(2\lambda - V)
Overexpectation:
Noted decrease in responding to a previously effective CS when presenting two CSs predicting a single US.
Conditioning calculations yield negative associative strength, indicating no efficiency of the CS.
Inhibition and Extinction Models
Inhibition: Implies negative associative values for a CS- in trials where the US is presented.
Extinction Process: Occurs when the CS continues without the US, leading to diminished CR, modeled mathematically:
\Delta V = k(0 - \lambda) = -k\lambda.
Take Home Message
Recognition that classical conditioning is complex and not solely relying on contiguity.
The residual impact of US expectancy on learning in CS-US associations is paramount.
The Rescorla-Wagner model demonstrates predictive capacity and explains various classical conditioning phenomena.
Study Tips
Memorize the equation: \Delta V = k(\lambda - V).
Note that if k\lambda > 0, V gains associative strength; if negative, V loses strength.
Acknowledge that λ can vary based on conditioning phases, affecting learning expectations.
Recognize constant k will always be between 0 and 1.
Homework Assignments
Complete calculations for ΔV and V across multiple trials as instructed.
Ensure discussions about the homework to clarify concepts in the forum.