Download Rescorla-Wagner Theory: Understanding Classical Conditioning's Underlying Processes and more Lecture notes Cognitive Neuroscience in PDF only on Docsity!
Chapter 5: Finishing up Classical
Conditioning
Underlying Processes : Rescorla-Wagner Theory
Lecture Outline
- Underlying processes in Pavlovian conditioning
- S-R vs. S-S learning
- Stimulus-substitution vs. Preparatory-response theory
- Compensatory response model
- Rescorla-Wagner model
- Practical applications of Pavlovian conditioning
- Understanding the nature of phobias
- Treating phobias
- Aversion therapy
Problem with the S-S theories
• While they can explain simple conditioning
phenomena, they can’t explain
Rescorla-Wagner Model
• Began looking more in-depth at what is
happening to the actual association between
the CS and the US during the process of
conditioning (af ter each trial).
• Driven to find an explanation f or blocking
phenomenon.
Rescorla-Wagner Theory
• These concepts were incorporated into a
mathematical f ormula:
- Change in the associative strength of a stimulus
depends on the existing associative strength of
that stis and all others present
- If existing associative strength is low, then
potential change
- If existing associative strength is high, then
very little change occurs
- The speed and asymptotic level of learning is
determined by the
Rescorla-Wagner Model
- The equation for the model:
∆V=k(λ - V)
- Where: ∆V = V = k = λ = “lambda” represents the maximum associative value that a CS and US can hold (the asymptote/max of learning) (λ - V) = surprise value of the US
- The equation is applied once for each learning
trial, to see how much learning will happen on
each trial.
∆V=k(λ - V)
- The learning curve:
- If CS-US pairings repeated the associative strength (V) increases
- Increase in V is not consistent over trials - Trial 1 – substantial - Subsequent Trails – progressively smaller (less surprise) - Eventually V approaches stable value (λ)
- ∆V – represents change in associative strength on a given trial
Trials ( n)
Trials ( n)
Associ ative Stre ngth
(V)
Associ ative Stre ngth
(V)
∆V=k(λ - V)
• Quantifying surprise:
- Focus on relationship between V and λ
- Beginning of conditioning V is much less than
•At the beginning, the participant does not expect US and considerable learning occurs
•Over trials, the occurrence of the US is progressively less surprising, and V approaches λ
•Index of surprise = (λ – V)
V
Trials (n)
Rescorla-Wagner Model cont.
- Both parameter values remain the same for successive applications of the same situation - e.g., trials
- Parameter values differ when equation is applied to different situations - e.g., different CS-US combinations, different contexts
Rescorla-Wagner Model cont.
• Evaluation:
- To calculate the model’s predictions for
learning on a given CS-US trial, need to
estimate values of k & λ
- Could run pilot test but extremely complex
(Hull, 1943)
- Can just use arbitrary values!!!
- Precludes quantitative data (e.g., how much saliva on a given trial)
- Can make qualitative predictions (e.g., whether saliva will increase or decrease on a given trial)
Rescorla-Wagner Model cont.
- Acquisition
- k = 0.30 (parameter for salience of the CS-US)
- λ = 1.00 (maximum associative value )
- V = associative strength on trial 1 = 0. Trial 1 ∆V=k(λ - V) = 0.30 (1.00 – 0.00) = 0. Trial 2 ∆V=k(λ - V) = 0.30 (1.00 – 0.30) = 0.
4 0.66 ∆ V = 0.30 (1.00 – 0.66) =
3 0.51 ∆ V = 0.30 (1.00 – 0.51) =
2 0.30 ∆ V = 0.30 (1.00 – 0.30) =
1 0.00 ∆ V = 0.30 (1.00 – 0.00) =
Trial V ∆ V = k ( λ –V)
Rescorla-Wagner Model cont.
- Extinction
- With repeated extinction trials λ will = 0 (maximum associative value )
- Use same parameters but insert λ = 0 Trial 1 ∆V=k(λ - V) = 0.30 (0.00 – 0.66) = - 0. Trial 2 ∆V=k(λ - V) = 0.46 (0.00 – 0.46) = - 0.
4 0.22 ∆ V = 0.30 (0.00 – 0.22) = -0.
3 0.32 ∆ V = 0.30 (0.00 – 0.32) = -0.
2 0.46 ∆ V = 0.30 (0.00 – 0.46) = -0.
1 0.66 ∆ V = 0.30 (0.00 – 0.66) = -0.
Trial V ∆ V = k ( λ –V)
After More Extinction Trials V = 0
Extinction
66 46 (^3222) (^20010) 1.
40
60
80
100
(^0 1 2) Trials 3 4 5 6
Vall
Overshadowing cont.
Dim S Br
Trial 1
Dim S Br
Dim S Br
Trial 2 Trial 3
Rescorla-Wagner & Blocking
° Phase 1 ° Group 2 the light (CS) perfectly predicts the shock (US) in phase 1 ° Conditioning reaches the asymptote ° Phase 2 ° Compound stimuli (Light + Tone) presented with US ° No learning to Tone because light perfectly predicts US
° Associative strength is shared between CSs
Group 2 Light : Shock [Light + Tone] : Shock Tone ???
Group 1 [Light + Tone] : Shock Tone ???
Phase 1 Phase 2 Test
0
10
2 0
3 0
4 0
Strength
of Fear CR to To
ne
Group 1 Controls
Group 2 Blocking
Blocking cont.
S
L
Trial 1
S
L
T S L
Phase 2 Trials
Group 2 Light : Shock [Light + Tone] : Shock
Group 1 [Light + Tone] : Shock
Phase 1 Phase 2 Test
S
L
Trial 2
Trial 3
etc
- Clearly, the trials in Phase 1 will result in
Problems with Rescorla-Wagner
- Model focuses exclusively on CS-US association but cannot account for other events before, during, or after the association is formed.
- Problem 1:
- CS preexposure produces slower conditioning to CS later (latent inhibition). - Example: play a tone a number of times before it is paired with a shock. (have a harder time conditioning the tone)
- Latent inhibition is not predicted by Rescorla-Wagner
- unless you assume that preexposure lowers the learning rate (k) by lowering salience.
