lesson 5

Question 1

Limitation of the Rescorla-Wagner (RW) Model

Answer 1

RW makes incorrect predictions for certain phenomena, particularly those involving inhibition and latent learning.

Question 2

Goal of Subsequent Learning Models

Answer 2

To address the shortcomings of the RW model and provide more accurate explanations of learning phenomena.

Question 3

Focus of the Lesson on New Models

Answer 3

Understanding the main principles of each model and how they differ from RW, rather than detailed mathematical reproduction.

Question 4

Commonality Among Newer Models

Answer 4

Many are modifications or extensions of the original Rescorla-Wagner framework.

Question 5

Current Status of Learning Models

Answer 5

Observation: Despite significant advancements, no single model perfectly describes all aspects of learning.

Question 6

Attention in Learning

Answer 6

mportance: The amount of attention paid to a stimulus can influence the speed and extent of learning about associations involving that stimulus.

Question 7

Term: Salience (α) in the RW Model

Answer 7

Representation of Attention: RW incorporates the idea of attention through the salience parameter (α), which is assumed to be a fixed property of the CS.

Question 8

Term: Limitation of Salience in RW

Answer 8

Issue: RW assumes salience is constant and does not change based on learning or experience during the experiment. Real-world attention can vary.

Question 9

Term: Mackintosh Model - Key Principle

Answer 9

Principle: Attention to a CS increases if that CS is a good predictor of the US (high contingency) and decreases if it is a poor predictor. Animals learn to “tune in” to relevant cues and “tune out” irrelevant ones.

Question 10

Term: Mackintosh Model - Main Difference from RW

Answer 10

Difference: The salience parameter (α) is not fixed but is updated on each trial based on the CS’s predictive accuracy.

Question 11

erm: Mackintosh Model - Explanation of Blocking

Answer 11

Explanation: Similar to RW, in Phase 1, CS A becomes a good predictor of the US, increasing attention to A. In Phase 2 (AB + US), because A already predicts the US, B is a redundant predictor. Attention to B will not increase (or may even decrease), leading to less learning about B compared to the control group where A was not pre-trained.

Question 12

Term: Mackintosh Model - Explanation of Latent Inhibition

Answer 12

Explanation: In Phase 1 (A [+nothing]), CS A predicts nothing. According to the Mackintosh model, attention to A (α
A
​
) will decrease because it has low contingency with the US (which is absent). In Phase 2 (A + US), the experimental group starts with lower attention to A compared to the control group (which had no Phase 1). Lower attention leads to slower learning about the A-US association.

Question 13

Pearce-Hall Model - Key Principle

Answer 13

Principle: Attention to a CS increases if the outcome (US) on the previous trial was surprising (high prediction error: ∣λ−∑V∣). Animals pay more attention to cues that have recently been associated with unexpected outcomes.

Question 14

Pearce-Hall Model - Main Difference from RW

Answer 14

Difference: The salience parameter (α) is not fixed but is updated on each trial based on the magnitude of the prediction error on the previous trial.

Question 15

Pearce-Hall Model - Explanation of Blocking

Answer 15

Explanation: Similar to RW, in Phase 1 (A + US), A becomes a good predictor, reducing prediction error. In Phase 2 (AB + US), the presence of A means the US is largely predicted. The prediction error is smaller compared to the control group (B + US). Because attention to B depends on the prediction error, B will receive less attention and thus less learning in the experimental group.

Question 16

Pearce-Hall Model - Explanation of Latent Inhibition

Answer 16

In Phase 1 (A [+nothing]), there is no US, so the prediction error is consistently zero. As a result, attention to A (α
A
​
) remains low (or does not increase). In Phase 2 (A + US), the experimental group starts with low attention to A, leading to slower learning of the A-US association compared to the control group which starts with a higher level of attention to the novel CS A.

Question 17

Positive Transfer (IDS Experiment)

Answer 17

: Prior learning about a relevant stimulus dimension facilitates learning a new association involving that same dimension. Explained by the Mackintosh model through increased attention to the relevant dimension.

Question 18

Negative Transfer (Jane & Julia Experiment)

Answer 18

Prior learning about a CS hinders the learning of a new association involving that same CS, often when the US changes in magnitude. Explained by the Pearce-Hall model through decreased attention due to reduced surprise in the initial phase.

Question 19

Hybrid Attentional Models

Answer 19

Concept: Models that combine the principles of both the Mackintosh model (attention to good predictors) and the Pearce-Hall model (attention to surprising outcomes), suggesting that both types of attentional processes contribute to learning.

Question 20

Elemental Models of Learning

Answer 20

Models (like RW, Mackintosh, Pearce-Hall) that assume animals learn associations between individual CSs and USs.

Question 21

Configural Models of Learning

Answer 21

Models that propose animals learn associations between the entire configuration of stimuli present and the US, rather than individual CS-US associations

Question 22

Pearce (2002) Configural Model - Key Idea

Answer 22

When multiple CSs are presented together, an association is formed between the compound stimulus (the configuration) and the US.

Question 23

Role of Generalization in Configural Models

Answer 23

Importance: Configural models heavily rely on generalization, the tendency to respond to stimuli similar to those trained on. Differences between training and test stimuli lead to a weaker response.

Question 24

Configural Model - Explanation of Overshadowing

Answer 24

The experimental group learns an association with the AB compound. At test, they are presented with A alone, which is different from the training stimulus. This difference leads to a generalization decrement, resulting in a weaker response to A compared to the control group trained only with A.

Question 25

Configural Model - Explanation of Blocking

Answer 25

In Phase 1 (A + US), an association forms between A and the US. In Phase 2 (AB + US), the experimental group learns about the AB compound. At test for B, the relevant training was with AB. The control group trained with just B in Phase 2 has a more direct association with B. Differences in the training configuration might lead to subtle differences in response, but configural models can also explain blocking through generalization and learned irrelevance of the added cue.

Question 26

Configural Model - Explanation of Latent Inhibition

Answer 26

Configural models, especially those focused on compound stimuli, may not directly address latent inhibition with a single CS. Predictions might be similar to RW, as there's no configuration to learn about in Phase 1. Some configural models might suggest that repeated exposure to A alone allows the animal to learn about "A in a no-US context," which then interferes with learning "A in a US context" later, due to the change in the "configuration" of A plus the context.

Question 27

Conditioning to Context

Answer 27

Learning an association between the background environmental cues (e.g., the experimental chamber, time of day) and the US.

Question 28

Contingency of Context with US

Answer 28

Characteristic: Usually weaker than that of specific CSs because the context is often present even when the US is not.

Question 29

Importance of Context in Learning

Answer 29

Significance: Context can serve as a cue itself, and learning can be specific to the training context. Changes in context at test can lead to a generalization decrement.

Question 30

Generalization Decrement

Answer 30

Definition: A reduction in the conditioned response when the test stimulus is different from the training stimulus (including changes in context).

Question 31

Comparison of Learning Models

Answer 31

Observation: Each model (RW, Mackintosh, Pearce-Hall, Configural) has strengths and weaknesses, explaining some data well while failing on others. A single, perfect model of learning does not yet exist.

Question 32

Reason for Imperfection in Learning Models

Answer 32

Explanation: The complexity of learning itself. The scientific process: new models make predictions leading to new experiments, which can reveal limitations of existing models and drive the development of new ones.

Question 33

Common Basis of Most Learning Models

Answer 33

Underlying Principle: The formation and manipulation of associations between mental representations of events (CSs and USs), whether these are individual elements or configurations.

Question 34

Mackintosh model Key principles:

Answer 34

Learning involves changes in the associative strength between CS and US (similar to RW). The salience/attention (α) of a CS is not fixed but changes over trials. Attention to a CS increases if it is a good predictor of the US (high contingency). Attention to a CS decreases if it is a poor predictor of the US (low contingency).

Question 35

mackintosh model- main differences from RW

Answer 35

RW assumes a fixed salience (α) for each CS. The Mackintosh model proposes a dynamic salience (α) that is updated based on the predictive value of the CS.

Question 36

mackintosh model- explanation o f blocking

Answer 36

Phase 1 (A + US): Attention to A increases as it becomes a good predictor of the US. Phase 2 (AB + US): Because A already reliably predicts the US, B is a redundant predictor. Attention to B does not increase (or may even decrease) because it doesn't add new predictive information. Test (B): Due to the lack of attention and learning about B in Phase 2, the response to B is weaker in the experimental group compared to the control group.

Question 37

mackintosh model-explanation of latent inhibition

Answer 37

Phase 1 (A [+Nothing]): CS A predicts no US. Attention to A decreases because it has low contingency with the US. Phase 2 (A + US): The experimental group starts Phase 2 with reduced attention to A compared to the control group (which had no Phase 1). Test (A): Lower attention to A in the experimental group during Phase 2 leads to slower learning about the A-US association and thus a weaker response to A at test compared to the control group.

Question 38

Pearce-Hall model Key principles:

Answer 38

Learning involves changes in the associative strength between CS and US (similar to RW). The salience/attention (α) of a CS is not fixed but changes over trials. Attention to a CS increases if the outcome (US) on the previous trial was surprising (high prediction error). Attention to a CS remains low for CSs that consistently predict the outcome, whether the US is present or absent.

Question 39

pearce-hall model- main differences from RW

Answer 39

RW assumes a fixed salience (α) for each CS. The Pearce-Hall model proposes a dynamic salience (α) that is updated based on the magnitude of the prediction error on the previous trial.

Question 40

Pearce-Hall modelExplanation of Blocking:

Answer 40

Phase 1 (A + US): A becomes a good predictor, reducing prediction error on subsequent A+US trials. Attention to A remains relatively stable (not necessarily increasing). Phase 2 (AB + US): Because A largely predicts the US, the addition of B results in a smaller prediction error compared to the control group (B + US). Test (B): Lower prediction error during learning about B in the experimental group leads to lower attention allocated to B and thus less learning about the B-US association, resulting in a weaker response to B at test.

Question 41

Pearce-Hall modelExplanation of Latent Inhibition:

Answer 41

Phase 1 (A [+Nothing]): There is no US, so the prediction error is consistently zero. As a result, attention to A (α A ​ ) remains low (or does not increase). Phase 2 (A + US): The experimental group starts Phase 2 with low attention to A because it was not associated with surprising outcomes in Phase 1. Test (A): Low initial attention to A in the experimental group during Phase 2 leads to slower learning about the A-US association and thus a weaker response to A at test compared to the control group.

Question 42

configuration models key principles

Answer 42

Animals learn associations between the entire configuration of CSs present and the US, not just individual CS-US associations. Generalization plays a crucial role: responses to a test stimulus depend on its similarity to the training configuration. The "stimulus" being learned about can be a unique pattern or combination of cues.

Question 43

configural models main diff from rw

Answer 43

RW is an elemental model, focusing on individual CS-US links. Configural models emphasize the learning of a holistic representation of the stimulus compound.

Question 44

configural model- explanation of blocking

Answer 44

Phase 1 (A + US): An association forms between the configuration of A (plus the background context) and the US. Phase 2 (AB + US): The experimental group learns an association with the AB compound (plus context). The control group learns about the configuration of B (plus context). Test (B): The experimental group's training involved the AB configuration. Testing with B alone represents a change in the stimulus, leading to a generalization decrement and a weaker response compared to the control group whose Phase 2 training was more similar to the test stimulus B.

Question 45

configural model- expl of latent inhibition

Answer 45

Phase 1 (A [+Nothing]): Repeated exposure to A alone allows the animal to learn about the configuration of "A in a no-US context." Phase 2 (A + US): Learning about "A in a US context" is hindered because the animal has already learned something about A predicting the absence of the US in a similar context. This creates interference or reduces the "novelty" of A, impacting learning in Phase 2.