27 Nov 2020
School
Department
Course
Professor
1
PSYC 325 Lecture Notes – Chapter 6
Generalization and Discrimination Learning
• Behavioral processes
• Brain substrates
• Clinical perspectives
Using the Past to Cope with the Present
• We often need to respond to stimuli that are completely novel.
– Will I like snowboarding? I’ve never tried it before!
• Rather than respond randomly, we can consider prior experience with stimuli we have
encountered.
– Snowboarding reminds me of skiing. I have tried that, and loved it.
Using the Past to Cope with the Present
There are two basic ways of using these past experiences:
2
Generalization versus Discrimination
• Three potential approaches to study:
• Similar stimuli that predict similar outcomes
– And consequently, dissimilar stimuli predict different outcomes
• Similar stimuli that predict different outcomes
• Dissimilar stimuli that predict similar outcomes
Similar Stimuli Can Predict Similar Consequences
3
First, pigeons learned to peck a yellow light for food.
Next, they were presented with novel lights of different colors.
Rate of responding was recorded for the trained (yellow) and test colors.
(Guttman and Kalish study)
Similar Stimuli Can Predict Similar Consequences
• Generalization gradient—graph showing how behavioral response changes
with respect to physical changes in stimuli
• Responding is highest for physically similar stimuli – gradually
declines for less similar stimuli
Similar Stimuli Can Predict Similar Consequences
• There is a balance between generalization and discrimination
– Very similar stimuli treated similarly to training light: generalization
– Very dissimilar stimuli (e.g., 520 nm) treated as completely different: discrimination
The Challenge of Incorporating Similarity into Learning Models
• Animals clearly respond to stimuli based on similarity with past experience.
• How can “similarity” be captured in a formal model of learning?
The Challenge of Similarity in Learning Models
• Discrete-component representation—each individual stimulus is represented by its own node or
“component.”