week 4 learning Flashcards
Why is learning important
To make predictions about events in an environment and to control them. Learning exists to allow an organism to exploit and benefit from regularities in the environment
-must know what cue to pay attention to
How to identify if events are related
• Degree of contingency
–One method is to examine how often the two event co-occur
• Degree of covariation or correlation
–A second method is to consider whether the events appear together or independently
Classical Conditioning
Unconditioned stimulus -> Unconditioned response
US + Stimulus -> Conditional repond (stimulus is a conditional stimulus)
US then would evoke CR on its own
Operant Conditioning
Behaviouris shaped by the learner’s history of experiencing rewards and punishments for their actions.
Positive and negative reward
- Positive Reinforcement = An animal will learn to produce a behaviour if the consequence of doing so is receiving something pleasant.
- Negative Reinforcement = An animal will learn to produce a behavior if the consequence of doing so is stopping something unpleasant.
Positive and Negative punishment
- Positive Punishment = An animal will learn stop producing a behaviour if the consequence of producing the behavior is an unpleasant stimulus.
- Negative Punishment (response cost) = An animal will learn to stop producing a behavior if the consequence of producing the behavior is that something desirable is taken away.
Operant Conditioning:
Blocking Paradigm
- A mouse have an early training of Red ligh -> Food
- Late training of red light + bell-> food and blue light + alarm -> juice
- In test, blue light + bell-> juice
- Due to early learning, bell is not registered with food
What does blocking paradigm show
The Blocking effect shows that learning involves more than just monitoring co-occurrences
–If co-occurrence were the sole factor, then we would expect 50:50 responses between the food and juice in the test phase, but animals prefer the juice
Non-associative learning:
Refers to processes including habituation, priming and perceptual learning
Non-associative learning: Habituation
-Done through a series of exposure
– learning to ignore a stimulus because it is trivial (e.g. screening out background noises).
-because it has been learned
Non-associative learning: Priming
- Prior exposure to a stimulus can improve later recognition
- demonstrated by a change in the ability to identify a stimulus as the result of prior exposure to that stimulus, or a related stimulus
Non-associative learning: Perceptual Learning: Unitization
- repeated exposure of visual stimuli cause them to be perceived in chuck, not individually
- occurs when repeated exposure enhances the ability to discriminate between two (or more) otherwise confusable stimuli.
experiment for Non-associative learning: Perceptual Learning: Unitization
-Task of visual search where single feature cannot aid search
- There are two conditions:
–Target is always the same: consistent mapping
–Target differs on every trial: varied mapping
When the same target is always presented, people can learn to unitize features of the target and find it very quickly
Learning contingencies
Outcome Absent
Cue present a b
Cue absent c d
Delta P+ P(0?C0- P(O/-C)
=a/a+b - c(c-d)
Delta-P
Delta-P is a measure of the strength of contingency between a cue and an outcome
If people learn optimally, then their responses should reflect the magnitude of Delta-P
Are people sensitive to
People rate contingent associations as having a higher rating but overestimate noncontingent associations
Bliket detection task one cause and two cause
One-cause condition
Red block activates the detector
Cyan brick does not activate the detector
when children is asked , red is blicket
Two-Cause Condition
-when two item activate the bliket both are consider bliket
Bliket detection task backward and indirect
Backward: both objects activate, A activate by itself, when asked only A is considered
Indirect Screening: Both object active but one (B) does not activete itself. When asked, A P=1, B p=2/3
Probabilistic contrast model
® Tells us that when the background is variable, we need to conditionalize on the background to understand the strength of the contingencies in the environment.
® This tells us that people are sensitive to the context in which information is presented
Problem with delta P
- When the background is variable, delta-P misestimates the strength of the correlation
- The probabilistic contrast model states that we need to conditionalize on the background to understand the strength of the contingencies in the environment.
Assumptions of the Standard Model
- The environment affords the existence of directional connections between pairs of elements (cause effect)
- The elements are the mental representations of events or features of stimuli that are activated by the presence (or suggestion) of stimuli
- The presence of an element modifies the state of activation of another element
- Learning involves the strengthening of connections between elements
The Rescorla-Wagner model
Red light (X) and bell (Y) acts as predictor for reward (cheese)
V is existing connection
EV: Sum of associative strengths across all stimuli
lamba is the strength of Unconditioned Stimulus
Difference between maximum strength and current strength (10-1.5) = 8.5
DeltaVx= gamma (lamda-EV)
gamma is learning rate. higer gamma or differences lead to bigger changes
Further evidence for the role of attention in learning
• Blocking –Can also be explained by RW model • Highlighting • Unidimensional category boundaries are easier than diagonal boundaries –Cannot be explained by RW model
Evidence for Attention II:
Highlighting
Early TrainingRed Light + Bell Food
Late TrainingRed Light + Bell Food
Red Light + Alarm Juice
TestBell + Alarm Juics
Explaination for hightlight exp
A and B are already paired with X
Attention is shifted to D because it alone predicts the unusual event Y
Cue D drives the final response
Evidence for Attention in Learning:
Unidimensional Rules
-Subjects are shown 1 of 8 different stimuli to categorize on each trial
-Stimuli vary on their Height and the position of the inset vertical Line
two condition
-Filtration, only one condition is pay attention
-Condensation two condition must be attention
-people learn filtration faster than condestion
Learning is a balance of belief and data (what is belief and what is data)
-Data is the evidence of associations in the world
• If your beliefs are very strong, you need more evidence before you’re willing to change your beliefs
Two ways in which prior knowledge
influences our hypotheses
• Holmesian Deduction - Once you eliminate the impossible, whatever remains, however improbable, must be the truth
–Only hypotheses which explain the data are plausible candidates for an explanation
• Judicial Exoneration - If one suspect confesses, then we let the other suspect go
–If one hypothesis clearly explains the data, then other candidates are considered less likely