W11: Learning & knowledge Flashcards
Learning involves attention
Not all of information is important for achieving a particular goal
Trade off between salience and validity
The nature of selective attention determines the mental representations we form about the world
=> weighted opinions and updating our hypothesis
Learning Involves Attention:
Animals: Bees
In the pursuit of some goal, organisms must overcome initial fascination (attention) to salient but irrelevant attributes
Organisms must figure out what is relevant for achieving a particular goal
Bees: Waggle Dance
Efficient communication on where the food source is
(where n far) => not complicated. however no other info like preditors.
Human’s evolvement of learning: Selective attention.
four experimental effects which show us that attention is important for learning (4)
1) Trade offs between salience and validity
2) Blocking
3) Highlighting
4) Learning rules of different complexity
Salience
Definition &
High/Low
how much does the cue grab your attention all other things being equal?
1) In the absence of validity, high salience cues will attract attention
2) Low salience cues will not attract attention
Salience and Validity
Modified Posner cueing task
Description
S13
Only 1 of many arrows correctly predict the location of coming stimulus.
Variations:
1) Probabilisitic cues (p = ~ 0.5 to 0.8 of being correct) => differ in predictive validity
2) Salience (high/low)
Task:
learn which of these cues to attend to respond to the light as quickly as possible
Measure: Utilization
Salience and Validity
Modified Posner cueing task
Results + conclusions
S14
Utilisation for high salience >= low salience for all varied validity.
As validity for low salience (LS) increases, utilisation of LS increases.
Conclusions:
1) increase validity => increase utilization (opp same)
2) increase salience => increase utilization (opp same)
3) Validity and salience interact
4) increased utilization of 1 cue decreases utilization of the other.
Attention determines learning and is affected by …
Attention is driven by both salience and validity
Attention determines what we learn
- Blocking
- Highlighting
- Unidimensional rules
Evidence for Attention:
Blocking
Using Mouse classical conditioning
A. D Similar stimuli (lights)
B, C (all sounds)
Blocking task: during early training: A -> results in Reward X During late training: A+B -> X C+D -> Y Testing: B+D results in preference of Y.
Explanation:
- Attention is shifted to A because it is important for predicting X
- No Attention is left for B when A.B are paired
- Cue D drives the final response
Blocking: what was previously learned for cue A blocks out cue B (new learning)
Evidence for Attention:
Highlighting
Using Mouse classical conditioning
highlighting task: early training: A+B -> X late training: A+B -> X A+D -> Y Testing: B+D results in preference of Y.
A+B is double trained compared to A+D. So Why D?
Attention is shifted to D because it alone predicts the unusual event Y
Cue D drives the final response (highlighting)
Simple Learning Theory VS
Attentional learning theory
Simple Learning Theory
Co-occurrences lead to a strengthening between cues and outcomes
Attentional learning theory differs by also incorporating the ability to differentially weight cues according to their relevance
Evidence for Attention in Learning: Unidimensional Rules
Details
Plus simple learning and attentional learning theories
Subjects are shown 1 of 8 different stimuli to categorize on each trial
Stimuli vary on their Height (up/down) and the position(left/right) of the inset vertical Line
Conditions:
Filtration: 1D either height or position
Condensation: 2D
The results is compared to the predictions of simple learning and attentional learning theories
Evidence for Attention in Learning: Unidimensional Rules
Plus simple learning and attentional learning theories
results
S42
Simple learning theory: does not predict any difference between filtration and condensation condition
Predictions of Attention learning theory:
learning is better for the filtration than condensation condition
conclusions:
- Learning involves not only learning what things are associated but also where to attend
- it takes time to build up connections between cues and their associates for both of the models of learning
People do not learn in the absence of any prior expectation.
Expectations provide hypotheses which we use to:
- Explain the thing we’re trying to learn
- Evaluate our experiences (allow us to update our expectations)
One-shot Generalization
3
single example of a new concept can be enough information to support
(i) generation of new examples,
(ii) parsing an object into parts and relations (parts segmented by color),
(iii) generation of new concepts from related concepts
One-Shot/Fast Mapping learning
Learning by exclusion, based on what is already known (our expectations)
