Psy 416: Reasoning and Problem Solving
Concept Learning and Induction

1. Induction: Uncertain, Generalizations from experience, Predictions of future experience. Induction is generally about trying to make sense out of data. Often thought of as going from specific to general, or specific to specific.
2. Deduction: Certain, Conclusions from premises, Logical consequences of arguments, mathematical derivations. Often thought of as going from general statements to the specific events or statements or from general statements to general statements.

1. A concept is usually, but not always, identified by a noun or noun phrase, which is often thought to be definable by a set of properties.
2. Properties (red, large, circle, alive, animal) are often operationalized as values of dimensions, such as color, size, number, etc.
3. Some concepts may be simple, such as, 'red things'; some may consist of a conjunction of properties such as 'two red circles.' Sometimes a concept may be defined "disjunctively" such as a strike in baseball. Sometimes it's hard to identify the dimension to which a property belongs or to identify the set of necessary and sufficient features that identify a concept.
4. Sometimes the best we can do is to identify whether an example is an exemplar of a concept probabalistically.
5. Sometimes we may not be able to identify the dimensions or the features that define a concept with accuracy.

1. Aristotelean (or Classical): Concepts are defined by necessary and sufficient set of properties
2. Prototype: Concepts are defined by similarity to an ideal representative
3. Exemplar: Concepts are defined by its set of members

1. Much of the best data supporting connectionist models of learning has been in the area of concept learning.
2. It is thought that concepts are learned primarily by being presented with an exemplar and learning whether it is a member of a particular concept or not. As the learner gets more and feedback from examples, she gets better and better at knowing which examples are category members or concept exemplars and which are not. These are seen to be values of some dimension.: dogs, red things, grammatical sentences, planets, a strike in baseball—Simple concepts, conjunctive concepts, disjunctive concepts.
3. Hull’s Chinese character learning. see example
4. Kendler and transfer of training: reversal and nonreversal shifts. see example
5. Frequency and probability of association are the primary principles of category learning. For those properties that are reinforced the association is strengthened.

1. Minimal effects of early trials on categorization: random to very good performance, non-efficacy of changing rule during learning phase.
2. Bruner, Goodnow and Austin and strategies of concept attainment. Different people use different strategies to try to identify the correct concepts. These strategies are specific task related. Reception strategies (e.g., "Wholist"--select intersection of hits; and "partist"--test hypotheses ) and Selection strategies (e.g., "scanning"--test hypotheses; and "focusing"--use positive instance as anchor). see example
3. Strategy: Confirmation bias : There is a tendency for people to focus on confirmation of hypotheses and to pay attention to confirmatory rather than disconfirmatory evidence, although they can learn make use of information of all types.

Kenneth Spence:  Eye blink conditioning with aware and distracted subjects

Marcia Johnson: Tower of Hanoi with normal and Korsakoff patients

Continuity and discontinuity?

Implicit and explicit processes?

Automatic and deliberative (conscious) processes?

1. Availability: Think of examples
2. Representativeness: How do properties fit those of category prototype?
3. Role of base rates?