Chapter 7 - Knowledge Flashcards
concept
unit of knowledge
mental representation that picks out a set of entities
Category
group of related things
-concepts link knowledge together
Categorization
process by which things are placed into groups called categories
understanding, prediction, learning, communication
Definitional approach (categorization)
determine category membership based on whether the object meets the definition of the category
-can work for certain abstract/constructed categories
Defining features (categorization)
features that any object MUST have to be a category member
-necessary and sufficient conditions
Prototype approach (categorization)
- family resemblance
- things in a category resemble one another in a number of ways
Prototype (categorization)
an average of category members encountered in the past
- an abstract representation of the typical member of a category
- not an actual member of the category
Characteristic features (categorization)
- features that objects in the category typically have
- the most salient features of category
- true of most instances in category
Exemplar approach (categorization)
- actual examples of the category (members)
- concept is represented by multiple examples
- to categorize, compare the new item to stored examples
typicality effect
more typical members of a category are processed faster and easier
-apple is a fruit vs pomegrranate is a fruit
Cognitive economy (semantic network)
shared properties are only stored at higher-level nodes
Inheritance (semantic network)
lower level items shared properties of higher level items
-robin has feathers
connectionism
parallel distributed processing (PDP)
-computer models for representing cognitive processes
loosely based on neurons and neural networks
Units
connections
- neurons in layers
- synapses between units
Output layer
hidden layers
input layer
- receive input from hidden layers, provide resulting signal from network
- receive input from input and hidden layers, send output to hidden and output layers
- activated by stimulation from environment, send signals to hidden layers
connection weight
strength of connection between two units
- determines how strongly an incoming signal will activate the next unit
- changes in connection weights happen slowly and represent learning and memory
activation
represents how active the neuron currently is
-changes in activation happen quickly and represent moment to moment info processing
error signal
difference between actual activity of each output unit and correct activity
backpropagation
algorithm to update connection weights based on error signal
-updating starts at the output units and works backwards toward input units
learning
properties are correctly assoicated with concepts
similarity
similar concepts (with similar properties) share similar representations (patterns of activation)
generalization
knowledge is shared among similar concepts
graceful degradation
performance decreases gradually as more and more of the network is damaged