Generalization, Discrimination & Concept Formation PDF

Generalizatio n, Discriminatio n Learning, and Concept Formation Chapter 6  Behavioral processes of generalization What are Discrimination learning we Concept formation, category learning, and theoretical discussing explanations ?...

Generalizatio n, Discriminatio n Learning, and Concept Formation Chapter 6  Behavioral processes of generalization What are Discrimination learning we Concept formation, category learning, and theoretical discussing explanations ? Brain substrates Clinical applications What is generalization? Generalization: the transfer of past learning to new situations and problems Example: fear of flying bugs What determines what insects are feared vs. not feared? Balance between specificity and generality Specificity – how narrowly a rule applies Generality – how broadly a rule applies What is generalization? Eventually, the child may learn No fear response some flying insects are safe Discrimination learning: the process by which animals or people learn to response Strong fear response differently to different stimuli Behavioral processes of generalization Goal: determine when stimuli have similar and different outcomes The degree of similarity or dissimilarity impacts behavioral outcomes Same outcome Different outcomes Similar stimuli Wasps and bees  Wasps  run away! When do we know run away! two things are Bees  No similar? response Dissimilar stimuli Wasps and lady Wasps  run away! How do people bugs  run away! mentally represent Lady bugs  No the similarities of response two objects? First, let’s talk about when similar stimuli predict similar outcomes Generalization: When similar stimuli predict similar outcomes When do humans or animals generalize vs. discriminate? Exploring how pigeons peck at lights for reinforcement Guttman & Kalish (1956) Generalization: When similar stimuli predict similar outcomes Guttman & Kalish (1956) Pigeons pecked the yellow light the most They also frequently pecked similar lights but not dissimilar lights! What does this mean? Generalization gradient: Curve showing how changes in physical properties of stimuli correspond in response changes Generalization: When similar stimuli predict similar outcomes But are pigeons just making mistakes? – No The gradient reflects the best estimate that the novel stimulus (i.e., new color) with have the same consequence (i.e., food) Pigeons must identify the stimuli that have the same consequence as the target stimulus (i.e., yellow light) Consequential region Generalization: When similar stimuli predict similar outcomes Key take-aways from Guttman & Kalish (1956) A generalization gradient can show perceptions of similarity If two stimuli are perceived as highly similar, there will be significant generalization between them Animals (and humans) learn and predict which set of stimuli will have the same consequences as previously encountered stimuli Example: Insects that sting vs. insects that do not sting What processes make this possible? Theoretical explanations of generalization Remember the Rescorla- Wagner model? Elements of the world represented by nodes Network model to explain associations between stimuli and responses Values represent the strength of an association between a stimulus and response Theoretical explanations of generalization But, there are some issues using this model to explain generalization Stimuli are represented as discrete components: Each stimulus has one single node Based upon this model, how frequently would animals respond to similar stimuli? Theoretical explanations of generalization According to the Rescorla- Wagner model, animals should only respond to the trained stimulus This is not similar to the generalization gradient from Guttman & Kalish (1956)! Discrete-components model does not explain generalization Theoretical explanations of generalization Must be shared representation between similar nodes Distributed representation: Stimuli are represented by nodes Similar stimuli (i.e. yellow and orange lights) activate similar nodes Information learned about one stimulus will generalize to a stimulus that activates all or some of the same nodes  Discrete-Components Generalization Gradient Guttman & Kalish (1956) Generalization Gradient Distributed-Representation Generalization Gradient So far, we’ve discussed when similar stimuli predict similar outcomes What about when similar stimuli predict different outcomes? When similar stimuli predict different outcomes Discrimination learning: the process by which animals or people learn to response differently to different stimuli A behavior is triggered by the presence or absence of some stimulus Example: traffic lights Red light = stop, yellow light = floor it Stimuli are used as cues that impact behavior Stimulus control Real-world examples of stimulus control Trouble sleeping? You may have too many stimuli associated with staying awake in your bedroom! TVs, phones, or other devices Also guided by doing activities other than sleeping (or sex) in the bedroom Playing video games Eating Doing homework Discrimination learning: behaviors and mechanisms When do animals (and humans) generalize vs. discriminate? Jenkins and Harrison (1962) Trained 2 different groups of pigeons How will the groups respond to similar tones? Group #1 Group #2 Training Phase Testing Test with new tones, ranging from 300-Hz to 3,500-Hz Phase Discrimination learning: behaviors and mechanisms Jenkins and Harrison (1962) The birds trained to respond to 1000-Hz but not 950-Hz showed a steeper gradient curve What does this mean? Discrimination training elicits more narrowly-focused responses Theoretical explanation: Less activation of similar nodes More distinct mental representation We learn to generalize and discriminate between stimuli. These are key processes for forming concepts!  Concept: mental ideas based upon experiences with the world Categories: Concepts grouping of objects or ideas Dependen and that have common t upon concept formation categories underlying features Often used Categories of trees: fern, coniferous, etc. synonymously, Concept of a tree: mental representation but they are based upon knowledge and experience with different! trees Concept formation and categorization Cognitive processes used to organize, describe, and generalize about the world Example: felines Use generalization to categorize a lion amongst felines Concept formation and categorization Cognitive processes used to organize, describe, and generalize about the world Example: felines Use generalization to categorize a lion amongst felines Use discrimination to further categorize it amongst wild cats and not domesticated cats Theories of Concept Formation Rule-based approaches: object is categorized by testing it against one or more rules If X, then Y approach to categorization Example: categorizing organic chemistry compounds If the compound has an OH attached, then it is a phenol compound Theories of Concept Formation Similarity-based approaches: judge the similarity between a target object (i.e., lion) and some standard in memory Grammostola tarantulas Poecilotheria tarantulas Theories of concept formation: types of similarity- based approaches Prototype approach: abstraction based upon previous experiences Prototype: the idealized member of a category Formed based upon prior experience with members of the category When presented with a new object, compare it with prototype Theories of concept formation: types of similarity-based approaches Limitations to the prototype-approach Typicality effect Some items are seen as better members of a category compared to others Example: think of a fruit Did you think of a tomato? Theories of concept formation: types of similarity- based approaches Exemplar approach: comparison of new objects with other members (i.e., exemplars) of a category No abstraction No prototypes, many examples to compare against Category membership based upon similarity to exemplars  Category learning Humans (and animals) can learn categories through 2 main methods: Blocking: viewing multiple exemplars of a category back-to- back Grammostola This method tarantulas is best for generalizing across categories Poecilotheria tarantulas  Category learning Humans (and animals) can learn categories through 2 main methods: Blocking: viewing multiple exemplars of a category back-to- back Interleaving: viewing a mixture of exemplars from different categories Grammostola tarantulas This method is best for discriminati ng between categories Poecilotheria tarantulas Our ability to form concepts and categories relies on generalization and discrimination learning. Does this matter outside of images and sounds?  Discrimination learning vs. discrimination Discrimination learning: ability to discriminate between 2 stimuli (red light vs. yellow light) Stereotypes, Discrimination: behavior (usually harmful) towards individuals based Discriminati upon the group to which they are perceived to belong on, and Racism Discriminatory behavior can be fueled by stereotypes: Beliefs about the attitudes of the members of a group Relies on generalization Stereotypes, Discrimination, and Racism The trouble with stereotypes arises when: People use generalizations that are not based upon data People use statistically valid generalizations to justify discrimination Example: Gender stereotypes and math ability Results impacted by stereotype threat: decrease in performance due to membership to a group hampered by a negative stereotype Attitudes towards women can impact math success Evidence that these findings may vary depending on if the math test is high vs. low stakes Lots of individual variation! Brain Substrates Cortical representations and generalization Reminder: specialized areas for cortical processing of sensory information Primary visual cortex (V1) Primary auditory cortex (A1) Primary somatosensory cortex (S1) Each region is organized topographically: Each area on each cortex responds to a particular stimulus Cortical representations and generalization Topography of the A1 Neurons specialized to respond to specific frequencies Can visualize the receptive field: Area when stimulated will produce a response of a neuron Example: receptive field 2,000Hz sound Cortical representations and generalization Topography of the S1 and M1 Neurons specialized to process specific physical sensations Can visualize the receptive field: Area when stimulated will produce a response of a sensory neuron Example: receptive field of the mouth Shared-elements model of receptive fields Reminder – distributed- representation theory of generalization: Stimuli are represented by nodes Similar stimuli (i.e. yellow and orange lights) activate similar nodes Will this cause overlapping neurons in receptive fields to fire? Shared-elements model of receptive fields Model for a tone played at 550 Hz The 550 Hz will be activated Nodes close to 550 Hz should be activated Neurons that share similar receptive field should be activated if this theory is supported Shared-elements model of receptive fields Model for a tone played at 550 Hz Tested by stimulating the neuron responding to 550Hz Generalization gradient viewed What does the figure suggest? The role of the primary cortex in generalization Explored by Richard Thompson General procedure Train animal to respond to a specific tone View generalization curve Remove primary cortex (A1, S1, or V1) Play tones View generalization curve Findings: removing the cortex results in overgeneralization to tones Generalization and the hippocampal region The hippocampus and related structures play an important role in generalization! Responsible for learning about relationships between stimuli The hippocampal region: hippocampus and associated brain regions Usually use this terminology with animals Medial temporal lobe: inner surface of temporal lobe that contains hippocampus, amygdala, and structures important for memory Usually use this terminology with humans Generalization and the hippocampal region Damage to the hippocampal region impacts generalization Inability to generalize between stimuli that have occurred in the past Particularly involved in how associations are represented Operates as an “informational gateway” Selects what information is allowed to enter memory Determines how it is to be encoded by other brain regions We will talk about the hippocampus a bit more in future lectures!  Clinical Perspectives  Schizophrenia: psychological disorder characterized by Generalizat disturbances in thinking, emotional ion deficits responsiveness, and behavior DSM-5 criteria for diagnosis: in Delusions schizophre Hallucinations nia Disorganized speech Disorganized or catatonic behavior Negative symptoms  Hippocampal dysfunction observed in those with schizophrenia A core feature of the illness Generalizat Abnormalities in hippocampal shape ion deficits Overall smaller hippocampal volume in Able to learn associations between schizophre stimuli nia Struggle with transferring associations to a new context (i.e. generalizing) Assessed using an acquired- equivalence task Acquired- equivalence task Phase 1: learn both individuals are equivalent Phase 2: learn new information Phase 3: quiz on pairings learned so far Individuals with schizophrenia typically make more errors (choose the yellow fish) compared to neurotypicals  Hippocampal dysfunction observed in those with schizophrenia A core feature of the illness Abnormalities in hippocampal shape Generalizatio Overall smaller hippocampal volume n deficits in Able to learn associations between stimuli schizophreni Struggle with transferring associations a to a new context (i.e. generalizing) Assessed using an acquired- equivalence task Challenges with transitive inference (see p. 269) Altered generalization in autism spectrum disorder Autism spectrum disorder (ASD): neurological and developmental disorder Persistent deficits in social communication: Social-emotional reciprosity Nonverbal communication Developing, maintaining, and understanding relationships Presence of restricted, repetitive behaviors Repetitive motor movements Inflexible routines Highly fixated interests Atypical reactivity to sensory aspects of the environment Altered generalization in autism spectrum disorder During generalization Diverse anatomical tasks, children with and functional neural ASD hyperfocus on abnormalities one feature of a stimulus Dysfunction in Results in impaired prefrontal cortex, generalization basal ganglia, temporal lobes, and limbic regions Altered connectivity within and between brain regions Study guide What is generalization? Know how specificity and generality is important for generalization. What is discrimination learning? How is it different compared to generalization? Be able to describe the method, results, and overall findings of the Guttman & Kalish (1956) study. What is a generalization gradient? What is a consequential region? What factors determine the shape of the generalization gradient? Be able to describe the Rescorla-Wagner model, discrete components model, and how the Rescorla-Wagner model is unable to account for generalization. What is a distributed-representation model? How is it different than and similar to the discrete- components model? How does discrimination training impact a generalization gradient? Know the method and results of Jenkins and Harrison (1962) Study guide (cont.) What is the difference between a concept and a category? Be able to describe each of the theoretical perspectives of concept formation. Know the difference between an exemplar and a prototype. Know the difference between blocking and interleaving. What is the difference between discrimination and stereotypes? What is stereotype threat? When does the use of stereotypes become problematic? What is the receptive field? How does this relate to the distributed-representation model? Know how the medial temporal lobe is important for generalization, including the brain areas and the hippocampus’ role in association representation What generalization deficits are present in individuals with schizophrenia and ASD? What is an acquired-equivalence task? What is transitive inference? (p. 269 in your book)