Memory and cog exam 2 Flashcards
Where are pyramidal cells?
cortex
where are stellate cells?
thalamus
where are purkinje cells?
cerebellum
What are the types of glial cells and what are their functions?
Ependymal - line ventricular walls and central canal of spinal canal, secrete CSF
Astrocyte - in CNS, create blood brain barrier
microglial - in CNS, immune surveillance and defense
Oligodendroglial - form myelin in CNS
Schwann - form myelin in PNS
types of integration of PSP:
Integration across space (spatial): PSP coming in via multiple synapses simultaneously
Integration across time (temporal): multiple PSP coming via one synapse in succession
Cancellation: EPSP and IPSP graded potentials cancel each other out
How are firing rates measured?
Measured in spikes per second
What was the major event in the history of cognitive science that popularized connectionism? What was it a precursor to?
the publication of the ‘parallel distributed processing’ volumes in 1986 by Rumelhart and McClelland. direct precursor to deep neural nets and deep learning.
Components of neural network models:
Implemented on computer, simulated neurons (units), simulated connections (links), connections have weights. Abstraction of real neurons.
What are feedforward networks?
inputs -> hidden layer –> output
What are the elements of a simulated neuron?
input activations x, output activation y, activations correspond to firing rates, w connection weights correspond to synaptic efficacy & can be inhibitory or excitatory, netinput = weighted sum of inputs to unit
In a distributed memory system, a mental state is:
a pattern of activation over the neurons
In a distributed memory system mental processing is:
transforming patterns of activations through input to output pattern transformation and weights
where is knowledge in a memory model?
weights. The weight matrix determined the mapping from inputs to outputs
What are memory traces in a distributed memory system?
changes in the weights
what is retrieval in a distributed memory system?
reinstatement of a prior pattern of activation
What is 1 bit
2^1 possibilities
what is 4 bit
2^4 possibilities
What are localist neural representations?
Each stimulus dimension (size, shape, etc) is represented by a seperate pool of neurons. Each particular value is represented by a dedicated ‘detector’ neuron. Can assign verbal labels.
What are distributed neural representations?
Each item (ie a color) is represented by a pattern of activity over multiple neurons. each neuron participates in the representation of multiple items. Only the pattern of activation differentiates the item, not the neurons. You can’t label individual neurons.
What is I/O mapping
long list of correspondences. each correspondence maps one pattern to another pattern. function y = f(x) in math. Essential building blocks of any behaving system.
What is the pattern associator network?
map each input pattern to a specific output pattern. Part of larger system that provides inputs and interprets outputs. Each output neuron calculates its activation y independently of all other output neurons. All outputs receive the same inputs x but with different weights w.
How is information carried out?
By transforming patterns of activation across populations of neurons. the synaptic weights of these connections between neurons govern how these transformations are carried out.
What is synaptic plasticity?
the strength and/or number of synaptic connections changes as a result of experience. can be strengthening or weakening. both excitatory and inhibitory connections.
AMPA receptor
opens when glutamate binds
permeable to sodium
transmission of activation
NMDA receptor
opens when glutamate binds AND depolarized membrane
permeable to sodium and calcium
co-activation detector
triggers learning
What is the aim of an autoassociator? How is it achieved?
to reproduce the same pattern at output that was presented at input. using the delta rule calculate the difference between the internal and external inputs and changing the weights of connections in the direction that will reduce the difference.
- The connections
are recurrent –
they loop back
onto the same units. - Repeat until convergence – that is,
until the pattern stops changing.
Autoassociator can ‘clean up’ degraded patterns
What is it called when more than one association is stored in the same weight matrix?
superposition
Distributed memories:
Distributed memories: Each association
is stored in the relative strengths of
multiple connections. Each connection
participates in the reconstruction of
multiple associations.
What is fault tolerance and graceful degradation?
Minor damage tends to cause a small
change in response to many inputs,
rather than a total loss of some
memories and no effect on others.
* Distributed memories are robust to
damage. This is called graceful
degradation
Benefits of distributedness:
- Generalization to similar patterns
- Similarity = feature overlap
- Robustness to damage
- Good (slightly degraded) reconstruction
from imperfect cues (noisy inputs) - Graceful degradation when units or
weights get damaged - Prototype extraction
- See McLeod et al. (1998, pg. 63).
- Prototypes will be covered in Lecture 16
Recurrent networks:
Recurrent networks have loops. It is possible to
start at some node, travel around the links, and
return back to the same node.
* The recurrent networks are dynamical systems
whose behavior depends not only on the stimulus
but also on their internal state.
feedforward networks:
can be drawn so that all
connections point in one direction. There are no
loops. The information flows from inputs to hidden
units to outputs.
Basic concepts of memory:
- Mental state as a pattern of activation
- Mental processing as transforming
patterns of activation - Memory traces as changes in the weights
- Retrieval as reinstatement of
a prior pattern of activation
Retrieval by reconstruction:
- Remembering is an active
process of reconstruction. - Memories are reconstructed
on the basis of partial cues
and the latent knowledge in
the connections between
neurons.
Retrieval is a process of reconstruction: the input activation
propagates to the output units, multiplied by the weights.
Brains relationship to the distinction between read (or retrieve) and write (or store) modes of operation:
- The brain typically operates in both modes simultaneously.
- That’s possible because the synaptic changes are small and
incremental. (This is modeled by setting the learning-rate
parameter to a small value: ) - The brain uses (“Reads”) the existing weights to produce
behavior and also updates (“Writes”) them a little according
to the current activation patterns.
Δw ji =
η ⋅ x i ⋅ y j
Removing the temporary scaffold from Slide #10 above:
Learning:
the process by which
relatively long-lasting changes occur in
behavioral potential as a result of experience.
Memory:
the product of learning
Sharpening errors
tendency to elaborate on details that were not made clear in a story (ie making up explanations and falsely recalling these as being in the story)
Flattening errors
Inability to remember details of the story, especially if they did not fit with the schema that the person had imposed on the story
One type of schema is a script. What is a script?
organized knowledge structures about typical actions and the order of such actions. frontal lobe damage patients remembered as many of the actions in each script as normal participants but were severely impaired in recalling the correct sequence of steps. Frontal lobes responsible for the planning or sequential nature of scripts.
What does Barlett consider remembering to be ?
remembering is reconstruction rather than retrieved
What are intrusions?
incorrectly “remembering” items consistent with schema
What was the experiment that showed relationship btwn expertise and?
chess experts shown 25-30 chess pieces from middle of a master game. Almost 100% remember compared w 40% for normal ppl. When the chess pieces are random, recall same for both groups. Memory based off of their skill set and recognizing good moves.
Self-reference effect
information bearing on the self is processed more deeply and remembered better
Subjects more likely to
remember the word “friendly”
if they had been asked
whether they were friendly
than whether they knew what
friendly meant or whether it
was a long word
Self-serving bias
interpreting events such that
the person claims credit for success but denies
blame for failure
Junk mail theory of self-deception
People control what events they attend
to and try to not attend to ‘bad news’:
* People remember good things better
than bad things (because of rehearsal).
* People try not to dwell on failures.
* People recognize “junk mail”
(e.g. bad news) and just throw it in
the trash (do not pay attention to it)
Positive illusions
People overestimate their
good qualities and
underestimate their faults
2. People overestimate their
perceived control over events
3. People are overly optimistic
in their estimates of good
things happening (or of
avoiding bad things
What is the negative cognitive triad?
- Negative view of oneself:
Perceiving oneself as worthless, deficient, inadequate,
unlovable, and lacking the skills necessary to achieve happiness. - Negative view of the world at large:
Perceiving the environment as imposing excessive demands
and/or presenting obstacles that are impossible to overcome,
leading to continual failure and loss. - Negative view of the future:
Perceiving the future as hopeless and believing that one is
powerless to change things for the better. One expects of the future
only continuing failure and unrelenting misery and hardship
Cognitive specificity hypothesis
- Most social situations are ambiguous and leave room for interpretation.
- Different mental-health disorders are characterized by different types
of automatic thoughts and different schemas for interpretation.
Cognitive distortions associated with depression:
- All-or-nothing thinking: Seeing events as either all good or all bad
- Overgeneralization: Believe that if a negative event occurs, it is
likely to occur again in similar situations in the future - Mental filter: focusing only on negative details of events, thereby
rejecting the positive features of one’s experiences - Disqualifying the positive: neutralizing or denying one’s
accomplishments - Jumping to conclusions: forming a negative interpretation of
events, despite a lack of evidence - Magnification and minimization: making mountains out of molehills
- Emotional reasoning: interpreting events on the basis of emotion
instead of reason - Should statements: creating unrealistic expectations
- Labeling and mislabeling: attaching negative labels to oneself and
others - Personalization: Assuming that one is responsible for other people’s
problems
What did they find with the washing clothes passage?
Presenting the schema BEFORE reading the passage aided recall, but not if it was presented AFTER reading the passage
What are the three positive illusions?
people overestimate their good qualities and underestimate their faults, overestimate their control over events, and are unrealistically optimistic
What was the period study?
ppl more likely to recall their period as aligning with their previous view of how mild or severe it was regardless of how they actually rated it while it was happening.
What is self-protection?
trying to avoid loss of self-esteem by avoiding failure
What is the difference between people with high vs low self-esteem?
they don’t want to fail, their ideas about themselves are conflicted and uncertain (self-concept confusion), they do not view themselves with strong negative terms but rather have an absence of strong positive views about themselves.
Self-deception strategies:
self-serving bias (claim credit for success but denies blame for failure), being more skeptical of negative feedback than positive feedback, selective attention and memory, downward comparisons, shifting criteria to fit your behavior (ie i’m a good girlfriend because I’m good at giving gifts)
2 benefits of high self-esteem
positive feelings & initiative, or openness to new experiences
What is the sociometer theory?
self-esteem is linked to social acceptance. Self-esteem is a sociometer because it measures the traits you have according to how much they qualify you for social acceptance
Mind-independent vs mind-dependent conventions
Some type of boundary physically imposed by nature (strait between an island and a country)
Boundaries as lines drawn by humans with no real physical reason (state lines)
Concepts vs categories
Concept: mental representation
Category: set of objects or events or entities in the world. a concept denotes a category
Reasons for having categories
they help us to make inferences about novel instances
conceptual hierarchy
Superordinate level: most general
Basic level: intermediate
Subordinate level: most specific
Three theoretical approaches to conceptual structure:
classical (rule based)
probabilistic (similarity based - prototype based concepts, instance based)
Knowledge based (theory-theory)
Classical view of concepts:
concepts are defined in terms of rules involving necessary (each member of this category MUST have this feature) and jointly sufficient (whatever object has this feature(s) must be a member of this category) features. the view that concepts have precise definitions based on defining features has come to
be called the classical view
Conjunctive vs disjunctive concepts
conjunctive: defined by a rule that only uses the logical connectives AND (ie the number three and color black)
easier to learn
disjunctive concept: defined by a rule that uses OR or EITHER OR (could also use AND and not)
harder to learn
limitations of the classical view
defining features are hard to come by
typicality effects (the tendency to judge typical items as members of a category more easily than atypical items) point to graded category membership
unclear (borderline) cases
similarity based approach:
each concept has a center and a periphery, category membership is determined by similarity
family resemblance theory:
members of a category have a family resemblance to one another. Graded structure: certain category members are rated more typical or representative of the category than others
Multi-dimensional (vector) spaces
each dimension is a feature which can take multiple values. each object is described by a large number of dimensions. each point corresponds to a point in a vector space. the distance between points measures the dissimilarity of the corresponding objects. The prototype is the center of mass (the average of all coordinate points (x, y, and z) in the category)
limitations of the prototype view
it discards information concerning category size, variability, and correlations. they fail to reflect the context sensitivity evident in human categorization.
Instance based representation
don’t combine training examples into an average prototype, instead store in memory the examples themselves. perform averaging during memory retrieval, not storage. this preserves information about the distribution: its mean, variance, correlations
Blurs the distinction between episodic memory (of instances) and semantic memory (of concepts)
Memory for specific instances
photos of skin lesions. Memory for the specific training exemplars was maintained even after a more abstract representation of the category had been developed
Semantic network
early model of semantic memory. each concept represented by a node. labeled links denote relationships between concepts.
Processing:
cognitive economy - shared properties are attached to nodes high in the hierarchy
Evidence for hierarchical semantic organization
sentence verification task: true or false?
category verification tests: “a canary is a bird”
property tests: semantic distance = 1 “A canary can fly”
RT quicker for the closer semantic distances
Problems with semantic nets:
verification time depends on many factors: typicality frequency, etc
a hierarchical effect often goes away when these factors are taken into account
reverse distance effects for false statements
semantic priming effects
the time to process a word depends on its similarity to a previously processed word
Spreading-activation model
associative network (not necessarily hierarchical)
links vary in strength, stronger links faster RT, accounts for typicality
accounts for semantic priming effects
Adaptive control of thought - rational
modular organization
declarative LTM organized as a semantic network with spreading activation
Declarative LTM in ACCT-R fan effects
the more links fan out from the node, the slower and weaker the activation
Concepts as points in a vector space
a large matrix of co-occurence statistics containing information about word meaning
Pre-training a large language model (LLM)
a large population of brain circuits generated gigabytes of text
eavesdrop on their conversation and use an incremental learning rul to train a deep neural net that can generate similar texts
Recovering the spatial layout of a cloud of points
can infer vector-space representation from word co-occurence stats
Hyperspace analogue to language (HAL)
input 300 mill words from internet discussion groups. slide 10-word window across length of text and detect co-occurences
Latent semantic analysis
machine learning technique. large corpora of text. vector-space representations of the similarity relationships among concepts. captures gists of passages. can pass TOEFL vocab test
Coherence
items within a category that tend to be similar to each other
differentiation
items from one category tend to be different from other categories
which level is most useful for making categorical inferences
the basic level
Concepts vs categories:
A concept is a
mental representation, made up of one’s knowledge about a type of object (e.g., DOGS )
or an idea (e.g., SHARING ). Categories, on the other hand, are the actual divisions that
we use in dividing up the world. The category DOG , for example, consists of those
objects in the world that we consider to be representatives of the concept DOG. The
concept is inside us, while the category is often out there.
What is the exemplar view?
assumes that there is no
single mental representation of a concept (prototype as in the family resemblence; rather, all one has in memory are the specific
items, or exemplars (examples) of the concept that one has encountered in one’s deal-
ings with the world
What did the skin lesion study show?
participants were more correct for old exemplars than for new ones (even
new ones that were perceptually similar to the training slides). Similar results were
found when participants were tested after a 2-week delay
after learning a category, people are most
efficient at categorizing new objects that are similar to previously seen (i.e., training)
exemplars
Do people categorize based off exemplars even when given rules?
yes
amnesics and exemplars
learning concepts depends on implicit memory, while recognition for the specific exem-
plars depends on explicit memory for episodes. Since amnesia involves a loss of explicit
memory, typically with retained implicit memory (see discussion in Chapter 2), Knowl-
ton and Squire (1993) concluded that concept learning could take place independently
of storage of the specific items. This conclusion raises problems for the exemplar view,
which predicts that categorization should be related to recognition of individual items
similarity of family-resemblence and exemplar theories
they both
assume that objects in the category are related to each other in terms of similarity. in the family-resemblance view, category judgments are made on
the basis of similarity to the prototype. The exemplar view proposes that an object
is placed in some category because of its similarity to one or more exemplars of the
concept.
similarity-based view of concepts
assumes that the simi-
larity is in the world to start with, and that people often rely on surface similarity—that
is, perceptual characteristics of the objects—as the basis for categorization. This is a
bottom-up view of concepts and category formation, since it assumes that similarities
and differences among objects are obvious, because of the way in which the world is
structured and the way in which our perceptual systems are built.
Example of when similarity does not affect categorization:
three inch circle seen as being more similar to a quarter but categorized as a pizza. factors more important than similarity sometimes
what is folk theory
postulates that many objects have
an underlying essence that dictates many characteristic features of those objects, and
determines their category membership.
psychological essentialism
people believe that many categories have some essential components
that explain how and why deep and surface features coalesce
elements of psychological essentialism
First, the theory predicts
that people will appeal to the “essential nature” of a concept when making category
decisions, and the extent to which an object possesses this essential nature will deter-
mine its category membership. This means that essentialist categories should show
all-or-none membership
secondly,
if organisms or objects have an essential nature, then they should maintain certain
traits even if placed in a different environment.
Thirdly, psychological essentialism predicts that underlying essences should out-
weigh surface similarity when people make category judgments or engage in concept-
based inferences, and that category membership will be used to infer nonobservable
or internal traits.
can essentialist and family-resemblence theory (graded category structure) coexist?
research findings of graded category structure may not contradict the essentialist view,
and it is possible that people may represent both the essential and the typicality of
concepts simultaneously
coexist for odd number and water
