Docstoc

Neural Networks

Document Sample
Neural Networks Powered By Docstoc
					Cognitive Information
Processing
      Historical context
Symposium on Information Theory at MIT
(September 10-12, 1959).
   Newell & Simon, Chomsky, Miller, Bruner, and
    many others (see Gardner, 1987, p.28)
“Behaviorism spoke to many needs in the
scientific community, including some that
were quite legitimate . . . Yet, in retrospect,
the price paid by strict adherence to
behaviorism was far too dear (Gardner, 1987,
p. 12).
    Historical context
“I [Bruner] think it should be clear to you
by now that we were not out to “reform”
behaviorism, but to replace it. As my
colleague George Miller put it some
years later, ‘We nailed our new credo to
the door, and waited to see what would
happen. All went very well, so well, in
fact that we may have been victims of
our success” (Bruner, 1990, p. 4).
        Key features
Representations
Computers
De-emphasis on affect, context, culture,
and history
Belief in interdisciplinary studies
Rooted in classical philosophical
problems (Gardner, 1987)
Information processing
  model of cognition
Stages of Information
     Processing
Sensory memory
Working memory
Long-term memory
       Sensory memory
Auditory longer than visual
Selective attention
   Process and select info while ignoring
    other
     Sensory memory
Selective attention
 Process and select info while ignoring
  other
 Meaning it holds for learner

 Similarity with other tasks

 Task difficulty

 Ability to control attention – differences by
  age, hyperactivity, intelligence, and LD
         Sensory memory
Automaticity
Pattern recognition
   Template matching – mental copies (what’s the
    problem with this?)
   Prototype model
   Feature model
   Gestalt principles – what we do when things fail to
    resemble their prototype
   Prior experience
        Stroop effect
Missing slides
Working memory
    Working memory
7±2
Chunking
Short term (15-30 seconds without
rehearsal)
Rehearsal
Cognitive load
Automaticity
    Working memory
Encoding – how do you do it?
Demonstrations
          Encoding
Getting the information from WM to LTM
Provide organized information
Arrange extensive and variable practice
Use strategies for encoding
Enhance self-control of information
processing (Metacognition – more on
this later)
           Retrieval
Non-cued (recall)
Cued (recognition)
 Strength of memory trace
 Context

Encoding specificity – influence of the
context of encoding
           Forgetting
Failure to encode
Failure to retrieve
Interference
 Retroactive (later learning interferes with
  previously learned material)
 Proactive (previous learning interferes with
  later learning and is related to the amount
  of practice on the original task)
Long-term memory
       Concept maps
What’s a concept map?
A way to organize concept words and
propositions.
How are they used?
Inspiration is a tool for creating concept
maps – but you can do them by hand!
Create a concept map
Neural network
Semantic/propositional network/concept
map
Schema
Scripts
Dual-encoding models
Long-term memory
             Conditional




Procedural                       Declarative



                  Episodic             ?     Semantic
               (personal experience)       (general knowledge)
      Dual code models
Visual/verbal – 2 systems of memory
representation
Paivio
NOTE: Working memory
   Baddley model - Phonological and visual
    loop
    Building blocks of
        cognition
Concepts
Propositions




                   Note: Hand out cards
                  Concepts
It’s a “thing” that     Examples
“we” have classified.      Apple
Classified by              Red
   Rule
   Prototype
   Probability
        Propositions
Smallest unit of meaning that can stand
as a separate assertion.
Judge as true or false
Relationship between two concepts.
 Apples are red.
 Apples are green.

 Green apples are sour.
       What do the blocks
             build?
  Declarative              Procedural
      Semantic Networks      Productions
      Neural Networks        Scripts
      Schema




What are they and what do they “look” like????
     Network models
Semantic/Propositional networks
Nodes have meaning
Spreading activation
May show hierarchical relationships
Concept maps
Learning – building the network
                                                                            Breathe

Network                                                        can



models
                                             ANIM AL           has           Skin
                                                               can
                                                                            M ove
                                         is a          is a
            Fly
                         can
           Wings         has          BIRD                    FISH
                         has

          Feathers
                               is a             is a




                     CANARY                            OSTRICH

                   can    is                      cannot        is

            Sing               Yellow           Fly                  Tall
                            Network models
                      1.6
                      1.5                                          A canary
Response time (sec)




                                                 A canary          has skin
                      1.4
                               A canary           can fly
                      1.3                                        A canary is
                               can sing
                      1.2                        A canary is     an animal
                                                    a bird
                      1.1
                               A canary is
                       1
                                a canary
                      0.9
                                  "Complexity" of sentence
                                 (gaining complexity to right)
     Ausubel’s model
Rote versus Meaningful
Reception versus Discovery
Meaningful Reception Learning
Hierarchical, integrated body of
knowledge.
Anchoring idea
Learning – gaining the cognitive
structure
          Ausubel’s model
Processes of meaningful learning
   Subsumption
        Derivative (illustrate the concept – e.g., examples of
         different types of dogs)
        Correlative (elaboration, extension, or modification of
         previously learned concept).
   Superordinate – new inclusive proposition or
    concept under which established ideas are
    subsumed.
   Combinatorial – new idea not related in a specific
    sense, but is generally relevant to the broad
    background information.
       Neural networks
Connectionism (recall GCR)
Parallel distributed processing
Biologically inspired
Subsymbolic
   The nodes don’t mean anything – connections are
    most important
   Activation pattern carries the “meaning”
Learning via strengthening/weakening
weights
Processes: Spreading activation
Neural
networks            Output layer




 Hidden layers




      Input layer
Neural networks
                                                               Backward
                                                               Error
                                                               Propagation
                               Output Units
learning


                               Hidden Units



                  Forward      Input Units
                  network
                  activation



                                  From Luger & Stubblefield (1993, p. 524)
          Productions
If-then rules
 If the apple is red, then it is good for eating.
 If the apple has a worm, then don’t eat it!

“Fire” automatically
Is ordinarily implicit memory (typically
not conscious thought)
Production systems – developed via
declarative, then procedural knowledge.
  Production systems
IF the engine is getting gas, and
the engine will turn over,
THEN the problem is spark plugs.
IF the engine does not turn over, and the
lights do not come on
THEN the problem is battery or cables.
IF the engine does not turn over, and
the light do come on
THEN the problem is the starter motor.
                           From Luger & Stubblefield (1993)
              ACT - R
Comprehensive network model of memory
Propositions (subject + predicate)
Declarative knowledge (initially – schemalike
structures)
Procedural knowledge (later - productions)
Working memory – where declarative K. is
processed.
Learning – gaining these propositions
Strengthening (frequently used, stored close)
                    Schema
Abstract descriptions of things/events
Data structure
Top-down processing – a means to use
schema
Bottom-up processing – building/tweaking
Learning – development of a schema
   Accretion
   Tuning
   Restructuring
               Schema
Initial research – reading schema


      Who:     Mom, daycare teacher . . .
     Where:    couch, on floor
     When:     bedtime, circle time
    Actions:   hold book, turn pages . . .
     Props:    books
             Scripts
Experientially oriented (episodic
memories initially)
Utilizes schema
Based in Artificial intelligence
Use scripts to understand situations
through social contact
Used with typical/logical/routine
behavior
               Scripts
Situational understanding


     Where:    doctor office
     Actors:   patients, doctors, nurses . . .
     How to:   sit on exam table . . .
      Props:   medical equipment . . .
      Scene:   waiting room . . .
          Mental models
Mental, built on the fly
Humans represent the world they are interacting with
through mental models.
Are schemata +
   Represent knowledge and
   Include perceptions about task demands and task
    performances.
Used to direct behavior
Tend to be incomplete
Have little control over them
Unstable, change over time
            Mental models
“In order to understand a real-world phenomenon, a
person has to hold what Johnson-Laird (1983) describes
as a working model of the phenomenon in his or her mind.
Mental models are not imitations of real-world
phenomena, they are simpler.”
“A mental model which explains all aspects of the
phenomenon that a person interacts with is an appropriate
one. In order to provide explanation, it has to have a
similar structure to the phenomenon it represents; it is this
similarity in structure which enables the holder of the
model to make mental inferences about the phenomenon
which hold true in the real world.”
“A structural analogy of the world”
<http://www.cs.ucl.ac.uk/staff/a.sasse/thesis/chapter03.html>
      Mental models
How do you know what a learner’s
mental model is?
 Observe them
 Ask them for an explanation

 Ask them to make predictions

 Ask them to teach another student
             Instruction
Activate prior knowledge
Advance organizers (Ausubel)
Comparative organizers and elaboration
Conceptual/Mental models (often teacher
created)
   Learnable
   Functional
   Usable
Other strategies (for learner and instructor)?
    Assessment
?
    Motivation
?
    Worldview
?
                  Expertise
Consider your concept maps
Novice and Experts
   Experts excel mainly in their domain
   Have superior short-term memory for material in their
    domain – why is this, given that their memory capacity does
    not change? Perceive large meaningful patterns in their
    domains
   Are fast – and generally solve problems with less errors than
    novices.
   Spend more time analyzing the problem qualitatively
   See and represent problems in their domain in a deeper
    way.
   Have strong self-monitoring skills.
        Experts/Novices
Have been lots of expert-novice comparisons.
   Teachers – have different understandings of
    viewed classroom situations. Would focus on
    different thing (expert – both visual and verbal,
    whereas novice mainly visual). Experts more
    likely explain than just describe. Planned much
    more for long term action, and much of the
    planning was done in the context of teaching.
    Novices had less extensive teaching schemata.
    Planning took much more time for novices. Expert
    teachers could improvise.
Metacognition
          Definition
May mean different things to different
people.
“One’s awareness of thinking and the
self-regulatory behavior” that
accompanies this awareness” (Driscoll,
p. 107).
In a nut shell, knowing what you know,
knowing when you know, and knowing
what you need to know.
    Elements/Processes
“The executive” in many theories of memory.
Predicting, checking, monitoring, reality
testing, coordination and control of deliberate
attempts to study, learn, or solve problems
   Use of study time
   Estimating readiness for test
MC enhancing processes can be taught
(when appropriate depending on age of
learner)
 Metacognitive ability
     depends on
Person variables
   What is the role of age?
Task variables
   Different instructional content
Strategy variables
   Ways to encode, store, and retrieve
    information
            Instruction for
            metacognition
Limitations of standard instructional
practices
 Emphasis on direct instruction
 Lack of on-line diagnosis
 Basic skills before understanding
        E.g., Overemphasis on decoding results in lack
         of comprehension skills
   Emphasis on subskills
        Skills practiced in isolation
        Instruction for
        metacognition
Absence of explicit strategy instruction
Differential treatment effect
   Cumulated and more pronounced for
    weaker students
Self-regulation
    Reciprocal teaching
Form of guided cooperative learning
Passing of responsibility to students through
a well-defined process
Process elements
   Predicting – hypothesize what the author will say
    next in the text.
   Question generating – identify the kind of things
    that make for a good question
   Summarizing – integrate information
   Clarifying – Clarify unclear vocabulary, referent
    words, and complicated concepts.
    Reciprocal teaching
Teacher’s role
   Explain the strategies the students will be using, why they
    are learning them, and where they will be able to use
    them (helpful situations), and how they will learn the
    strategies.
   Then give instruction on the four strategies (only one day
    on these two points)
   Model the strategies
   Use guided practice in which more responsibility is given
    for the students to be the teacher.
   Offer praise
   Use teacher judgment to determine when more modeling
    and instruction is needed.

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:15
posted:9/1/2012
language:English
pages:57