Chapter 7. Memory and Training

1
Chapter 7. Memory and Training

• OVERVIEW
• Working memory temporary, attention-demanding
  store to retain new info until we use it
• examine, evaluate, transform,
  compare, mental arithmetic, predict
• hold new info until encoding
  it into long term memory
• Long-term memory storehouse of facts about the
  world and how to do things
• memory three stage presentation
• encoding the process of putting things into
  memory system
• storage the way in which information is held or
  represented in the two memory systems
• retrieval our ability to access information in
  memory (forgetting)

Working memory
Verbal
Retrieval
Spatial
Encoding
(Learning and training)
Storage
Long-term memory
Procedural, declarative
organization, mental models
2

• WORKING MEMORY
• three core components
• verbal component
• phonological store info in linguistic form,
  words and sounds
• articulatory loop rehearsed by articulating
  words and sounds
• spatial component visuospatial sketchpad
• info in an analog, spatial form, often typical of
  visual images
• stores info in a particular form, or code
• central executive controls WM activity and
  assigns attentional resources to subsystems
• Code Interference
• verbal-phonetic and visual-spatial codes function
  cooperatively than competitively
• Brooks (1968) -- tasks should be designed so that
  the disruption by different concurrent activities
  does not occur (the greater resource competition
  with spatial working memory in the spatial
  response condition produced greater interference
  and vice versa)
• Interference in the Central Executive
• four core functions of the central executive
  (Baddeley, 1996)
• to coordinate performance on multiple tasks
• to temporarily hold and manipulate information
  stored in LTM
• to change retrieval strategies from LTM
• to attend selectively to stimuli

3

• a task demanding central executive resources
  should not be performed concurrently with other
  tasks drawing on those resources
• Matching Display with Working Memory Code
• principle of stimulus/central-processing/response
  compatibility best association of display
  formats to codes of working memory
• S display modality (auditory and visual)
• C two possible central-processing codes (verbal
  and spatial)
• R two possible response modalities (manual and
  vocal)
• S-C compatibility (fig 7.3)
• Limitations of Working Memory Duration and
  Capacity
• Duration
• how long does information in WM last if it is not
  rehearsed?
• Brown-Peterson paradigm decay function (fig
  7.4)
• without continuous rehearsal, little info is
  retained beyond 10 to 15 seconds
• transience applied to both spatial and verbal
  working memory
• augment the initial transient stimulus with a
  long-lasting visual display memory aid
• Capacity and Chunking
• capacity limit interacts with time (fig 7.4)
  the faster the rehearsal speed, the larger the
  capacity
• memory span the limiting number recalled with
  full attention
• max. capacity of WM -- 7 2 chunks of information
• chunk a set of adjacent stimulus units tied
  together by associations in the LTM

4

• chunking recording info. by semantically
  associating low-level elements (letters better
  than digits)
• chunking may be facilitated by parsing
  (physically separating likely chunks)
• optimum chunk size three to four for arbitrary
  alphanumeric strings
• Interference and Confusion
• MTBR lost from WM through interference from info
  learned at another time (fig 7.5)
• PI (proactive interference) activity engaged in
  prior to encoding the MTBR disrupts its retrieval
• pronounced, especially in a series of memory
  tasks with little time between them
• RI (retroactive interference) activity during
  the retention interval disrupts retrieval of the
  MTBR
• items in WM sometimes forgotten because they are
  confused with items held in the same time
• confusion similarity -- typically acoustic but
  sometimes semantic
• to minimize memory interference and confusion
• avoid creating large strings of similar sounding
  chunks
• use different codes (verbal vs. spatial) for
  different sources of information
• use digit strings that are particularly easy to
  remember
• ensure that the intervals before, during, and
  after storage are free of unnecessary activity
  using the same code (spatial or verbal) as the
  stored info
• Running Memory
• A sequence of stimuli is presented to the
  operator, who neither knows its length nor is
  expected to retain the entire sequence. Instead,
  a different response must be made to each
  stimulus or series of stimuli at some lag after
  they occur. memory span less than 72

5

• EXPERTISE AND MEMORY
• Expertise
• domain specific advantages in a specified
  domain
• acquired through practice or training in a domain
• provides a measuring performance advantage
• involve specialized, rather than generic,
  knowledge
• a task that defines the domain of expertise is
  called intrinsic
• a task that is not central to the domain of
  expertise, but greater expertise in the domain
  improves performance nonetheless, is called
  contrived
• Expertise and Chunking
• chunking strategies can be acquired through
  expertise
• store relevant stimulus material in WM in terms
  of chunks rather than lowest-level unit
• our ability to chunk information depends on our
  expertise in the subject domain
• Skilled Memory
• two aspects of performance in skilled tasks that
  are difficult for traditional view of WM
• skilled activities can be interrupted, and later
  resumed, with little effect on performance lt-gt
  transient
• performance in skilled tasks requires quick
  access to a large amount of info lt-gt limited
  capacity
• Ericsson and Kintsch (1995) WM includes another
  mechanism based on skilled use of storage in LTM
  long-term working memory (LT-WM)
• info. in LT-WM is stable, but accessed through
  temporarily active retrieval cues in WM

6

• domain specific skills allow LT-WM for the
  skilled activity (medical diagnosis, waiting
  tables, mental arithmetic)
• expert chess player -- retrieval structure (a set
  of retrieval cues) stored in WM to access the
  info stored in LT-WM
• chunking is thought of as a particular kind of
  retrieval structure
• PLANNING AND PROBLEM SLOVING
• a plan can be defined as a strategy for solving a
  problem
• planning and problem solving occur in the central
  executive
• planning difficulty increases when there are more
  choices available for action (fewer constraints)
• satisficing heuristic selects the current best
  plan with no guarantee that it is the absolute
  best
• implementation cost the cost associated with
  performing the actions resulting from a
  particular plan ? the amount of time required
• opportunistic planning -- the problem solving
  occurs by following the most promising leads at
  any point in time can lead to solutions bit are
  not optimal locally optimal but globally
  suboptimal
• automated planning system useful with multiple
  solutions but decreases exploratory behavior and
  lead to deterministic interpretation of presented
  info
• SITUATION AWARENESS
• situation awareness resides in WM (experts in
  LT-WM)
• relatively domain specific
• SA has important application for design
• it has implications for display design
• it has implications for automation

7

• LEARNING AND TRAINING
• Development of Expertise Learning
• two general cognitive models to explain learning
  ACT-R and Soar
• ACT-R production system model
• currently active information is compared to a set
  of production rules (IF-THEN statements)
• Procedural knowledge knowledge of how to do
  things (knowing how)
• declarative knowledge knowledge of facts
  (knowing what) chunks
• production rules embody procedural knowledge, but
  their conditions and actions are defined in terms
  of declarative knowledge in terms of chunks
• formation of production rules is a key part of
  learning
• emphasis on practice on specific example to
  improve learning in the ACT-R framework
• Soar a general cognitive architecture relying
  on production rules
• Soar does not distinguish between declarative and
  procedural knowledge
• Soar learns the particular response that must be
  made, and stores that as a chunk
• common characteristics of ACR-T and Soar of how
  cognitive learning occur
• emphasis on instances
• recall of the instance through chunking
• recall strategies for training and experience
• novices for open-ended strategies such as
  means-end analysis
• intelligent tutors based on ACT-R that teach
  cognitive skills

8

• Transfer of Training
• training efficiency
• the best learning in the shortest time
• longest retention
• is cheapest?
• transfer of training
• how much a new skill can capitalize what has been
  learned before?
• Measuring Transfer
• positive, zero, negative transfer
• transfer (control time transfer
  time)/(control time) x 100 savings/(control
  time) x 100
• transfer effectiveness ratio
• TER (amount of savings)/(transfer group time in
  training program) x 100
• training cost ratio
• TCR training cost in target environment (per
  unit time)
• training cost in the training program
  (per unit time)
• if TER x TCR gt 1, then the program is cost
  effective -- (if not, consider safety
  considerations)
• Training System Fidelity
• maximum positive transfer if all elements of a
  task were identical to the target task
• expensive , added realism not necessarily add to
  their TER

9

• instead of total fidelity, which components of
  training similar to the target task
• critical task components, processing demands, or
  task-relevant perceptual consistency
• Virtual Environments
• cost effective method for training
• tradeoff between TER and TCR -- lowering TER but
  increasing TCR
• more effective training technique for navigation
  of space
• Negative Transfer
• negative transfer is related to stages of
  processing (table 7.1)
• the lack of standardization in the control
  arrangements
• Training Techniques
• Practice and Overlearning
• practice makes perfect but how much practice
• after zero performance errors
• the speed of performance increases at a rate
  proportional to the log of the of trials
• attention or resource demand decreases
  automated fashion
• overlearning decreases the rate of forgetting of
  the skill
• Elaborative Rehearsal
• rehearsal is an active process, necessary to
  maintain chunks of info in WM
• maintenance rehearsal good way to maintain
  info. in WM but ineffective for transferring to
  LTM

10

• Reducing Cognitive Load
• cognitive load theory the effects of high
  demands on working memory on training
  effectiveness
• diagram, proximity-compatibility principle, dual
  modality presentation, worked sample
• Part-Task Training
• segmentation useful when segments of the skill
  vary greatly in their difficulty
• fractionization separate practice on two or
  more components
• allows attention to be focused on each component,
  reducing cognitive load and allowing automatic
  processing to develop
• separating task components prevents the
  development of time-sharing skills
• effective with independently broken off
  components and learnable consistencies
• effective if task components draw on different WM
  subsystems
• Guided Training
• error prevention
• training wheels approach offers explicit and
  immediate feedback about the error
• errors should not be eliminated completely
  long-term learning impaired
• Knowledge of Results (KR)
• feedback about the quality of performance, useful
  for motivation and performance
• delayed KR and the interval with other activities
  performance declined through RI
• KR during performance less well than after
  completion divided attention

11

• Learning by Example
• case studies annotations and elaboration
• not excessive cognitive load processing example
• Consistency of Mapping
• between target info. and the trainees response
• varied mapping increases in performance during
  training
• search for various targets in various contexts,
  feature learning procedure between consistent and
  varied mapping
• LTM
• Knowledge Representation
• procedural and declarative knowledge, semantic
  and episodic memory
• Knowledge Organization
• information is not stored in a random collection
  of facts, rather in structure and organization
• system features are congruent with the operators
  organization of that knowledge
• Methods for Representing Long-Term Knowledge
• organization of an expert users knowledge
  useful for training program, improving the design
  of an interface, or building a menu/index
  structure
• multiple knowledge acquisition techniques
• scaling technique gives a sense of how domain
  concepts are related to each other, usually by
  having experts rate pairs of concepts
• protocol technique perform typical tasks and
  think aloud
• interviews, observation, and document analysis

12

• conceptual graph analysis a representation of a
  users knowledge of a system
• Mental Models
• a mental structure that reflects the users
  understanding of a system may be created
  spontaneously by the user or carefully formed and
  structured though training
• incorporating a mental model into a training
  program can be effective -- visibility
• Memory Retrieval and Forgetting
• Recall and Recognition
• recall knowledge in the head
• recognition knowledge in the world (yes or no
  response or a choice response) more sensitive
• failure of recall and recognition (forgetting)
• RI, PI, similarity (confusion), the absence of
  retrieval cues, passage of time (recency)
• one particular kind of recall related to
  remembering to do something prospective memory
  checklists
• Skill Retention
• Degree of Overlearning
• additional practice after performance reaches
  error free -- automation
• Skill Type
• perceptual-motor skill very little forgetting
  over long time
• procedural skill more rapidly forgotten
• Individual Differences
• fast learners better retention than slow learner
  chunking skills