Cognitive Modeling Flashcards
2 types of arrival patterns
standard; non-standard
2 types of service patterns
state-independent
state-dependent
Three queuing principles
- FCFS
- LCFS
- pre-emptive
Processing capacity
how many customers can be served at once
A/B/X/Y/Z
A = arrival pattern (M, N, D, G) B = service pattern (M, N, D, G) X = processing capacity / number of parallel lines/channels Y = buffer size Z = queue principle (FCFS, etc)
Traffic intensity function
P = utilization of server gamma = arrival rate mu = service speed/rate P = gamma/mu
W
Wait time
L
Expected number of customers
Mental workload
directly proportionate to average utilization (linear)
Queuing applications in human performance
- quantitative mechanism to understand human performance
- experiment may be too hard/expensive/impossible
- unify many experiments (MHP)
- used in interface design
Ideal human processing model expectations
- address internal mechanisms (neuroscience/cog structure)
Parietal cortex
memory, language
Occipital lobe
visual signals
Wernicke’s area / Broca’s area
language
Inner ear
- hairs transmit signals
- balance
EEG vs. ERP
EEG - no obvious external stimuli
ERP - triggered by event
Reaction time
1/mu
RT
realization time + wait time
SOA
Time between arrival times
Relation between neuroscience and cognitive modeling
- Human behavior is the result of interaction between physical and mental activities in body and brain
- Human neural system is the ground of cognitive model; we no longer work on the black-box assumption
ACT-R evolution
HAM->ACT-R 1.0 -> ACT-R/PM
Perceptual and motor from epic
QN-MHP
- Cognitive subnetwork
- One serial server (F), other servers are parallel
Keys to building your own cognitive simulation model (planning)
(1) how to model human operators
(2) how to model interaction between the human operators and the world
(3) how many servers you need
(4) how to connect those servers
Keys to building your own cognitive simulation model (coding)
(1) processing time of servers -> using Qn-MHP or MHP
(2) Capacity of servers
(3) Response recording (response, workload)
Keys to building your own cognitive simulation model (key principles)
- Always set the model’s parameters according to: MHP, Literature, Cognitive mechanisms, experimental settings
- Never code to match your data
Measurements of mental workload
- Performance-based
- Subjective
- Psychophysiological
Advantages to psychophysiological
- do not add a new task to measure workload
- real-time
P300 amplitude
- reliable and real-time index of mental workload
- ERP
- a positive component of ERP with latency between 300 and 800
Existing adaptive workload management systems
- phone adaptive system
- voice adaptive system
- in-vehicle message system
How to set parameters
- literature
- experimental settings
- model’s initial setting (MHP - ex)
- free parameters
Benefits of math model
- rigorous quantification of the system
- equations are easy to be reused, extended, transferred
- produce analytical solutions
Disadvantages of math model
- system’s with high complexity are hard to model
- deterministic - only have point estimation
Benefits of simulation model
- no limitation of complexity of system to be modeled
- can be used when math model is NP hard
- distribution prediction
Disadvantages to simulation model
- lots of iterations -> closer to analytical solution
- codes are not easily transferable to math equation
Exponential model
E(avg-min, 1, min), then change to seconds
Core assumptions of EPIC
- no limitation on cognitive capacity
- perceptual motor have limited capacity
Core assumptions of ACT-R
Cognitive processing serial (capacity = 1)
One production rule fired at a time
Core assumption of SOAR
no limitations anywhere
Frequency in promodel
inter-arrival time
Processing logic in promodel
when your entities arrive at current location, what are the processes? - happens automatically
Move logic in promodel
happens only when destination server is available
Things learned from papers
- must have R^2 and RMS
- must have experimental validation
- detailed analysis of human cognition
- include codes
- show model and experimental results together
- no single server model
attribute
label specific towards this entity
Improve model
- parameters consistent with task
- understand which part of model influences data a lot
- check R^2 and RMS
- free parameters
How to apply/use model
- Goal 1: improve human performance/reduce workload (use optimization/scheduling to develop new systems)
- Goal 2: explain more experimental results (change model’s input/output)
3 different ways to model human brain network
(1) neurological
(2) math
(3) simulation
Promodel process
- locations
- entities
- processing (processing logic and move logic)
- arrivals (inter-arrival time)
- (variables)
- (general information)/(simulation options)
Additional promodel process for clock
- attribute
- arrays
- arrival logic
- processing logic
- Global variable
Johnson’s rule
doesn’t know which arrives first, just calculates the optimal
