View
220
Download
0
Embed Size (px)
Citation preview
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Syllabus
http://gandalf.psych.umn.edu/~schrater/schrater_lab/courses/MathMod06/
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Mathematical Model• Definition of a “Model”: • A model is a simplified representation
of some aspect of the real world. • Mathematical Models
– representation of relationships between numerical or symbolic representations of measurements and world properties.
– Relation example: • weight relations between two objects
x R y if and only if x is heavier than y
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Role of models in science
The World The Model
PredictionsData
Derivation
Interpretation
Experimentation
Abstraction Relations betweenDifferent World properties
Relations betweenWorld properties And measurementCollection of measurements
With the procedure for Gathering it
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Overview• Mathematical models of Human Behavior
– Types of model• Descriptive: Relations between measurements• Predictive: Relations between world model and
measurements• Causal: Predictive with directed relations between world
properties– Goal of models
• Representing behavior in symbols and relations• Predicting behavior• Summarizing large bodies of data• Making assumptions and theories explicit and testable
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Measurements and world properties• Example: Measuring the mind
– What does it mean to measure the mind? How do we abstract mind?
• Relations between Measurements– Why can’t we add IQs? A silly idea-The IQ of a committee add in parallel:
€
IQgroup =1
1
IQ1
+1
IQ2
+ ...+1
IQN
€
IQgroup =1
1
120+
1
140+
1
150
= 45For Example:
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
What kind of model is this?• Fitts' law is a model of human psychomotor behavior for speed/accuracy
tradeoffs in rapid, aimed movement (not drawing or writing). According to Fitts’ Law, the time to move and point to a target of width W at a distance A is a logarithmic function of the spatial relative error (A/W)
• MT = a + b log2(2A/W + c)
• where• MT is the movement time• a and b are empirically determined constants, that are device dependent.• c is a constant of 0, 0.5 or 1 • A is the distance (or amplitude) of movement from start to target center• W is the width of the target, which corresponds to “accuracy”
= log2(2A/W + c)
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Causal theories specification of causes
Natural or man-made?
How can we describe what generated these patterns?
Some questions can’t be addressed without causal assumptions.Relations between elements in the theory are not symmetric.
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Abstracting Human Behavior• World Abstractions
– Events (relations between time, place, and objects/agents)– Outcomes (relations between actions and world)
• Behavior Abstractions– Goals/Values (Rewards, gains, losses)
• defined over outcomes• Relations betweens goals/values- ( utility/preferences)
– Beliefs (Subjective probability)• Defined over events• Relations between beliefs (certainty)
– Actions (Moves, choices, decisions, communication,etc.)• Relation between events, actor, and outcomes• Relations between actions (plans, causes)
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Behavior Modeling• Behavior theory-
– Define relations between goals, values, and beliefs– Derive actions from goals, values and beliefs
• Behavior measurement– Methods for quantifying actions– Only actions are measurable-all other behavior properties
are theoretical, and require a predictive model to connect to measurables.
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Speech Generation• Goal- Deliver a message• Events- utterances• Outcomes - sound fidelity to intention, comprehension• Beliefs- Ideas -> words; words -> sounds• Actions- Facial, esophageal, rib cage muscle
movements
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Example: Speech generationVoice Puppetry, M. Brand;
Siggraph’99
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Can we fit natural language behavior in this paradigm?
Goal of language behavior?
Beliefs?
Actions?
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Can we fit natural language behavior in this paradigm?
Goal of language behavior?
Convey some meaning
Beliefs?
Meaning generated by other’s parsing ofthe sentence
Actions?
Sentence generation
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Signs of a good theory• Using a small number of principles, be able to derive
detailed consequences that can be specialized to many different situations.
• Moreover, these consequences can be converted into measurable predictions that can be compared to experiment.
• Example from physics: Classical Mechanics and the principle of least action:– The path taken by an object will minimize the “action” (the
conversion of potential to kinetic energy).
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Least action demo
http://www.eftaylor.com/software/ActionApplets/LeastAction.html
Which path will the ball take?
Kinetic Energy
K = M v2
U = g y
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Are there similar principles for human behavior?
• Some of you may operate according to these principles:– Sloth principle: Minimize effort. Only do what you have to?– Hedonic principle: Maximize those good times?– Power principle: Maximize influence? Only I can rule the
world.– Evolutionary principle: Maximize survival/number of
progeny?• Serious proposal
– Maximize value, the expected utility of an action.
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Theoretical framework underlying almost all models of human behavior
• Decision Theory/Game TheoryModel human behavior via a Maximization Principle: Behavior
achieves goals by maximizing value for the organism. 1. People model the world internally and formulate beliefs about it.2. People ascribe values to different world states and actions
John von Neumann John Nash Duncan Luce
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Overview• Modeling Beliefs
– Belief representation– Belief formation– Belief revision
• Modeling Utility for different domains– Utility for simple cognitive judgments– Utility for simple perceptual judgments– Utility for interpersonal interactions– Utility for simple motor actions (e.g. reaching)– Utility for mate selection
• Modeling Learning
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Modeling BeliefsRoger N. Shepard
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Roger N. Shepard
Example: Modeling Beliefs
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Roger N. Shepard
Example: Modeling Beliefs
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Example: Modeling BeliefsRoger N. Shepard
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Beliefs involve representing certainty about the presence of abstracted world properties internally
What are the world properties?
What is the abstraction?
What is the belief?
Pigment changesSurface changesMaterial changes
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Homework requires Matlab• “BASIC for people who like linear algebra”• Full programming language
– Interpreted language (command)– Scriptable– Define functions (compilable)
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Data
• Basic- Double precision arraysA = [ 1 2 3 4 5]A = [ 1 2; 3 4]B = cat(3,A,A) %three dimensional array
Advanced- Cell arrays and structuresA(1).name = ‘Paul’A(2).name = ‘Harry’
A = {‘Paul’;’Harry’;’Jane’}; >> A{1} => Paul
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Almost all commands Vectorized• A = [ 1 2 3 4 5 ] ; B = [ 2 3 4 5 6]
– C = A+B – C = A.*B – C = A*B’– C = [A;B]– sin( C ), exp( C )
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Useful commands• Colon operator
– Make vectors: a = 1:0.9:10; ind = 1:10– Grab parts of a vector: a(1:10) = a(ind)– A = [ 1 2; 3 4]– A(:,2)– A(:) = [ 1
324]
Vectorwise logical expressionsa = [ 1 2 3 1 5 1]a = =1 => [ 1 0 0 1 0 1]
size( ), whos, help, lookforls, cd, pwd, Indices = find( a = =1 ) => [ 1 4 6 ]
PSY 5018H: Math Models Hum Behavior, Prof. Paul Schrater, Spring 2006
Stats Commands• Summary statistics, like
– Mean(), Std(), var(), cov(), corrcoef()• Distributions:
– normpdf(),• Random number generation
– P = mod(a*x+b,c)rand(), randn(), binornd()
• Analysis tools– regress(), etc