Академический Документы
Профессиональный Документы
Культура Документы
Levels of analysis
Tom Griffiths
Admin
Sections are on this week (sections 107
and 108, go to the other section in the
same timeslot)
Office hours are up on the syllabus on
bCourses
Announcement gives some links for
resources to transition to Python,
special session planned for 9/8
Levels of analysis
At what level should we try to model
human cognition?
Outline
Levels of analysis
Break
The computational level and cognition
David Marr
David Marr
1945-1980
1982
Implementation
How can the representation and algorithm be
realized physically?
Analyzing information
processing systems
What is being computed?
Why is it being computed?
Computational theory
1. What is being computed?
2. Why is it being computed?
where 1. is a solution to the computational
problem specified by 2.
Analyzing information
processing systems
What representation?
What algorithm?
What representation?
Many formal systems solve the same
computational problems
e.g. 2+2=4 and 010+010=100
Analyzing information
processing systems
What is the physical
implementation of the
system?
Computation
What is the goal of the computation, why is it
appropriate, and what is the logic of the strategy
by which it can be carried out?
constrains
Implementation
How can the representation and algorithm be
realized physically?
Implementation
How can the representation and algorithm be
Neuroscience
realized physically?
Implementation
How can
the representation
and algorithm
be
Models
based
on how neurons
compute
realized physically?
An example: Memory
An example: Memory
Implementation
How can the
representation
algorithm
be
Explaining
human
memoryand
based
on the
realizedformed
physically?
circuits
by hippocampal neurons
What computations are
supported by different
kinds of neurons?
What is the capacity of
memory systems made
from these neurons?
(Treves & Rolls, 1994)
An example: Memory
Representation and algorithm
How can the
representation
algorithm
be
Explaining
human
memoryand
based
on the
realized physically?
representation
of items with binary features,
and a rule for determining familiarity
Item 1: 00101011
Item 2: 01010010
Item 3: 11011010
An example: Memory
Computation
How can the
representation
algorithm
be
Explaining
human
memoryand
as an
optimal
realized physically?
solution
to the statistical problem of
identifying items likely to be needed again
(Anderson, 1990)
Break
Up next:
The computational level and cognition
Purposive: why?
function/problem
optimal solution
output
input
output
input
output
An example: Memory
Computation
How can the
representation
algorithm
be
Explaining
human
memoryand
as an
optimal
realized physically?
solution
to the statistical problem of
identifying items likely to be needed again
(Anderson, 1990)
Questions
How does one go about conducting a
computational-level analysis?
What is the equivalent of aerodynamics
for cognition?
Are there any dangers of pursuing
explanations at the computational level?
Questions
How does one go about conducting a
computational-level analysis?
What is the equivalent of aerodynamics
for cognition?
Are there any dangers of pursuing
explanations at the computational level?
Optimization
The key to producing a purposive explanation
Provides a potential connection between
function and behavior, if theres a reason to
believe that behavior should be optimal
if people solve a problem badly, then thats not
the answer to a why question
Adaptation is typically
explicitly evolutionary
(people can learn too)
Questions
How does one go about conducting a
computational-level analysis?
What is the equivalent of aerodynamics
for cognition?
Are there any dangers of pursuing
explanations at the computational level?
Cognitive aerodynamics
What is the mathematical theory that
characterizes optimal solutions for the
computational problems that people face?
Depends on the computational problem
deductive reasoning: logic
inductive reasoning: statistics
Questions
How does one go about conducting a
computational-level analysis?
What is the equivalent of aerodynamics
for cognition?
Are there any dangers of pursuing
explanations at the computational level?
Levels of analysis
At what level should we try to model
human cognition?
computational
problem
algorithm
implementation
Thursday
The start of rules and symbols
read Haugeland on formal systems
read AIMA for background on logic