Connections between Computer Science and Psychology

1
Connections between Computer Science and
Psychology
2
connections

• The brain is a remarkable computer (Hinton)
• computer tool design based on psychological
  theories
• intelligent tutoring systems
• treating autism
• artificial intelligence can computers understand
  speech? recognize faces? be creative?
• neural network algorithms in computer science

3
PAT Practical Algebra Tutor

• PAT has the defining feature of containing a
  psychological model of the cognitive processes
  behind successful and near-successful student
  performance...this cognitive model is written as
  a system of if-then rules...when help is needed,
  the tutor...can provide hints that are
  individualized to the students particular
  approach
• Intelligent Tutoring Goes To School in the Big
  City, Koeginger, Anderson, Hadley, and Mark

4
computers and autistic children

• there is a growing interest in using computers
  with autistic children, both for general
  education and for stimulating communication.
  There appear to be three main reasons why
  children with autism are attracted to the
  computer (1) it involves no social factors (2)
  it is consistent and predictable and (3) it
  allows a child to take active control and work at
  his/her own pace.
• John Swettenham, Departments of Experimental
  Psychology and Psychiatry, U. Cambridge

5
understanding the brain

• the brain performs many computational tasks
• pattern matching

6
how does the brain compute?

• one approach
• guess at models of the brain
• study the computational power and limitations of
  these models (theory simulation)
• compare with capabilities of the brain
• computers ideal for model simulation
• computer science also provides ways to develop
  the theory

7
neural network models of the brain

• gross idealizations
• still, useful to test theories of how brain
  processes function
• rule out poor theories
• hone in on important features

8
neuron
9
signals between neurons
10
artificial neuron
input
weight
weighted input
1/2
neural unit
1/2
1/3
output
0
sum
sum gt 1?
-1/2
0
-1
-1
0
0
threshold function if sum gt 1, output 1 if sum
lt 1, output 0
multiply the input and weight to get the weighted
inputs
11
artificial neural network
neural units
outputs
inputs
...
...
arrows have weights
12
example

• recognizing a simple pattern checker
• algorithm input is four bits, representing the
  colour of each square, ordered from top left to
  bottom right 1001
• algorithm output should be "1" on input 1001
• algorithm output should be "0" on all other inputs

13
program to recognize checker

• if (bit 1 is 1) and
• (bit 2 is 0) and
• (bit 3 is 0) and
• (bit 4 is 1) then output yes
• else output no

14
neural network to recognize checker

• modular design (partially completed)

inputs 2 and 3 are 0?
0
-1
1
both conditions are true?
1
1
inputs 1 and 4 are 1?
0
15
unit for inputs 2 and 3 are 0
input
weight
weighted input
0
neural unit
1
output
sum
sum gt 1?

1
0
16
unit for inputs 2 and 3 are 0outputs "0" on
input 1001
input
weight
weighted input
0
neural unit
0
1
output
0
sum
sum gt 1?
0
0
0
1
0
0
17
unit for inputs 2 and 3 are 0outputs "1" on
input 1101
input
weight
weighted input
0
neural unit
0
1
output
1
sum
sum gt 1?
1
1
0
1
0
0
18
neural network to recognize checker
if yes output 0 if no output 1
inputs 2 and 3 are 0
0
-1
1
both conditions are true
1
1
inputs 1 and 4 are 1
if yes output 1 if no output 0
0
19
unit for inputs 1 and 4 are 1?

• what weights would you use?

input
weight
weighted input
neural unit
output
sum
sum gt 1?

20
unit for inputs 1 and 4 are 1

• let's try weights of 0 for the "irrelevant"
  inputs and weights of 1 for the "relevant" inputs
  1 and 4.

input
weight
weighted input
1
neural unit
0
output
sum
sum gt 1?

0
1
21
unit for inputs 1 and 4 are 1

• unit does the right thing on input 1001!

input
weight
weighted input
1
neural unit
1
0
output
0
sum
sum gt 1?
2
1
0
0
1
1
22
unit for inputs 1 and 4 are 1

• but the unit does the wrong thing on input 1000
  ?

input
weight
weighted input
1
neural unit
1
0
output
0
sum
sum gt 1?
1
1
0
0
0
1
23
correct unit for inputs 1 and 4 are 1
input
weight
weighted input
1/2
neural unit
1/2
0
output
0
sum
sum gt 1?
1
1
0
0
1/2
1/2