PowerPoint Presentation Computational Aesthetics and the Science of Fun

1
Computational Aesthetics And the Science of Fun
2
(No Transcript)
3
(No Transcript)
4
Birkhoffs Aesthetic Measure
A O/C
O - Order C - Complexity
Horizontal, Vertical, Proportion,Tangent
5
(No Transcript)
6
T. Staudek On Birkhoffs Aesthetic , fimu-rs-99
7
More on Aesthetic Measures
A O/C
O - Order H(x)-H(xy) C - Complexity H(x)
x - present y - past / prior knowledge
A H(x)-H(xy)/H(x)
8
Info-Rate Model
Aesthetic perception as a communication process
Information that influences the cognitive
state of the information receiver
The information paradox Discover more by
listening more. (you gain by learning, not get
bored!)
A I(x,y)/H(x)
9
Info-Rate (cont.)
Assuming a space M of mental models (prior) and
space M of actual (posterior) models, we can
write
I(x,y) I(x,yM)H(M)D(M,M)
Information rate Prediction Explanation
Cost Likelihood

• I(x,yM) prediction using past data y and
  model M (Coding Gain)
• H(M) Model uncertainty (size of model space)
• D(M,M) Distance between mental and actual
  model distribution

(Model Cost)
10
Itti Baldi Bayesian Surprise nips 06
11
Wow and Aha
12
(No Transcript)
13
The Atoms of EVE A Bayesian Basis for Aesthetic
Analysis Artificial Intelligence for Engineering
Design, Analysis and Manufacturing (AIEDAM)
Expectation Violation Explanation
Set-up Punchline Get-it?
Garden Eaten Tragedy?
Analyzing humor is like dissecting a frog. Few
people are interested
and the frog dies of it.
Comedy?
14
Aha!
Wow!
15
Finding Interest Points
MEX (Memory and Expectation) Patterns
Aha Wow
16
(No Transcript)
17
Machine Improvisation
Corea Orig Impro
Memex Piano Violin
18
Music Models
19
Shlomo Dubnov Thoughts About Memex.
http//music.ucsd.edu/sdubnov
Memex Music
Memex, the machine (Bush 1945), was a futuristic
device, For creating and recalling associations
In the form of memory trails. Memex, the music
(Dubnov 2006), is an algorithmic
composition, Designed to create new music from
old music By associations along probabilistic
trails.
20
Lets say that the current note in Memex is G -
taken from Bach. To get the next note The
machine will step forward with probability Q - or
jump backward with probability P Where jump
backward is to the same note (different song)
with most similar history.
Bach C D F E C G C
Q
P
Beethoven A F E C G A C
4
Mozart D C G B B A
2
If it steps, the next note is C. If it jumps, the
next note is A. Beethovens 4 gt Mozarts 2.
21
(No Transcript)
22
(No Transcript)
23
Kahneman and Tversky (1979). Prospect Theory An
Analysis of Decision Under Risk. Econometrica,
Vol. 47, No. 2, pp 263-291.
24
Apparent Contribution from Aesthetic Utility W
- P
25
Fun (flow) function computed by EVE with same
G/G as for slots. All E was assumed to be
positive and Resolution (R) was set to Q2.
P was tweaked by the human creator until the
machine composition sounded best, which turned
out to be much like slots a P of 13.
26
(No Transcript)
27
Measure of Expectation (E)
This E is a negative entropy
28
Explanation (E)
Slots
E - H P log P R H- Q log Q
R-
H/H- sense of humor, R Bayesian resolution
Memex
E - Q2 P log P Q log Q
29
Thanks!