data-powered algorithms - i bernard chazelle princeton university bernard chazelle princeton...

74
Data-Powered Data-Powered Algorithms - I Algorithms - I Bernard Chazelle Bernard Chazelle Princeton University Princeton University

Post on 20-Jan-2016

218 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Data-PoweredData-PoweredAlgorithms - IAlgorithms - IData-PoweredData-PoweredAlgorithms - IAlgorithms - I

Bernard ChazelleBernard Chazelle

Princeton UniversityPrinceton University

Bernard ChazelleBernard Chazelle

Princeton UniversityPrinceton University

Page 2: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University
Page 3: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Lord Kelvin ( 1824-1907 )

“ X-rays will prove to be a hoax"

“ Radio has no future. "

Page 4: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Albert Einstein ( 1932 )

“ There is not the slightest indication that

nuclear energy will ever be obtainable. "

Page 5: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

"I think there is a world market

for maybe five computers."

Thomas WatsonIBM Chairman

(1943 )

Page 6: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

“Computing power doubles

every two years."

Gordon MooreIntel Co-founder

(1965 )

Moore’s LawMoore’s Law Moore’s LawMoore’s Law

Page 7: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University
Page 8: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University
Page 9: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Moore’s Law Moore’s Law repealedrepealed

Moore’s Law Moore’s Law repealedrepealed

In a few decades…

Page 10: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

“ There’s nothing to be discovered

in physics today. "

Lord Kelvin ( 1824-1907 )

Page 11: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Lord Chazelle (2006)

“ There’s nothing to be discovered

in computer science today. "

Page 12: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Lord Chazelle (2006)

“ There’s nothing to be discovered

in computer science today. "

Page 13: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

“ “ Computing will be the Computing will be the most most disruptive scientific disruptive scientific paradigmparadigm since quantum since quantum mechanics." mechanics."

Lord Chazelle (2006)

Page 14: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

“ … “ … and the end of Moore’s and the end of Moore’s Law willLaw will make this even more make this even more obvious." obvious."

Lord Chazelle (2006)

Page 15: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

What is computing ?

Universality

Self-reference

Duality

Tractability

Page 16: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University
Page 17: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

controlcontrol

data

program

Page 18: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Let ‘em eat cakePrint this

Let ‘em eat cake

Page 19: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Before Turing…Before Turing…

data

Page 20: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Before Turing…Before Turing…

data

Page 21: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Fishing …Fishing …

Page 22: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Fishing …Fishing …

Fishing manual

program data

Page 23: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Fishing …Fishing …

Confucius

Fishing manual

program data

Page 24: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

needs to know

how to fish

Page 25: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Fishing manual

program data

needs to know nothing about

fishing

Page 26: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

controlcontrol

data

program

knows nothin

g

Page 27: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

001010100010100010011111010001010

turn bits into sounds

Page 28: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

001010100010100010011111010001010

display/organize email

Page 29: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

0010101000101000100111110100010

10algebra

Earth simulator

Page 30: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University
Page 31: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University
Page 32: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Let ‘em eat cakePrint this

Let ‘em eat cake

signified

signifier

program data

Saussure(1857-1913)

Page 33: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Let ‘em eat cakePrint this

This is not a pipe

Magritte

Page 34: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Let ‘em eat cakePrint this

WHO’S ON FIRST ?

Abbott and Costello

Page 35: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University
Page 36: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Print thisPrint this

Print this

Page 37: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Print this twicePrint this twice

Print this twice Print this twice

Self-replication

Page 38: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

James Watson – Francis Crick, 1953

Page 39: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Self-reference: base pairs

Duality: gene/protein

signifier

signified

Page 40: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University
Page 41: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

all seem intractable

Protein folding

Scheduling

Theorem provingTraveling

salesman

Andrew Wiles

Map coloring

Page 42: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

equivalent

Protein folding

Scheduling

Theorem provingTraveling

salesman

Andrew Wiles

Map coloring

Page 43: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

intractable ?

Protein folding

Theorem provingTraveling

salesman

Andrew Wiles

Map coloring

E-commerce security

Page 44: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Two Amazing Two Amazing Consequences of Consequences of

IntractabilityIntractability

Two Amazing Two Amazing Consequences of Consequences of

IntractabilityIntractability

Zero Knowledge Zero Knowledge

Probabilistically Checkable Proofs Probabilistically Checkable Proofs

Page 45: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University
Page 46: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Are you richer

than me ?

dunno, but I won’t tell you how

much I’m worth

Bill

Bob

Page 47: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

I won’t tell you either

So, who’s richer ?

Bill

Bob

Page 48: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Bill

There exists a dialogue…

Bob

Page 49: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

blah blah blahblah blah blahblah blah blahblah blah blah

blah blah blahblah blah blahblah blah blahblah blah blah

Bill

Bob

Page 50: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Bill

at the end of which…

Bob

Page 51: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

1. They will know who is richer

2. They will have learned nothing else

( with probability 0.99999999999 )

Bob

Bill

Page 52: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Zero Knowledge

I have no nukes !

Prove it!

Page 53: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

1. No UN inspections

2. Both parties try to cheat

Page 54: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University
Page 55: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Who willbelieve

me?

My Proof of Riemann’s Hypothesis

Step 1 write proof in special format

Step 2 verifier picks 5 random words

Page 56: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

compiler

Page 57: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Verifier

There’s something wrong.

I REJECT !

Page 58: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Verifier

Everything looks fine.

I ACCEPT !

Verifier is correct withprobability 0.9999999

Page 59: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Verifier

There’s something wrong.

I REJECT !

If my 2000-page proof is wrong in only one step, how can verifier spot an error in 5 random words?

Page 60: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Verifier

Everything looks fine.

I ACCEPT !

How does verifier know I proved Riemann’s hypothesis and not

2+2=4 ?

Page 61: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University
Page 62: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Very little does a lot

Page 63: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

32x 17

22432

= 544

grade school

Page 64: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

FFT

signal processing

Page 65: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

RSA

encryption e-commerce

Page 66: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

PageRank

web search

Page 67: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Biomedical imaging

Sloan Digital Sky

Survey10 petabytes10 petabytes

(~1MG)(~1MG)10 petabytes10 petabytes

(~1MG)(~1MG)

10 10 petabytes/yrpetabytes/yr

10 10 petabytes/yrpetabytes/yr

150 petabytes/yr150 petabytes/yr150 petabytes/yr150 petabytes/yr10,000 times theLibrary of Congress

Page 68: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Understanding Biological Function

• 100s of sequenced genomes• Function of many genes unknown

– 30% for yeast

Genome Proteome

Page 69: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Interaction Networks• High-throughput experiments (Yeast two-hybrid,

etc)• Form networks of interactions

Page 70: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

protein-protein interaction networks

Page 71: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Barabasi, AL. et al. (2004) Nat. Rev. Genet.Spirin, V. et al. (2003) PNAS

Analysis of Interaction Networks

Yeger-Lotem, E. et al. (2004) PNAS

Page 72: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Sciences of The Formula

math, physics, chemistry

Page 73: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

Sciences of The Algorithm

Abu Abdullah Muhammad bin Musa al-Khwarizm (780-850)

Page 74: Data-Powered Algorithms - I Bernard Chazelle Princeton University Bernard Chazelle Princeton University

“ If Google is a religion, what is its God?

It would have to be The Algorithm. “