universal semantic communication brendan juba (harvard and mit) with madhu sudan (msr and mit) &...

49
Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Upload: dean-bexley

Post on 14-Dec-2015

220 views

Category:

Documents


2 download

TRANSCRIPT

Page 1: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Universal Semantic Communication

Brendan Juba (Harvard and MIT)with Madhu Sudan (MSR and MIT)

& Oded Goldreich (Weizmann)

Page 2: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)
Page 3: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)
Page 4: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

110100

110100

HOW DO WE DEFINE THE

“MEANING OF THE COMMUNICTATION?

??”

TO BE CONTINUED…

Page 5: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

MAN, WHAT THE EFF??A FAILURE

TO COMMUNICA

TE!

Page 6: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

I. Model of communicationII.Theory of finite

communicationIII.Example: computationIV.Model for infinite

communication

Outline

Page 7: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

“Meaning” = Usage

ENVIRONMENT

=

Page 8: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Printer

Printing, formally

Printer driver Printer firmware

ENVIRONMENTINTERFACE FIXED IN ADVANCE!

GOAL OF COMMUNICATION

Page 9: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

“USER”

“SERVER”

ENVIRONMENT

BEHAVIOR DEFINED

WITH GOAL

Abstract goals of communication“G = (ENV,R)”

FINITE GOAL OF COMMUNICATION: “USER

ACHIEVES GOAL” IF USER “HALTS” WHEN R = 1

R: g {0,1}

environment internal state

σu2σu1 σs

2σs1

U: Ωu × {0,1}* g Ωu × {0,1}* dist.over S: Ωs × {0,1}* g Ωs × {0,1}* dist.

over

Page 10: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Goal of computation (function f)

ENVIRONMENT

x

f(x)

R = “user message = f(x)?”

Page 11: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

1.Goal of Communication

2.Universal user

3.Sensing function

4.Helpful server

Key Concepts

Page 12: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Bob’s problem

??

I DON’T KNOW WHICH

ONE! P

BOB WANTS TO PRINT SUCCESSFULLY, REGARDLESS OF WHICH PRINTER HE IS USING

Page 13: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)
Page 14: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Universal user

NOTE: WE SHOULD SUCCEED FROM ANY STATE

ENVIRONMENT

P-Universal user for printing

P

Page 15: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

ENVIRONMENT

1101

ENVIRONMENT

1101

I’M THROUGH WITH YOU

THAT’S ALL I NEEDED TO

HEAR!

FROM ANY STATE??

I SURE BLEW

THAT…

Page 16: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Summary: universal user

Definition. A universal user for a goal G = (ENV,R) and a class of servers S is a user strategy s.t. for every server S in S and every initial state of S and ENV, the user achieves G.

That is, halts when R = 1

(w.h.p.)

WE WILL SAY THAT THE UNIVERSAL USER IS “EFFICIENT” IF, WITH EACH SERVER S IN S,THE USER RUNS IN SOME POLYNOMIAL TIME DEPENDING ON S, WITH THE GOAL-SPECIFIC SIZE PARAMETER DEPENDING ON ENV.

Page 17: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

I. Model of communicationII.Theory of finite

communicationIII.Example: computationIV.Model for infinite

communication

Outline

Page 18: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

IT’S ALL ABOUT THE FEEDBACK!!

Page 19: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

1.Goal of Communication

2.Universal user

3.Sensing function

4.Helpful server

Key Concepts

Page 20: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

ENVIRONMENTI CAN

STOP!

Sensing functions: “safety”

SENSING FUNCTION:

V : user’s view g {0,1}“V IS SAFE”:

V = 1 e R = 1 (w.h.p.)

RECALL, REFEREE:R : environment’s view g {0,1}

Page 21: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Sensing functions: “viability”

ENVIRONMENT

M

I CAN STOP

!

“V IS VIABLE” IF THERE EXISTS SOME USER STRATEGY THAT

RELIABLY OBTAINS V = 1

Page 22: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Theorem 1. If there is an efficiently computable S-safe and S-viable sensing function for a goal, then there is an efficient S-Universal user for that goal.

ENUMERATE ALL USER ALGORITHMS, RUN EACH WITH CONSTANT FACTOR OVERHEAD: SAFE & VIABLE SENSING FUNCTION INDICATES WHEN TO HALT

Achieving Universal Communication

Each algorithm of length l gets ≈ 1/l22l-

share of the total running time

Page 23: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Theorem 2. There is a natural class of 2l servers S s.t. a S-Universal user for any goal that requires the server to act experiences an overhead of Ω(2l) rounds.

IT TAKES ≈2l ROUNDS TO SEND

ALL 2l PASSWORDS OF LENGTH l!

NOTE: QUALITATIVELY OPTIMAL IN TERMS OF PROGRAM LENGTHS!

Theorem 2. There is a natural class of 2l servers S s.t. a S-Universal user for any goal that requires the server to act experiences an overhead of Ω(2l) rounds.

Might still consider restricted classes where we can be

efficient…

Page 24: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

So what is Theorem 1 good for??

CHARACTERIZATION IN TERMS OF SENSING FUNCTIONS CAN BE

USEFUL

Page 25: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Helpful servers

ENVIRONMENT

“S IS HELPFUL” IF THERE EXISTS SOME USER STRATEGY THAT

RELIABLY SUCCEEDS AT G

KEY DEF. #4…

Page 26: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

SG

Page 27: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

SG-Universal user for G

ENVIRONMENT

SG

NO COMMON KNOWLEDGE NECESSARY!

Page 28: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Theorem 3. If there is an efficient S-Universal user for a goal, then there is an efficiently computable S-safe and S-viable sensing function for that goal.

THE FUNCTION THAT TELLS A UNIVERSAL USER WHEN TO HALT IS A SAFE & VIABLE SENSING FUNCTION

Page 29: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Main Theorem. There is an efficient S-Universal user for a goal if and only if there is an efficiently computable S-safe and S-viable sensing function for the goal.

MORAL: SAFE & VIABLE SENSING FUNCTIONS ARE PRECISELY THE FUNCTIONS THAT TELL UNIVERSAL USERS WHEN TO HALT!

Page 30: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Theorem 4. If a sensing function is SG-safe for a goal G, then it is safe for G with all servers, even malicious and unhelpful ones.

CAN CONSTRUCT A HELPFUL SERVERTHAT BREAKS SAFETY WHENEVER SOME ADVERSARY CAN

Page 31: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

SG

Proof sketch: Theorem 4

ENVIRONMENTI CAN

STOP!

NOT SG-SAFE

FOR G

Page 32: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

RECAP: 1. Sensing is necessary and sufficient

2. Sensing with helpful servers must also be

safe with all servers

We’ll see a more concrete interpretation of these theorems

next…

Page 33: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

I. Model of communicationII.Theory of finite

communicationIII.Example: computationIV.Model for infinite

communication

Outline

Page 34: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Goal of computation (function f)

ENVIRONMENT

x

f(x)

R = “user message = f(x)?”

Page 35: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

For which problems can solutions be communicated

without common knowledge?

Page 36: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

SCompetitive Proof Systems

(Bellare-Goldwasser ‘94)

“x S”

SOUNDNESS(STANDARD)

PROVE IT!

YOU AREN’T FOOLING ANYONE!

COMPLETENESS(“COMPETITIVE

PROVER”)

WELL, I’M CONVINCED! EFFICIENT,

GIVEN ORACLE FOR

S

Page 37: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Theorem 5. Let G be the goal of deciding membership in a set S.

Then there is a SG-universal user for G iff there are competitive proof systems for both S and Sc.

Corollary. If there is a SG-universal user for G then S is in PSPACE.

Page 38: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

ENVIRONMENT

S

Theorem 5: obtaining a competitive proof system from a universal user

SG

x

S(x)

“x S”

NOT FOOLED: THEOREMS 3&4

TIME’S UP…

Page 39: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Theorem 5: obtaining a universal user from a competitive proof system

S

“x S”

x

HELPFUL SERVER

I WON’T BE FOOLED!

Page 40: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Computational problems with universal users

• Any PSPACE-complete problem [Shamir’92]• Any #P-complete problem [LFKN’92]• Graph Isomorphism [GMW’91]• Total functions in NP (solvable by

Levin’s universal search algorithm [Levin’73])– Integer Factoring– Discrete Logarithm– many more…

Page 41: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

I. Model of communicationII.Theory of finite

communicationIII.Example: computationIV.Model for infinite

communication

Outline

Page 42: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

REPEATING FINITE COMMUNICATION STRATEGY:PROBABILITY p OF FAILURE EACH SESSION…

REPEATING FINITE COMMUNICATION STRATEGY:PROBABILITY p OF FAILURE EACH SESSION…

Multi-session goals

ENV

SESSION 1 …SESSION 2 SESSION 3

INFINITE SESSION STRATEGY: ZERO ERRORS AFTER FINITE NUMBER OF ROUNDS

Page 43: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

Sensing for infinite goals

SESSION 1 …SESSION 2 SESSION 3

ENV

I’D BETTER TRY SOMETHING

ELSE!!

SAFETY: ERRORS DETECTED WITHIN FINITE # OF ROUNDSVIABILITY: FAILURES CEASE WITHIN FINITE # OF ROUNDS FOR AN APPROPRIATE COMMUNICATION STRATEGY

Page 44: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

This weaker version of sensing suffices to construct universal

users for infinite goals.

But is it necessary??

Page 45: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

110011011110

An impossible finite goal

ENVIRONMENT

I WONDER IF IT PRINTED…

RECALL: WE SHOULD STOP IN FINITE TIME

Page 46: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

110011011110

A possible infinite goal

ENVIRONMENT

PASSWORD FOUND IN FINITE # OF ROUNDSMORAL: FEEDBACK IS

UNNECESSARY!

Page 47: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

We saw a model for capturing problems of misunderstanding in communications systems.

We also saw some limits of “strong” solutions to this problem.

Page 48: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)

THERE EXISTS SOME USER STRATEGY THAT RELIABLY SUCCEEDS AT G

1.Goal of Communication

2.Helpful server

3.Universal user

4.Sensing function

Key Concepts

G = (ENV,R: g {0,1})

environment internal state

FOR EVERY SERVER S IN S AND EVERY INITIAL STATE OF S AND ENV, THE USER ACHIEVES G

V : user’s view g {0,1}

SAFETY: ERRORS DETECTED WITHIN FINITE # OF ROUNDSSAFETY: V = 1 e R =

1VIABILITY: FAILURES CEASE WITHIN FINITE # OF ROUNDS FOR AN APPROPRIATE COMMUNICATION STRATEGY

VIABILITY: THERE EXISTS SOME USER

STRATEGY THAT RELIABLY OBTAINS V =

1

Page 49: Universal Semantic Communication Brendan Juba (Harvard and MIT) with Madhu Sudan (MSR and MIT) & Oded Goldreich (Weizmann)