Toward the KRW conjecture Cubic Formula Lower Bounds via - - PowerPoint PPT Presentation

toward the krw conjecture
SMART_READER_LITE
LIVE PREVIEW

Toward the KRW conjecture Cubic Formula Lower Bounds via - - PowerPoint PPT Presentation

Toward the KRW conjecture Cubic Formula Lower Bounds via Communication Complexity Irit Dinur Or Meir Irit Dinur, Or Meir Toward the KRW conjecture Formulas (de-Morgan) Formulas are circuits with fan-out 1 . Cannot store intermediate results.


slide-1
SLIDE 1

Toward the KRW conjecture

Cubic Formula Lower Bounds via Communication Complexity Irit Dinur Or Meir

Irit Dinur, Or Meir Toward the KRW conjecture

slide-2
SLIDE 2

Formulas

(de-Morgan) Formulas are circuits with fan-out 1. Cannot store intermediate results. The formula complexity L(f) is the size of the smallest formula for f.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-3
SLIDE 3

Formulas

Would like: Explicit f with L(f) = nω(1).

Irit Dinur, Or Meir Toward the KRW conjecture

slide-4
SLIDE 4

Formulas

Would like: Explicit f with L(f) = nω(1). State-of-the-art: Andreev’s function [A87] has complexity ˜ Ω(n3) [H98].

Irit Dinur, Or Meir Toward the KRW conjecture

slide-5
SLIDE 5

Formulas

Would like: Explicit f with L(f) = nω(1). State-of-the-art: Andreev’s function [A87] has complexity ˜ Ω(n3) [H98]. Proof based on shrinkage [S61, IN93, PZ93. H98, T14]. We now have an optimal analysis of shrinkage.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-6
SLIDE 6

Formulas

Would like: Explicit f with L(f) = nω(1). State-of-the-art: Andreev’s function [A87] has complexity ˜ Ω(n3) [H98]. Proof based on shrinkage [S61, IN93, PZ93. H98, T14]. We now have an optimal analysis of shrinkage. So shrinkage is unlikely to get us any further.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-7
SLIDE 7

Composition

[KRW91]: we need to understand composition.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-8
SLIDE 8

Composition

[KRW91]: we need to understand composition. Let f : {0, 1}m → {0, 1}, g : {0, 1}n → {0, 1}. The composition f ⋄ g : {0, 1}m×n → {0, 1} is

m n

X

Irit Dinur, Or Meir Toward the KRW conjecture

slide-9
SLIDE 9

Composition

[KRW91]: we need to understand composition. Let f : {0, 1}m → {0, 1}, g : {0, 1}n → {0, 1}. The composition f ⋄ g : {0, 1}m×n → {0, 1} is

g m n g g ..........

X

a

Irit Dinur, Or Meir Toward the KRW conjecture

slide-10
SLIDE 10

Composition

[KRW91]: we need to understand composition. Let f : {0, 1}m → {0, 1}, g : {0, 1}n → {0, 1}. The composition f ⋄ g : {0, 1}m×n → {0, 1} is

g f (fg)(X) m n g g ..........

X

a

Irit Dinur, Or Meir Toward the KRW conjecture

slide-11
SLIDE 11

The KRW conjecture

g f (fg)(X) m n f g g

X 1 X m

......... ......... L(g) L(f) g g ..........

X

a

L(g)

Clearly, L(f ⋄ g) ≤ L(f) · L(g).

Irit Dinur, Or Meir Toward the KRW conjecture

slide-12
SLIDE 12

The KRW conjecture

g f (fg)(X) m n f g g

X 1 X m

......... ......... L(g) L(f) g g ..........

X

a

L(g)

Clearly, L(f ⋄ g) ≤ L(f) · L(g). KRW conjecture*: ∀f, g : L(f ⋄ g) ≈ L(f) · L(g).

Irit Dinur, Or Meir Toward the KRW conjecture

slide-13
SLIDE 13

The KRW conjecture — Prior Work

KRW conjecture*: ∀f, g : L(f ⋄ g) ≈ L(f) · L(g). Implies super-polynomial formula lower bounds.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-14
SLIDE 14

The KRW conjecture — Prior Work

KRW conjecture*: ∀f, g : L(f ⋄ g) ≈ L(f) · L(g). Implies super-polynomial formula lower bounds. [KRW91] defined the universal relation U. Like a function, but simpler. Suggested to prove the conjecture for U ⋄ U.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-15
SLIDE 15

The KRW conjecture — Prior Work

KRW conjecture*: ∀f, g : L(f ⋄ g) ≈ L(f) · L(g). Implies super-polynomial formula lower bounds. [KRW91] defined the universal relation U. Like a function, but simpler. Suggested to prove the conjecture for U ⋄ U. Conjecture was proved for U ⋄ U by [EIRS91] Alternative proof by [HW93]. Recently: [GMWW14] proved conjecture for f ⋄ U.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-16
SLIDE 16

Our results

We prove conjecture for f ⋄ g where g is parity: L(f ⋄ g) ≥ L(f) · L(g)/2

˜ O(√m)

Irit Dinur, Or Meir Toward the KRW conjecture

slide-17
SLIDE 17

Our results

We prove conjecture for f ⋄ g where g is parity: L(f ⋄ g) ≥ L(f) · L(g)/2

˜ O(√m)

Also, we give a structural result:

Irit Dinur, Or Meir Toward the KRW conjecture

slide-18
SLIDE 18

Our results

We prove conjecture for f ⋄ g where g is parity: L(f ⋄ g) ≥ L(f) · L(g)/2

˜ O(√m)

Also, we give a structural result:

KRW conjecture: naive formula is optimal.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-19
SLIDE 19

Our results

We prove conjecture for f ⋄ g where g is parity: L(f ⋄ g) ≥ L(f) · L(g)/2

˜ O(√m)

Also, we give a structural result:

KRW conjecture: naive formula is optimal. Our result: naive formula is essentially the only optimal formula.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-20
SLIDE 20

Our results

We prove conjecture for f ⋄ g where g is parity: L(f ⋄ g) ≥ L(f) · L(g)/2

˜ O(√m)

Also, we give a structural result:

KRW conjecture: naive formula is optimal. Our result: naive formula is essentially the only optimal formula.

Actually, the lower bound for f ⋄ g already follows from [H98]. However, our proof is very different, and seems to be more generalizable for other g’s.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-21
SLIDE 21

Our results

We prove conjecture for f ⋄ g where g is parity: L(f ⋄ g) ≥ L(f) · L(g)/2

˜ O(√m)

Also, we give a structural result:

KRW conjecture: naive formula is optimal. Our result: naive formula is essentially the only optimal formula.

Actually, the lower bound for f ⋄ g already follows from [H98]. However, our proof is very different, and seems to be more generalizable for other g’s. Also: new proof of the state-of-the-art cubic lower bound

  • f [H98].

Irit Dinur, Or Meir Toward the KRW conjecture

slide-22
SLIDE 22

1

Introduction

2

Background

3

Proof Strategy

4

New tools

Irit Dinur, Or Meir Toward the KRW conjecture

slide-23
SLIDE 23

Outline

1

Introduction

2

Background

3

Proof Strategy

4

New tools

Irit Dinur, Or Meir Toward the KRW conjecture

slide-24
SLIDE 24

Karchmer-Wigderson Relations [KW90]

Relate L(f) to complexity of a communication problem KWf.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-25
SLIDE 25

Karchmer-Wigderson Relations [KW90]

Relate L(f) to complexity of a communication problem KWf. The KW relation KWf is defined as follows:

Alice gets x ∈ f −1(0). Bob gets y ∈ f −1(1).

Irit Dinur, Or Meir Toward the KRW conjecture

slide-26
SLIDE 26

Karchmer-Wigderson Relations [KW90]

Relate L(f) to complexity of a communication problem KWf. The KW relation KWf is defined as follows:

Alice gets x ∈ f −1(0). Bob gets y ∈ f −1(1). Clearly, x = y, so ∃i s.t. xi = yi. Want to find such i.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-27
SLIDE 27

Karchmer-Wigderson Relations [KW90]

Relate L(f) to complexity of a communication problem KWf. The KW relation KWf is defined as follows:

Alice gets x ∈ f −1(0). Bob gets y ∈ f −1(1). Clearly, x = y, so ∃i s.t. xi = yi. Want to find such i. Want to talk as little as possible.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-28
SLIDE 28

Karchmer-Wigderson Relations [KW90]

Relate L(f) to complexity of a communication problem KWf. The KW relation KWf is defined as follows:

Alice gets x ∈ f −1(0). Bob gets y ∈ f −1(1). Clearly, x = y, so ∃i s.t. xi = yi. Want to find such i. Want to talk as little as possible.

Only deterministic protocols!

Irit Dinur, Or Meir Toward the KRW conjecture

slide-29
SLIDE 29

Karchmer-Wigderson Relations [KW90]

Relate L(f) to complexity of a communication problem KWf. The KW relation KWf is defined as follows:

Alice gets x ∈ f −1(0). Bob gets y ∈ f −1(1). Clearly, x = y, so ∃i s.t. xi = yi. Want to find such i. Want to talk as little as possible.

Only deterministic protocols! This talk: Assume C(KWf) = log L(f).

Irit Dinur, Or Meir Toward the KRW conjecture

slide-30
SLIDE 30

Karchmer-Wigderson Relations [KW90]

Relate L(f) to complexity of a communication problem KWf. The KW relation KWf is defined as follows:

Alice gets x ∈ f −1(0). Bob gets y ∈ f −1(1). Clearly, x = y, so ∃i s.t. xi = yi. Want to find such i. Want to talk as little as possible.

Only deterministic protocols! This talk: Assume C(KWf) = log L(f). KRW conjecture: C(KWf⋄g) ≈ C(KWf) + C(KWg).

Irit Dinur, Or Meir Toward the KRW conjecture

slide-31
SLIDE 31

KRW and KW

Can we use KW relations to attack the KRW conjecture? How does KWf⋄g look like? Recall: f ⋄ g maps {0, 1}m×n to {0, 1}.

Alice Bob m m n n

X Y

Irit Dinur, Or Meir Toward the KRW conjecture

slide-32
SLIDE 32

KRW and KW

Can we use KW relations to attack the KRW conjecture? How does KWf⋄g look like? Recall: f ⋄ g maps {0, 1}m×n to {0, 1}.

Alice Bob m m n n g g g .......... g g g ..........

a b

X Y

Irit Dinur, Or Meir Toward the KRW conjecture

slide-33
SLIDE 33

KRW and KW

Can we use KW relations to attack the KRW conjecture? How does KWf⋄g look like? Recall: f ⋄ g maps {0, 1}m×n to {0, 1}.

Alice f Bob f 1 m m n n g g g .......... g g g ..........

a b

X Y

Irit Dinur, Or Meir Toward the KRW conjecture

slide-34
SLIDE 34

KRW and KW

Can we use KW relations to attack the KRW conjecture? How does KWf⋄g look like? Recall: f ⋄ g maps {0, 1}m×n to {0, 1}.

Alice f Bob f 1 m m n n g g g .......... g g g ..........

a b

X Y

Irit Dinur, Or Meir Toward the KRW conjecture

slide-35
SLIDE 35

KRW and KW

Can we use KW relations to attack the KRW conjecture? How does KWf⋄g look like? Recall: f ⋄ g maps {0, 1}m×n to {0, 1}.

Alice f Bob f 1 m m n n g g g .......... g g g ..........

a b

X Y

Irit Dinur, Or Meir Toward the KRW conjecture

slide-36
SLIDE 36

KRW and KW

Can we use KW relations to attack the KRW conjecture? How does KWf⋄g look like? Recall: f ⋄ g maps {0, 1}m×n to {0, 1}.

Alice f Bob f 1 m m n n g g g .......... g g g ..........

a b

X Y

Irit Dinur, Or Meir Toward the KRW conjecture

slide-37
SLIDE 37

KRW and KW

Can we use KW relations to attack the KRW conjecture? How does KWf⋄g look like? Recall: f ⋄ g maps {0, 1}m×n to {0, 1}.

Alice f Bob f 1 m m n n g g g .......... g g g ..........

a b

X Y KRW conjecture: C(KWf⋄g) ≈ C(KWf) + C(KWg) The obvious protocol is essentially optimal.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-38
SLIDE 38

Outline

1

Introduction

2

Background

3

Proof Strategy

4

New tools

Irit Dinur, Or Meir Toward the KRW conjecture

slide-39
SLIDE 39

Proof strategy

We wish to show that the obvious protocol is optimal.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-40
SLIDE 40

Proof strategy

We wish to show that the obvious protocol is optimal. Fix a protocol Π for KWf⋄g. We will show that Π must behave roughly like the obvious protocol.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-41
SLIDE 41

Proof strategy

We wish to show that the obvious protocol is optimal. Fix a protocol Π for KWf⋄g. We will show that Π must behave roughly like the obvious protocol. Partition each transcript π of Π to two parts: π = π1 ◦ π2. π1 and π2 correspond to the two stages of the obvious protocol.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-42
SLIDE 42

Proof strategy

Show:

Irit Dinur, Or Meir Toward the KRW conjecture

slide-43
SLIDE 43

Proof strategy

Show:

1

∀ π1 that has not solved KWf, ∃ π2 of length C(KWg).

Irit Dinur, Or Meir Toward the KRW conjecture

slide-44
SLIDE 44

Proof strategy

Show:

1

∀ π1 that has not solved KWf, ∃ π2 of length C(KWg).

2

∃ π1 of length almost C(KWf) that has not solved KWf.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-45
SLIDE 45

Proof strategy

Show:

1

∀ π1 that has not solved KWf, ∃ π2 of length C(KWg).

2

∃ π1 of length almost C(KWf) that has not solved KWf.

Together, the two items imply a lower bound of ≈ C(KWf) + C(KWg).

Irit Dinur, Or Meir Toward the KRW conjecture

slide-46
SLIDE 46

Proof strategy

Show:

1

∀ π1 that has not solved KWf, ∃ π2 of length C(KWg).

2

∃ π1 of length almost C(KWf) that has not solved KWf.

Together, the two items imply a lower bound of ≈ C(KWf) + C(KWg). The first item gives our structural result. This will be the focus of the rest of this talk.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-47
SLIDE 47

Structural Result - Basic Idea

Alice f Bob f 1 m m n n g g g .......... g g g ..........

a b

X Y

In the 2nd stage, they need to solve KWg on some Xi, Yi.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-48
SLIDE 48

Structural Result - Basic Idea

Alice f Bob f 1 m m n n g g g .......... g g g ..........

a b

X Y

In the 2nd stage, they need to solve KWg on some Xi, Yi. In principle, this requires C(KWg) bits.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-49
SLIDE 49

Structural Result - Basic Idea

Alice f Bob f 1 m m n n g g g .......... g g g ..........

a b

X Y

In the 2nd stage, they need to solve KWg on some Xi, Yi. In principle, this requires C(KWg) bits. However, they already communicated almost C(KWf) bits in the 1st stage.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-50
SLIDE 50

Structural Result - Basic Idea

Alice f Bob f 1 m m n n g g g .......... g g g ..........

a b

X Y

In the 2nd stage, they need to solve KWg on some Xi, Yi. In principle, this requires C(KWg) bits. However, they already communicated almost C(KWf) bits in the 1st stage. Need to show: does not make the 2nd stage easier.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-51
SLIDE 51

Structural Result - Basic Idea

The players communicated less than C(KWf) ≤ m bits in the first stage.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-52
SLIDE 52

Structural Result - Basic Idea

The players communicated less than C(KWf) ≤ m bits in the first stage. On average: less than 1 bit per row.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-53
SLIDE 53

Structural Result - Basic Idea

The players communicated less than C(KWf) ≤ m bits in the first stage. On average: less than 1 bit per row. Thus, on the typical row, they did not make much progress.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-54
SLIDE 54

Structural Result - Basic Idea

The players communicated less than C(KWf) ≤ m bits in the first stage. On average: less than 1 bit per row. Thus, on the typical row, they did not make much progress. If they solve a typical row in the 2nd stage, they still need to send ≈ C(KWg) bits.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-55
SLIDE 55

Structural Result - Basic Idea

There might be non-typical rows on which Alice and Bob communicated a lot.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-56
SLIDE 56

Structural Result - Basic Idea

There might be non-typical rows on which Alice and Bob communicated a lot. Might be easy to solve KWg on those rows.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-57
SLIDE 57

Structural Result - Basic Idea

There might be non-typical rows on which Alice and Bob communicated a lot. Might be easy to solve KWg on those rows. Ensure: Players do not solve KWg on non-typical rows.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-58
SLIDE 58

Structural Result - Basic Idea

There might be non-typical rows on which Alice and Bob communicated a lot. Might be easy to solve KWg on those rows. Ensure: Players do not solve KWg on non-typical rows. Force Xi = Yi on non-typical rows.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-59
SLIDE 59

Structural Result - Basic Idea

Forcing Xi = Yi is hardest part of the proof.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-60
SLIDE 60

Structural Result - Basic Idea

Forcing Xi = Yi is hardest part of the proof. In particular, need to force g(Xi) = g(Yi).

Irit Dinur, Or Meir Toward the KRW conjecture

slide-61
SLIDE 61

Structural Result - Basic Idea

Forcing Xi = Yi is hardest part of the proof. In particular, need to force g(Xi) = g(Yi). Issue: What if the players already found that g(Xi) = g(Yi) in the first stage?

Irit Dinur, Or Meir Toward the KRW conjecture

slide-62
SLIDE 62

Structural Result - Basic Idea

Forcing Xi = Yi is hardest part of the proof. In particular, need to force g(Xi) = g(Yi). Issue: What if the players already found that g(Xi) = g(Yi) in the first stage? This cannot happen, since we know they did not solve KWf in the first stage.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-63
SLIDE 63

Credit

This proof strategy was developed by Edmonds, Impagliazzo, Rudich and Sgall [EIRS91]. Used it to prove the KRW conjecture for the composition of universal relations U ⋄ U.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-64
SLIDE 64

Credit

This proof strategy was developed by Edmonds, Impagliazzo, Rudich and Sgall [EIRS91]. Used it to prove the KRW conjecture for the composition of universal relations U ⋄ U. Our novelty: extending this strategy to f ⋄ g, where g = ⊕.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-65
SLIDE 65

Credit

This proof strategy was developed by Edmonds, Impagliazzo, Rudich and Sgall [EIRS91]. Used it to prove the KRW conjecture for the composition of universal relations U ⋄ U. Our novelty: extending this strategy to f ⋄ g, where g = ⊕. This setting is considerably more challenging.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-66
SLIDE 66

Main Difficulty

Universal relations are “information-theoretic objects”.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-67
SLIDE 67

Main Difficulty

Universal relations are “information-theoretic objects”. Direct correspondence between information and complexity:

If Alice sends a message that has t bits of information, the complexity of the problem decreases by exactly t bits.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-68
SLIDE 68

Main Difficulty

Universal relations are “information-theoretic objects”. Direct correspondence between information and complexity:

If Alice sends a message that has t bits of information, the complexity of the problem decreases by exactly t bits.

KW relations are “computational objects”.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-69
SLIDE 69

Main Difficulty

Universal relations are “information-theoretic objects”. Direct correspondence between information and complexity:

If Alice sends a message that has t bits of information, the complexity of the problem decreases by exactly t bits.

KW relations are “computational objects”. There are examples KWf in which

Alice sends little information, and decreases the complexity by a lot.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-70
SLIDE 70

Main Difficulty

We deal with this difficulty separately for KWf and KWg.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-71
SLIDE 71

Main Difficulty

We deal with this difficulty separately for KWf and KWg. For KWf, we prove a general “fortification lemma”, which creates a correspondence between information and complexity.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-72
SLIDE 72

Main Difficulty

We deal with this difficulty separately for KWf and KWg. For KWf, we prove a general “fortification lemma”, which creates a correspondence between information and complexity. For KWg, we use the fact that the lower bound for parity has an information-theoretic proof.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-73
SLIDE 73

Main Difficulty

We deal with this difficulty separately for KWf and KWg. For KWf, we prove a general “fortification lemma”, which creates a correspondence between information and complexity. For KWg, we use the fact that the lower bound for parity has an information-theoretic proof. Therefore, can analyze directly how information and complexity interact for parity.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-74
SLIDE 74

Another difficulty: the “randomized barrier”

Every KW relation has a randomized protocol of complexity 2 log n. Hence, cannot use techniques that work against randomized protocols.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-75
SLIDE 75

Another difficulty: the “randomized barrier”

Every KW relation has a randomized protocol of complexity 2 log n. Hence, cannot use techniques that work against randomized protocols. KW relations have no hard distributions (with complexity > 2 log n).

Irit Dinur, Or Meir Toward the KRW conjecture

slide-76
SLIDE 76

Another difficulty: the “randomized barrier”

Every KW relation has a randomized protocol of complexity 2 log n. Hence, cannot use techniques that work against randomized protocols. KW relations have no hard distributions (with complexity > 2 log n). Hard to use information-theoretic techniques. Also, most rectangle bounds don’t work [KKN95].

Irit Dinur, Or Meir Toward the KRW conjecture

slide-77
SLIDE 77

Another difficulty: the “randomized barrier”

Every KW relation has a randomized protocol of complexity 2 log n. Hence, cannot use techniques that work against randomized protocols. KW relations have no hard distributions (with complexity > 2 log n). Hard to use information-theoretic techniques. Also, most rectangle bounds don’t work [KKN95]. Our work is the first to bypass this barrier.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-78
SLIDE 78

Outline

1

Introduction

2

Background

3

Proof Strategy

4

New tools

Irit Dinur, Or Meir Toward the KRW conjecture

slide-79
SLIDE 79

Fortification

“Dream version”: If Alice sends t bits of information, then complexity decreases by t.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-80
SLIDE 80

Fortification

“Dream version”: If Alice sends t bits of information, then complexity decreases by t. Counter-example:

Suppose exactly half of Alice’s inputs start with 0. Suppose all of Bob’s inputs start with 0.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-81
SLIDE 81

Fortification

“Dream version”: If Alice sends t bits of information, then complexity decreases by t. Counter-example:

Suppose exactly half of Alice’s inputs start with 0. Suppose all of Bob’s inputs start with 0.

If Alice says “my first bit is 1” then

she sent 1 bit of information, but complexity decreased to 0.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-82
SLIDE 82

Fortification

“Dream version”: If Alice sends t bits of information, then complexity decreases by t. Counter-example:

Suppose exactly half of Alice’s inputs start with 0. Suppose all of Bob’s inputs start with 0.

If Alice says “my first bit is 1” then

she sent 1 bit of information, but complexity decreased to 0.

Can fix it: discard in advance all Alice’s inputs that start with 1.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-83
SLIDE 83

Fortification

Recall: we model states of the protocol by rectangles A × B.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-84
SLIDE 84

Fortification

Recall: we model states of the protocol by rectangles A × B. We say a rectangle A × B is “fortified” if

whenever Alice’s sends t bits of information, the complexity decreases by at most t + O(log n).

Irit Dinur, Or Meir Toward the KRW conjecture

slide-85
SLIDE 85

Fortification

Recall: we model states of the protocol by rectangles A × B. We say a rectangle A × B is “fortified” if

whenever Alice’s sends t bits of information, the complexity decreases by at most t + O(log n).

Fortification lemma Every rectangle A × B has a fortified subrectangle whose complexity is close to that of A × B.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-86
SLIDE 86

Averaging Argument for Information

“On the typical row, they did not make much progress.”

Irit Dinur, Or Meir Toward the KRW conjecture

slide-87
SLIDE 87

Averaging Argument for Information

“On the typical row, they did not make much progress.” This requires an “averaging argument” for information.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-88
SLIDE 88

Averaging Argument for Information

“On the typical row, they did not make much progress.” This requires an “averaging argument” for information. How do we model information?

Irit Dinur, Or Meir Toward the KRW conjecture

slide-89
SLIDE 89

Averaging Argument for Information

“On the typical row, they did not make much progress.” This requires an “averaging argument” for information. How do we model information? For entropy: follows immediately from sub-additivity.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-90
SLIDE 90

Averaging Argument for Information

“On the typical row, they did not make much progress.” This requires an “averaging argument” for information. How do we model information? For entropy: follows immediately from sub-additivity. [EIRS91]: proved it for “predictability”.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-91
SLIDE 91

Averaging Argument for Information

“On the typical row, they did not make much progress.” This requires an “averaging argument” for information. How do we model information? For entropy: follows immediately from sub-additivity. [EIRS91]: proved it for “predictability”. This work: we prove it for min-entropy.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-92
SLIDE 92

Summary

Result: We (re-)prove the KRW conjecture for the special case f ⋄ g where g is parity.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-93
SLIDE 93

Summary

Result: We (re-)prove the KRW conjecture for the special case f ⋄ g where g is parity. Corollary: an alternative proof for the cubic lower bound for formulas.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-94
SLIDE 94

Summary

Result: We (re-)prove the KRW conjecture for the special case f ⋄ g where g is parity. Corollary: an alternative proof for the cubic lower bound for formulas. New tools: fortification lemma, averaging argument for min-entropy.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-95
SLIDE 95

Summary

Result: We (re-)prove the KRW conjecture for the special case f ⋄ g where g is parity. Corollary: an alternative proof for the cubic lower bound for formulas. New tools: fortification lemma, averaging argument for min-entropy. Future direction: replace with a general function?

Irit Dinur, Or Meir Toward the KRW conjecture

slide-96
SLIDE 96

The 1-out-of-m problem [BBKW14]

Alice and Bob are given m distinct instances of KWg. They need to solve one of them. Prove: not easier than solving a single instance.

Irit Dinur, Or Meir Toward the KRW conjecture

slide-97
SLIDE 97

Thank you!