Reed-Muller testing and approximating small set expansion & - - PowerPoint PPT Presentation

reed muller testing and approximating small set expansion
SMART_READER_LITE
LIVE PREVIEW

Reed-Muller testing and approximating small set expansion & - - PowerPoint PPT Presentation

Reed-Muller testing and approximating small set expansion & hypergraph coloring Venkatesan Guruswami Carnegie Mellon University (Spring14 @ Microsoft Research New England) 2014 Bertinoro Workshop on Sublinear Algorithms May 27, 2014


slide-1
SLIDE 1

Reed-Muller testing and approximating small set expansion & hypergraph coloring

Venkatesan Guruswami

Carnegie Mellon University

(Spring’14 @ Microsoft Research New England)

2014 Bertinoro Workshop on Sublinear Algorithms May 27, 2014

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 1 / 22

slide-2
SLIDE 2

Linear codes and testing

Binary linear code C is a subspace of Fn

2

Dual space C ⊥ = {y ∈ Fn

2 | y, c = 0 ∀c ∈ C}.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 2 / 22

slide-3
SLIDE 3

Linear codes and testing

Binary linear code C is a subspace of Fn

2

Dual space C ⊥ = {y ∈ Fn

2 | y, c = 0 ∀c ∈ C}.

Property test for code “x ∈ C”

Pick y randomly from a subset T ⊆ C ⊥ and check x, y = 0. T = C ⊥ rejects x / ∈ C with prob. 1/2.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 2 / 22

slide-4
SLIDE 4

Linear codes and testing

Binary linear code C is a subspace of Fn

2

Dual space C ⊥ = {y ∈ Fn

2 | y, c = 0 ∀c ∈ C}.

Property test for code “x ∈ C”

Pick y randomly from a subset T ⊆ C ⊥ and check x, y = 0. T = C ⊥ rejects x / ∈ C with prob. 1/2. Soundness error of test on input x :=

  • Ey∈T
  • (−1)x,y
  • Venkatesan Guruswami (CMU)

RM testing and approximability May 2014 2 / 22

slide-5
SLIDE 5

Linear codes and testing

Binary linear code C is a subspace of Fn

2

Dual space C ⊥ = {y ∈ Fn

2 | y, c = 0 ∀c ∈ C}.

Property test for code “x ∈ C”

Pick y randomly from a subset T ⊆ C ⊥ and check x, y = 0. T = C ⊥ rejects x / ∈ C with prob. 1/2. Soundness error of test on input x :=

  • Ey∈T
  • (−1)x,y
  • Focus on restricted/structured set of dual codewords for test:

q query tests: T ⊆ C ⊥

q (low-weight dual codewords)

Hope to have low soundness error when x is far from C.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 2 / 22

slide-6
SLIDE 6

Locally testable codes (LTC)

Goal: Construct codes C of good rate with a low-query test Always accept codewords, and reject strings far from the code with good prob. Most work has focused on q = O(1) case. Best known constant-query LTC has dimension o(n) (n/poly(log n))

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 3 / 22

slide-7
SLIDE 7

Locally testable codes (LTC)

Goal: Construct codes C of good rate with a low-query test Always accept codewords, and reject strings far from the code with good prob. Most work has focused on q = O(1) case. Best known constant-query LTC has dimension o(n) (n/poly(log n)) Recently, due to connections to approximability, there has been interest in the regime: q ≈ εn (“small linear locality”), and codes of large (n − o(n)) dimension.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 3 / 22

slide-8
SLIDE 8

Binary Reed-Muller codes

Let P(m, u) be the F2-linear space of all multilinear polynomials in X1, X2, . . . , Xm of degree u (coefficients in F2)

Reed-Muller code

RM(m, u) = {f (a)a∈Fm

2 | f ∈ P(m, u)}.

Code length = 2m. Dimension = u

j=0

m

j

  • (number of monomials of degree u)

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 4 / 22

slide-9
SLIDE 9

Binary Reed-Muller codes

Let P(m, u) be the F2-linear space of all multilinear polynomials in X1, X2, . . . , Xm of degree u (coefficients in F2)

Reed-Muller code

RM(m, u) = {f (a)a∈Fm

2 | f ∈ P(m, u)}.

Code length = 2m. Dimension = u

j=0

m

j

  • (number of monomials of degree u)

Distance = 2m−u (A min. wt. codeword: f (X) = X1X2 · · · Xu)

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 4 / 22

slide-10
SLIDE 10

Large rate Reed-Muller codes

Our focus: Large degree u = m − r − 1 (think r fixed, m → ∞) Code distance = 2r+1. (Poly f (X) = X1X2 · · · Xm−r−1) Dual space is RM(m, r)

f ∈ P(m, m − r − 1) and g ∈ P(m, r) = ⇒ f · g ∈ P(m, m − 1) = ⇒

x f (x)g(x) = 0

Dual codewords of minimum weight (= 2m−r): L1L2 . . . Lr, product of r degree 1 polys (affine forms).

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 5 / 22

slide-11
SLIDE 11

Reed-Muller testing

Canonical test for proximity of f : Fm

2 → F2 to deg. m − r − 1 polys:

1

Pick linear independent affine forms L1, L2, . . . , Lr u.a.r, and set h = r

j=1 Lj (random min. wt. dual codeword)

2

Check f , h =

x f (x)h(x) = 0 (≡ deg(f · h) < m)

# queries = 2m−r = εn for ε = 2−r.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 6 / 22

slide-12
SLIDE 12

Reed-Muller testing

Canonical test for proximity of f : Fm

2 → F2 to deg. m − r − 1 polys:

1

Pick linear independent affine forms L1, L2, . . . , Lr u.a.r, and set h = r

j=1 Lj (random min. wt. dual codeword)

2

Check f , h =

x f (x)h(x) = 0 (≡ deg(f · h) < m)

# queries = 2m−r = εn for ε = 2−r.

Theorem ([Bhattacharyya, Kopparty, Schoenebeck, Sudan, Zuckerman’10])

If f is 2r-far from P(m, m − r − 1), then error of above test is bounded away from 1; i.e., for some absolute constant ρ < 1

  • E[(−1)f ,L1L2···Lr]
  • ρ

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 6 / 22

slide-13
SLIDE 13

A beautiful connection: LTCs and SSEs

[Barak, Gopalan, H˚ astad, Meka, Raghavendra, Steurer’12] made a beautiful

connection between locally testable codes (LTCs) and small set expanders (SSEs). Instantiating with Reed-Muller codes, they constructed SSEs with currently largest known count of bad eigenvalues.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 7 / 22

slide-14
SLIDE 14

Small set expansion problem

SSE(µ, ε) problem

Given graph G = (V , E) on n vertices, distinguish between: Yes instance: ∃ small non-expanding set, i.e., ∃S ⊂ V , |S| = µn, EdgeExp(S) ε No instance: All small sets expand, ∀S, |S| = µn, EdgeExp(S) 1/2.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 8 / 22

slide-15
SLIDE 15

Small set expansion problem

SSE(µ, ε) problem

Given graph G = (V , E) on n vertices, distinguish between: Yes instance: ∃ small non-expanding set, i.e., ∃S ⊂ V , |S| = µn, EdgeExp(S) ε No instance: All small sets expand, ∀S, |S| = µn, EdgeExp(S) 1/2.

SSE intractability hypothesis [Raghavendra, Steurer’10]

∀ε > 0, ∃µ such that SSE(µ, ε) is hard.

(Implies many other intractability results, including Unique Games conjecture.)

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 8 / 22

slide-16
SLIDE 16

A spectral necessity

A subset S with EdgeExp(S) ε can be “found” in the eigenspace

  • f eigenvalues 1 − ε (of graph’s random walk matrix).

[Arora, Barak, Steurer’10]: this eigenspace has dimension nε for

No instances (when the graph is a small set expander) ⇒ exp(nε) time algo for SSE problem

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 9 / 22

slide-17
SLIDE 17

A spectral necessity

A subset S with EdgeExp(S) ε can be “found” in the eigenspace

  • f eigenvalues 1 − ε (of graph’s random walk matrix).

[Arora, Barak, Steurer’10]: this eigenspace has dimension nε for

No instances (when the graph is a small set expander) ⇒ exp(nε) time algo for SSE problem

Necessary requirement for SSE intractability hypothesis

Existence of small set expanders (SSEs) with nΩε(1) “bad” eigenvalues 1 − ε.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 9 / 22

slide-18
SLIDE 18

SSEs with many bad eigenvalues [BGHMRS’12]

Noisy hypercube

Vertex set V = {0, 1}t. Edge x ∼ y if HamDist(x, y) = εt. Has t = log |V | eigenvalues ≈ 1 − ε.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 10 / 22

slide-19
SLIDE 19

SSEs with many bad eigenvalues [BGHMRS’12]

Noisy hypercube

Vertex set V = {0, 1}t. Edge x ∼ y if HamDist(x, y) = εt. Has t = log |V | eigenvalues ≈ 1 − ε.

Derandomization via Reed-Muller code

Take subgraph induced by V ′ = RM(m, r) (t = 2m, ε = 2−r). Vertices P(m, r), degree r polynomials Edges f ∼ g if f − g = L1L2 · · · Lr. Easy: Graph retains Ω(t) eigenvalues ≈ 1 − ε.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 10 / 22

slide-20
SLIDE 20

SSEs with many bad eigenvalues [BGHMRS’12]

Noisy hypercube

Vertex set V = {0, 1}t. Edge x ∼ y if HamDist(x, y) = εt. Has t = log |V | eigenvalues ≈ 1 − ε.

Derandomization via Reed-Muller code

Take subgraph induced by V ′ = RM(m, r) (t = 2m, ε = 2−r). Vertices P(m, r), degree r polynomials Edges f ∼ g if f − g = L1L2 · · · Lr. Easy: Graph retains Ω(t) eigenvalues ≈ 1 − ε. But now |V ′| ≈ 2mr = 2(log t)r, so we have 2(log |V ′|)Ωε(1) bad eignevalues.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 10 / 22

slide-21
SLIDE 21

SSE property of Reed-Muller graph

Fourier analysis over P(m, r)

Express function A : P(m, r) → R as A(f ) =

β

A(β)(−1)β,f . “frequencies” β range over cosets of P(m, m − r − 1) (dual group of P(m, r)). Weight of frequency β = Hamming dist. of β to P(m, m −r −1)

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 11 / 22

slide-22
SLIDE 22

SSE property of Reed-Muller graph

Fourier analysis over P(m, r)

Express function A : P(m, r) → R as A(f ) =

β

A(β)(−1)β,f . “frequencies” β range over cosets of P(m, m − r − 1) (dual group of P(m, r)). Weight of frequency β = Hamming dist. of β to P(m, m −r −1)

SSE proof has two ingredients

Take A = indicator of a small set

1

A has very little Fourier mass on low frequencies (Hypercontractivity of low-degree polynomials)

2

Contribution of high frequency A(β) killed by edges of graph (testing of RM(m, m − r − 1)), leading to expansion.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 11 / 22

slide-23
SLIDE 23

LTCs of constant absolute distance

Consider C ⊆ Fn

2 of minimum distance d.

Think d fixed, and n → ∞. Largest possible dimension (sphere packing bound): ≈ n − d

2 log n.

Achieved by BCH codes! However, BCH code is not testable even with 0.49n queries.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 12 / 22

slide-24
SLIDE 24

LTCs of constant absolute distance

Consider C ⊆ Fn

2 of minimum distance d.

Think d fixed, and n → ∞. Largest possible dimension (sphere packing bound): ≈ n − d

2 log n.

Achieved by BCH codes! However, BCH code is not testable even with 0.49n queries. Reed-Muller code RM(m, u) of length n = 2m and u ≈ m − log d Dimension ≈ n − (log n)log d, Testable with 2n/d queries (rejecting d/3-far strings with Ω(1) prob.)

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 12 / 22

slide-25
SLIDE 25

Price of local testability?

Where in the spectrum between Reed-Muller and BCH does the best dimension of distance d code testable with O(n/d) queries lie? Dimension n − O(d log n) vs. n − (log n)log d

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 13 / 22

slide-26
SLIDE 26

Price of local testability?

Where in the spectrum between Reed-Muller and BCH does the best dimension of distance d code testable with O(n/d) queries lie? Dimension n − O(d log n) vs. n − (log n)log d

[Guo, Kopparty,Sudan’13] Lifted codes, with dimension n −

  • log n

log d

log d slightly improving Reed-Muller codes.

[G.,Sudan,Velingker,Wang’14] For a class of affine-invariant codes

containing Reed-Muller, dimension n −

  • log n

log2 d

log d .

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 13 / 22

slide-27
SLIDE 27

RM testing application II: Hypergraph Coloring

Best known algorithms to color 3-colorable graphs use nΩ(1) colors Dream result: Matching hardness?

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 14 / 22

slide-28
SLIDE 28

RM testing application II: Hypergraph Coloring

Best known algorithms to color 3-colorable graphs use nΩ(1) colors Dream result: Matching hardness? Less dreamy: Hardness of nΩ(1)-coloring 2-colorable hypergraphs?

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 14 / 22

slide-29
SLIDE 29

RM testing application II: Hypergraph Coloring

Best known algorithms to color 3-colorable graphs use nΩ(1) colors Dream result: Matching hardness? Less dreamy: Hardness of nΩ(1)-coloring 2-colorable hypergraphs? Recent result based a “structured” Reed-Muller testing result:

Theorem ([Dinur, G.’13], [G., Harsha, H˚

astad, Srinivasan, Varma’14])

Coloring a 2-colorable 8-uniform hypergraph with exp(2

√log log n)

colors is quasi NP-hard.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 14 / 22

slide-30
SLIDE 30

RM testing application II: Hypergraph Coloring

Best known algorithms to color 3-colorable graphs use nΩ(1) colors Dream result: Matching hardness? Less dreamy: Hardness of nΩ(1)-coloring 2-colorable hypergraphs? Recent result based a “structured” Reed-Muller testing result:

Theorem ([Dinur, G.’13], [G., Harsha, H˚

astad, Srinivasan, Varma’14])

Coloring a 2-colorable 8-uniform hypergraph with exp(2

√log log n)

colors is quasi NP-hard. Previous hardness only ruled out (log n)O(1) coloring.

Very recently, [Khot, Saket’14] improved bound to exp((log n)Ω(1)) via different use of the [Dinur, G.’13] RM testing result.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 14 / 22

slide-31
SLIDE 31

Won’t be able to describe the underlying PCP in any detail, but will try to give a glimpse of where Reed-Muller tesing fits in.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 15 / 22

slide-32
SLIDE 32

The Long Code

PCPs encode assignments {0, 1}m to enable efficient testing. The most influential code underlying almost all strong PCP results: Long Code [Bellare, Goldreich, Sudan’95]

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 16 / 22

slide-33
SLIDE 33

The Long Code

PCPs encode assignments {0, 1}m to enable efficient testing. The most influential code underlying almost all strong PCP results: Long Code [Bellare, Goldreich, Sudan’95]

Definition (Long Code encoding a ∈ {0, 1}m)

LONG(a) := f (a)f :{0,1}m→{0,1} . Gives value of every Boolean function on a: the most redundant encoding.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 16 / 22

slide-34
SLIDE 34

The Long Code

PCPs encode assignments {0, 1}m to enable efficient testing. The most influential code underlying almost all strong PCP results: Long Code [Bellare, Goldreich, Sudan’95]

Definition (Long Code encoding a ∈ {0, 1}m)

LONG(a) := f (a)f :{0,1}m→{0,1} . Gives value of every Boolean function on a: the most redundant encoding. The improvements in hypergraph coloring

(and also earlier integrality gaps in [BGHMRS’12], [Kane-Meka’13])

due to a “shorter” Reed-Muller based substitute of the long code.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 16 / 22

slide-35
SLIDE 35

The low-degree long code

Definition (Degree-r long code)

The degree-r long code encoding of a ∈ {0, 1}m is f (a)f ∈P(m,r) . Puncturing of long code to locations indexed by degree r fns. ⇐ ⇒ derandomization of hypercube to Reed-Muller codewords. Encoding length ≈ 2mr instead of 22m for the long code. For r ≈ log m, almost exponential savings.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 17 / 22

slide-36
SLIDE 36

Hypergraph gadget on low-degree long code

Underlying hypergh coloring hardness is a “low-degree long code test” Query patterns give hypergraph on vertex set P(m, r)1 such that:

1

(Completeness) If A : P(m, r) → {0, 1} is a codeword of the degree-r long code, i.e., ∃a ∈ Fm

2 such that ∀f , A(f ) = f (a),

then A is a 2-coloring without any monochromatic hyperedge.

1degree r polynomials over F2 in m variables

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 18 / 22

slide-37
SLIDE 37

Hypergraph gadget on low-degree long code

Underlying hypergh coloring hardness is a “low-degree long code test” Query patterns give hypergraph on vertex set P(m, r)1 such that:

1

(Completeness) If A : P(m, r) → {0, 1} is a codeword of the degree-r long code, i.e., ∃a ∈ Fm

2 such that ∀f , A(f ) = f (a),

then A is a 2-coloring without any monochromatic hyperedge.

2

(Soundness) If I : P(m, r) → {0, 1} is the indicator function of an independent set of measure µ, then ∃ a “sizeable” Fourier coefficient | I(β)| for some β of “low” weight (= distance to

P(m, m − r − 1))

1degree r polynomials over F2 in m variables

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 18 / 22

slide-38
SLIDE 38

8-uniform hypergraph gadget

Vertex set = P(m, r). Hyperedges on 8-tuples: e1 e1 + f1 e2 e2 + f1 + g · h + 1 e3 e3 + f2 e4 e4 + f2 + g · h′ + 1 . ∀ei, fi ∈ P(m, r), g, h, h′ ∈ P(m, r/2). Completeness: Ensured by (g · h)(a) = 0 or (g · h′)(a) = 0 for every a ∈ Fm

2 .

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 19 / 22

slide-39
SLIDE 39

8-uniform hypergraph gadget

Vertex set = P(m, r). Hyperedges on 8-tuples: e1 e1 + f1 e2 e2 + f1 + g · h + 1 e3 e3 + f2 e4 e4 + f2 + g · h′ + 1 . ∀ei, fi ∈ P(m, r), g, h, h′ ∈ P(m, r/2). Completeness: Ensured by (g · h)(a) = 0 or (g · h′)(a) = 0 for every a ∈ Fm

2 .

Soundness: Orthogonality to g · h is a good Reed-Muller test that kills high frequencies.

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 19 / 22

slide-40
SLIDE 40

Structured Reed-Muller testing

Recap: To test proximity to P(m, m − r − 1), check orthogonality to some degree r polys (the dual space).

Theorem (Dinur, G.’13)

If β : Fm

2 → F2 is 2r-far from P(m, m − r − 1), then

Eg,h

  • (−1)β,g·h

2−2Ω(r) , where g, h ∈R P(m, r/2).

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 20 / 22

slide-41
SLIDE 41

Structured Reed-Muller testing

Recap: To test proximity to P(m, m − r − 1), check orthogonality to some degree r polys (the dual space).

Theorem (Dinur, G.’13)

If β : Fm

2 → F2 is 2r-far from P(m, m − r − 1), then

Eg,h

  • (−1)β,g·h

2−2Ω(r) , where g, h ∈R P(m, r/2). Compare with [BKSSZ]: Test function L1L2 · · · Lr, constant soundness error (and 2m/2r queries) Here, test function g · h, soundness error doubly exponentially small in r (and typically 2m/4 queries)

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 20 / 22

slide-42
SLIDE 42

Proof idea of RM testing result

Need to understand when β, gh = 0 ⇐ ⇒ βg, h = 0, given β is far from P(m, m − r − 1).

Eh[(−1)βg,h] =

  • 1

if deg(βg) m − r/2 − 1

  • therwise

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 21 / 22

slide-43
SLIDE 43

Proof idea of RM testing result

Need to understand when β, gh = 0 ⇐ ⇒ βg, h = 0, given β is far from P(m, m − r − 1).

Eh[(−1)βg,h] =

  • 1

if deg(βg) m − r/2 − 1

  • therwise

1

For fixed β, {g ∈ P(m, r/2) | deg(βg) m − r/2 − 1} is a subspace of P(m, r/2)

⇒ Must prove co-dimension 2Ω(r) (for β far from P(m, m − r − 1))

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 21 / 22

slide-44
SLIDE 44

Proof idea of RM testing result

Need to understand when β, gh = 0 ⇐ ⇒ βg, h = 0, given β is far from P(m, m − r − 1).

Eh[(−1)βg,h] =

  • 1

if deg(βg) m − r/2 − 1

  • therwise

1

For fixed β, {g ∈ P(m, r/2) | deg(βg) m − r/2 − 1} is a subspace of P(m, r/2)

⇒ Must prove co-dimension 2Ω(r) (for β far from P(m, m − r − 1))

2

[BKSSZ] ⇒ If β is D-far from P(m, m − r − 1), then ∃ a linear

form L s.t. β|L=0 & β|L=1 are both D

3 -far from P(m − 1, m − r − 1).

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 21 / 22

slide-45
SLIDE 45

Proof idea of RM testing result

Need to understand when β, gh = 0 ⇐ ⇒ βg, h = 0, given β is far from P(m, m − r − 1).

Eh[(−1)βg,h] =

  • 1

if deg(βg) m − r/2 − 1

  • therwise

1

For fixed β, {g ∈ P(m, r/2) | deg(βg) m − r/2 − 1} is a subspace of P(m, r/2)

⇒ Must prove co-dimension 2Ω(r) (for β far from P(m, m − r − 1))

2

[BKSSZ] ⇒ If β is D-far from P(m, m − r − 1), then ∃ a linear

form L s.t. β|L=0 & β|L=1 are both D

3 -far from P(m − 1, m − r − 1).

3

Use 2. to lower bound co-dimension by sum of two similar co-dimensions (recursively for Ω(r) inductive steps)

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 21 / 22

slide-46
SLIDE 46

Summary

Ability to test high rate Reed-Muller codes is the basis for: Quantitative improvements via the low-degree long code Applications to approximability: SSE with many eigenvalues, improved

integrality gaps for sparsest cut, hardness of hypergraph coloring, size-efficient PCPs.

More applications?

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 22 / 22

slide-47
SLIDE 47

Summary

Ability to test high rate Reed-Muller codes is the basis for: Quantitative improvements via the low-degree long code Applications to approximability: SSE with many eigenvalues, improved

integrality gaps for sparsest cut, hardness of hypergraph coloring, size-efficient PCPs.

More applications? Even better testable codes than Reed-Muller codes? Limits of testability in the “small linear locality” (εn queries) regime? Is BCH or RM closer to the largest possible dimension?

Venkatesan Guruswami (CMU) RM testing and approximability May 2014 22 / 22