Oshman-slides

ON THE COMMUNICATION COMPLEXITY
OF PROPERTY TESTING IN GRAPHS
Joint work with Orr Fischer and
Shay Gershtein
THE β€œCONGEST” NETWORK MODEL
Network graph: 𝐺 = 𝑉, 𝐸
ο‚­Input to node 𝑣 ∈ 𝑉: its neighbors in 𝐺
ο‚­Computation in synchronous rounds
𝑂 log 𝑛 bits per edge per round
Goal: does 𝐺 satisfy property 𝑃?
ο‚­Yes: all nodes accept
ο‚­No: some node rejects
EXAMPLES OF HARD PROBLEMS
Θ 𝑛 rounds:
ο‚­Distinguishing diameter 4 from 5 [Holzer and Wattenhofer β€˜12]
ο‚­ 3 2 βˆ’ πœ– -approx of the diameter [Abboud, Censor-Hillel, Khoury ’16]
ο‚­πΆπ‘˜ -freeness for odd π‘˜ [Drucker, Kuhn, O. ’14]
Θ
𝑛 rounds:
ο‚­MST, … [Das Sarma et. al. ’12]
ο‚­3/2-approx of the diameter [Holzer and Wattenhofer β€˜12], [Frishknecht
et al. β€˜12]
𝐢4 -freeness [Drucker, Kuhn, O. ’14]
OPEN PROBLEM
Is there a β€œnatural” graph property that requires πœ” 𝑛
rounds?
WHAT ABOUT PROPERTY-TESTING?
Does 𝐺 satisfy 𝑃, or is 𝐺 πœ–-far from satisfying 𝑃?
Need to add/remove πœ– 𝐸 edges
to get a graph satisfying 𝑃
TESTING TRIANGLE-FREENESS, 𝐸 = Θ 𝑛
Observation:
If 𝐺 is πœ–-far from triangle-free, then
𝐺 contains Θ πœ–π‘›2 edge-disjoint triangles
Algorithm:
ο‚­Sample each edge w.p. 𝑂 1/ πœ–π‘›2
ο‚­Announce sampled edges to neighbors
ο‚­Locally check for triangles
2
TESTING SUBGRAPH-FREENESS
[Censor-Hillel, Fischer, Schwartzman,Vasudev’16]: 𝑂 1 πœ– 2
for triangle-freeness (general model)
[Fraigniaud,Rapaport,Salo,Todinca’16]:
𝑂 1 πœ– 2 for all connected 4-node graphs
What about 𝐢5 ? 𝐾5 ? …
LOWER BOUNDS FROM COMMUNICATION COMPLEXITY
Alice
Bob
COMMUNICATION COMPLEXITY OF
GRAPH PROPERTY TESTING
THE MODEL
ο‚­π‘˜ players with private inputs 𝐸1 , … , πΈπ‘˜ βŠ† 𝑛 2
ο‚­Does 𝐺 = 𝑛 , π‘˜π‘–=1 𝐸𝑖 satisfy 𝑃, or is it πœ–-far?
ο‚­Communication: shared blackboard
ο‚­β€œGeneral model”
ο‚­Upper bounds: edges can be duplicated
ο‚­Lower bounds: no edge duplication
β€œBUILDING BLOCKS”
Can efficiently implement:
ο‚­Sampling a random subgraph
ο‚­Estimating the average degree
ο‚­Random walk
…
TESTING TRIANGLE-FREENESS
ο‚§Upper bound: 𝑂 π‘˜ 𝑛𝑑
β€˜06])
1 4
+ π‘˜ 2 (adapting [Alon et. al.
ο‚­Also: simultaneous protocol with CC of 𝑂 π‘˜ 𝑛 when 𝑑 = 𝑂
and 𝑂 π‘˜ 𝑛𝑑 1 3 when 𝑑 = Ξ© 𝑛
𝑛
TESTING TRIANGLE-FREENESS: LOWER BOUNDS
𝑑 = Θ 1 , two players, one-way: Ξ©
𝑛
ο‚§By reduction from Boolean Hidden Matching
ο‚§β€œEvidence of hardness”: finding a triangle edge when 𝐺 is
πœ–-far from βˆ†-free
𝑑 = Θ
𝑛 , three players, one-way*: Ξ© 𝑛1
4
ο‚­ Alice and Bob talk back-and-forth
ο‚­ Charlie observes, then outputs the answer
𝑑 = Θ
𝑛 , three players, simultaneous: Ξ©
𝑛
THE HARD DISTRIBUTION [INSPIRED BY ALON ET. AL. β€˜06]
Charlie
Tripartite graph 𝐺 = π‘ˆ βˆͺ 𝑉1 βˆͺ 𝑉2 , 𝐸
The edges are iid 𝐡𝛾
𝑉1
𝑛
π‘ˆ
E #triangles = Θ 𝑛3/2
W.h.p., 𝐺 is Θ 1 -far from triangle-free
𝑉2
Let 𝑝 be the probability that 𝑒 ∈ Ξ”
THE HARD DISTRIBUTION [INSPIRED BY ALON ET. AL. β€˜06]
Tripartite graph 𝐺 = π‘ˆ βˆͺ 𝑉1 βˆͺ 𝑉2 , 𝐸
The edges are iid 𝐡𝛾
𝑛
Θ 𝑛1
4
Θ 𝑛1
4
𝑉1
π‘ˆ
𝑒
Upper bound:
ο‚­Fix 𝑒 ∈ π‘ˆ
ο‚­Alice and Bob send Θ 𝑛1
4
edges from 𝑒
𝑉2
THE PROTOCOL’S GOAL
Charlie should output {𝑖, 𝑗} ∈ 𝑋𝐢 such that:
𝑖, π‘˜ ∈ 𝑋𝐴
Pr βˆƒπ‘˜:
𝑀𝐴 , 𝑀𝐡 β‰₯ 1 βˆ’ 𝛿
𝑗, π‘˜ ∈ 𝑋𝐡
𝑖, 𝑗 is β€œcovered”
Alice and Bob should provide Charlie
with Ξ©
𝑛 edges that β€œlook covered”.
𝑖
π‘˜
𝑗
THE PROTOCOL’S GOAL
Charlie should output {𝑖, 𝑗} ∈ 𝑋𝐢 such that:
𝑖, π‘˜ ∈ 𝑋𝐴
Pr βˆƒπ‘˜:
𝑀𝐴 , 𝑀𝐡 β‰₯ 1 βˆ’ 𝛿
𝑗, π‘˜ ∈ 𝑋𝐡
𝑀𝐴 , 𝑀𝐡 are β€œgood” if:
π‘˜
(Pr 𝑒 covered 𝑀𝐴 , 𝑀𝐡 βˆ’ 𝑝) β‰₯ Ξ©
π‘’βˆˆπ‘‰1 ×𝑉2
𝑖
𝑛
Prior probability of being
In a triangle
𝑗
BOUNDING THE SUM OF COVER-PROBABILITIES
Pr 𝑖, 𝑗 covered 𝑀𝐴 , 𝑀𝐡
≀
π‘˜ Pr
𝑖, π‘˜ ∈ 𝑋𝐴 ∧ 𝑗, π‘˜ ∈ 𝑋𝐡 𝑀𝐴 , 𝑀𝐡
=
π‘˜ Pr
𝑖, π‘˜ ∈ 𝑋𝐴 |𝑀𝐴 β‹… Pr 𝑗, π‘˜ ∈ 𝑋𝐡 |𝑀𝐡
π‘’βˆˆπ‘‰1 ×𝑉2 (Pr
≀
𝑒 covered 𝑀𝐴 , 𝑀𝐡 βˆ’ 𝑝)
𝛾
Sum of
𝐿1 distance
between
Pr
𝑒 ∈ 𝑋𝐴 |𝑀
𝐴 βˆ’ 𝑛
π‘’βˆˆπ‘ˆ×𝑉
1
posterior and prior on Alice’s input
𝛾
Sum of
𝐿1 distance
between
Pr
𝑒 ∈ 𝑋𝐡 |𝑀
𝐡 βˆ’ 𝑛
π‘’βˆˆπ‘ˆ×𝑉
2
posterior and prior on Bob’s input
BOUNDING THE SUM OF COVER-PROBABILITIES
Sum of 𝐿1 distance between
posterior and prior on Alice’s input
≀
π‘’βˆˆπ‘ˆ×𝑉1 𝐷𝐾𝐿 (
𝑋𝑒 |𝑀𝐴 βˆ₯ 𝑋𝑒 )
(β€œLinear Pinsker”)
≀ 𝐷𝐾𝐿 ( 𝑋|𝑀𝐴 βˆ₯ 𝑋 )
With 𝑛1
4
communication bits: typically ≀ 𝑛1
4
BOUNDING THE SUM OF COVER-PROBABILITIES
π‘’βˆˆπ‘‰1 ×𝑉2 (Pr
𝑒 covered 𝑀𝐴 , 𝑀𝐡 βˆ’ 𝑝)
𝛾
Sum
of
𝐿
distance
between
≀ π‘’βˆˆπ‘ˆ×𝑉1 Pr1 𝑒 ∈ 𝑋𝐴 |𝑀𝐴 βˆ’
posterior and prior on Alice’s 𝑛input
β€œUsually” O 𝑛1
≀𝑂
𝑛
4
𝛾
Sum
of
𝐿
distance
between
π‘’βˆˆπ‘ˆ×𝑉2 Pr 𝑒1∈ 𝑋𝐡 |𝑀𝐡 βˆ’ 𝑛
posterior and prior on Bob’s input
β€œUsually” O 𝑛1
4
SIMULTANEOUS PROTOCOLS?
Tripartite graph 𝐺 = π‘ˆ βˆͺ 𝑉1 βˆͺ 𝑉2 , 𝐸
The edges are iid 𝐡𝛾
𝑛
Θ 𝑛1
4
Θ 𝑛1
4
𝑉1
π‘ˆ
𝑒
Upper bound:
ο‚­Fix 𝑒 ∈ π‘ˆ
ο‚­Alice and Bob send Θ 𝑛1
4
edges from 𝑒
𝑉2
OPEN PROBLEMS: SUBGRAPH-FREENESS
ο‚§β€œReal” lower bounds for testing triangle-freeness
ο‚§Multi-round
ο‚§Testing triangle-freeness, not β€œfinding a triangle”
ο‚§Going to larger subgraphs
ο‚§Embedding back in the CONGEST model