Presentation is loading. Please wait.

Presentation is loading. Please wait.

Direct product testing

Similar presentations


Presentation on theme: "Direct product testing"β€” Presentation transcript:

1 Direct product testing
Irit Dinur, Inbal Livni Navon Weizmann Institute

2

3 Direct product operation
𝑓 βŠ—π‘˜ π‘₯ 1 , π‘₯ 2 ,…, π‘₯ π‘˜ =𝑓 π‘₯ 1 ,𝑓 π‘₯ 2 ,…𝑓 π‘₯ π‘˜ Direct product operation Basic operation on functions Useful for hardness amplification β€œone new instance is as hard as k”

4 Testing direct products
Given a function, 𝑓: 𝑛 π‘˜ β†’ 0,1 π‘˜ is it a direct product? Related to parallel repetition: the honest players’ strategy is a direct product; if we knew that the player strategy is a DP then perfect parallel repetition is immediate, with perfectly exponential decay Natural test:

5 Testing direct products
How good is this test? Suppose it succeeds with probability 99%, how close is f to DP ? This is a property testing question. Hope: f is 99% close to DP Warmup: for each π‘§βˆˆ[𝑛] choose the most popular value An averaging argument shows that for 90% of tuples ( 𝑧 1 ,…, 𝑧 π‘˜ ), f outputs the popular value on about 90% of the coordinates Aside: [D-Steurer 14] showed a stronger result: for 99% of the tuples, f agrees with m on every entry

6 Testing direct products
Studied and used for some PCP constructions; initially suggested as a combinatorial alternative to β€œlow degree test” [Goldreich-Safra, D-Reingold, D-Goldenberg, Impagliazzon-Kabanets- Wigderson, D-Steurer] Main interest: the 1% regime, aka β€œsmall soundness” or β€œlist decodingβ€œ regime

7 The 1% regime Suppose f passes the test with probability 1%, can anything be said about its structure? [unlike standard property testing set-up, we cannot afford boosting success by repetition] …cannot expect a conclusion of the form β€œthere is a single DP g that approximates f”. we must allow a mix of several valid DP functions (aka a β€œlist”) Examples of properties that have 1% tests: Low degree tests Long code tests Gowers norm and additive combinatorics (these are related to low degree tests) Direct product tests For what properties do we have tests that imply meaningful structure even if they pass with low β€œ1%” probability? (Can be viewed as a stronger form of property testing) For which value of β€œ1%” can structure be found? Ultimately= anything above random

8 The 1% regime for Direct Product tests
D-Goldenberg: if f passes the V test with probability πœ– > 1/π‘π‘œπ‘™π‘¦(π‘˜), then it is π‘π‘œπ‘™π‘¦(πœ–) correlated to a DP Cannot expect such structure for πœ– below 1/π‘π‘œπ‘™π‘¦(π‘˜) : There is a function f that is not close to any DP, but passes the test w prob πœ– Impagliazzo Kabanets Wigderson: add a third query, and the counter-example goes away IKW theorem: for any πœ– > exp⁑(βˆ’βˆšπ‘˜), if f passes the Z test with probability β‰₯πœ–, it is π‘π‘œπ‘™π‘¦(πœ–) correlated to a DP New (almost a decade later): same holds for all πœ– > exp⁑(βˆ’π‘˜),

9 Main Result If 𝑓: 𝑛 π‘˜ β†’ 0,1 π‘˜ passes the Z-test with probability πœ–β‰₯ exp βˆ’π‘˜ , then it is π‘π‘œπ‘™π‘¦(πœ–) close to a direct product function The result is β€œballpark” tight Number of queries: it is impossible with 2 queries, as per [DG] example; so 3 is it Soundness: cannot go below exp⁑(βˆ’π‘˜) a random function where 𝑓(π‘₯) is chosen independently for each π‘₯∈ 𝑛 π‘˜ passes the test with probability exp βˆ’π‘˜ Open: can we get soundness test down to the randomness threshold? ( 2 βˆ’π‘˜ +𝛿 and not only (1.001) βˆ’π‘˜ ) β€œclose to DP” : there is approximate closeness, and exact closeness,

10 Proof Given 𝑓: 𝑛 π‘˜ β†’ 0,1 π‘˜ that passes the Z-test with probability πœ– β‰₯ exp βˆ’π‘˜ The proof will construct a DP that is close to f. How? Taking a popularity vote on a restricted set of x’s Choose a random restriction A= π‘₯ 1 , π‘₯ 2 , π‘₯ 3 , π‘₯ 4 ,βˆ—, βˆ—,βˆ—,βˆ— Choose a fixed answer 𝛼 ∈ 0,1 𝐴 , e.g. β€œ1001”, Consider only π‘₯’s that 𝑓 π‘₯ 𝐴 =𝛼 Take a majority vote on these x’s, (assume for now that f is invariant for permutations) 𝑔 𝑦 =π‘π‘œπ‘π‘’π‘™π‘Žπ‘Ÿ 𝑓 π‘₯ 1 , π‘₯ 2 , π‘₯ 3 , π‘₯ 4 ,𝑦,βˆ—,βˆ—,βˆ— Need to prove that inside this restriction, indeed f is a DP, i.e. the value of f on the i-th coordinate, depends only on the i-th coordinate.

11 Proof IKW: used sampling properties of k-sets and t-sets, which hold as long as πœ–β‰₯ exp βˆ’ π‘˜ 𝑑≀ π‘˜ When πœ–β‰ˆ exp βˆ’π‘˜ this fails. Instead, we perform β€œdensification”

12 Inside the restriction:
We have partial function 𝑓: 𝑛 π‘˜ β€² β†’ 0,1 π‘˜ β€² βˆͺ{βŠ₯}, that is defined on some πœ–β‰₯exp βˆ’π‘˜β€² fraction of the inputs We know that whenever the function is defined on two β€œintersecting” inputs, the answers agree whp β€œdensify” the function by using majority on small balls Prove that the densified function has similar properties to the original function, (using: reverse hyper-contractivity to reason about expansion of small sets in the test graph) With the dense function we can finish like before

13 Restriction decoding Assuming success 99%  Unique decoding
List decoding ? Restriction decoding and β€œzoom ins” Third query – brings us back to list decoding

14 (possibly: some form of high dimensional expansion)
Agreement tests The question of DP testing belongs to a family of testing questions called β€œagreement testing” Given a collection of partial views of a domain, that typically pairwise agree, is there a global view they all agree on? E.g. plane vs. plane low degree test E.g. tuple vs. tuple tests that we saw today Such agreement theorems are an important ingredient in all PCPs What is the connection between these questions and the geometry of the set of β€œlocal views” ? (possibly: some form of high dimensional expansion)


Download ppt "Direct product testing"

Similar presentations


Ads by Google