Download presentation
Presentation is loading. Please wait.
1
On the Error Parameter in Dispersers Ronen Gradwohl, Omer Reingold, and Amnon Ta-Shma. Joint work with Guy Kindler Microsoft Research
2
On the Error Parameter in Dispersers Ronen Gradwohl, Omer Reingold, and Amnon Ta-Shma. Joint work with Guy Kindler Microsoft Research
3
this talk: this talk: Goal: better explicit bipartite Ramsey constructions We have: some seeded dispersers and extractors. Observe: bipartite Ramsey µ strong seeded dispersers. Draw a path from where we are to where we want to go. Make some steps on that path.
4
Entropy (not really…) Define: The entropy of a set X by H(X)=log 2 ( | X | ) X (n-bit strings)
5
Ram 0/1 Bipartite Ramsey Graphs A function Ram: { 0,1 } n x { 0,1 } n ! { 0,1 } is ( k, k ) bipartite Ramsey if 8 X,Y µ { 0,1 } n, H(X),H(Y)>k, Ram(X,Y)= { 0,1 }. (n-bit strings) |X| ¸ 2 k |Y| ¸ 2 k x y
6
Bipartite Ramsey Graphs A function Ram: { 0,1 } n x { 0,1 } n ! { 0,1 } is ( k, k ) bipartite Ramsey if 8 X,Y µ { 0,1 } n, H(X),H(Y)>k, Ram(X,Y)= { 0,1 }. Known to exists for k=O(log n). [GV 88] k=n/2 [?] (O(log n),n/2 )- bipartite Ramsey graph. [BKSSW 05] k= n [BRSSW 06] k=n
7
Seeded Dispersers D : { 0,1 } n x { 0,1 } s ! { 0,1 } m is a (k, ) disperser if for H(X) > k, (m-bit strings) D x |X| ¸ 2 k (s-bit string) r If s > m, take D(x,r)=r [m] Interesting only when m > s !
8
Strong Dispersers D : { 0,1 } n x { 0,1 } s ! { 0,1 } m is a (k, ) strong disperser if for H(X) > k, D x |X| ¸ 2 k (s-bit string) r (m-bits)(s-bits)
9
Strong Dispersers D : { 0,1 } n x { 0,1 } s ! { 0,1 } m is a (k, ) strong disperser if for H(X) > k, D x |X| ¸ 2 k (s-bit string) r (m-bits)(s-bits) . For all but fraction of r ’s,
10
0/1 Strong Dispersers D : { 0,1 } n x { 0,1 } s ! { 0,1 } m is a (k, ) strong disperser if for H(X) > k, D x |X| ¸ 2 k (s-bit string) r (s-bits) . For all but fraction of r’s,
11
D : { 0,1 } n x { 0,1 } s ! { 0,1 } m is a (k, ) strong disperser if for H(X) > k, For all but fraction of r ’s, | Y |> ¢ 2 s, ! 9 r 2 Y s.t. 0/1 Strong Dispersers D x |X| ¸ 2 k (s-bit string) (s-bits) r
12
0/1 Strong Dispersers D : { 0,1 } n x { 0,1 } s ! { 0,1 } m is a (k, ) strong disperser if for H(X) > k, D x |X| ¸ 2 k (s-bit string) r For all but fraction of r ’s, | Y |> ¢ 2 s, ! 9 r 2 Y s.t. D is (k,s-log(1/ )) Ramsey!
13
D : { 0,1 } n x { 0,1 } s ! { 0,1 } is a (k, ) strong disperser. D is (k,s-log(1/ )) Ramsey. k ¸ (log n) s=O(log n)+log(1/ ) In that case D is (k,O(log n)) -Ramsey! For extractors: s ¸ O(log n)+ 2 ¢ log(1/ ) If s=log(n)+ 2 ¢ log(1/ )=n, D is (k,sqrt(n)) -bipartite. Parameters of Strong Dispersers
14
So can we get s=O(log(n))+log(1/ ) ? no. Can we get s=s n +log(1/ ) for some small function s n ? (would imply a (k,s n ) -bipartite Ramsey construction) no. So what do we get?? An almost strong disperser…
15
Almost-strong dispersers A (k, ) disperser D is strong in t bits if (s=t+u bits) (t-bits) D x |X| ¸ 2 k r r [t] (m-bits) Only interesting if m>u.
16
Almost-strong dispersers A (k, ) disperser D is strong in t bits if Our construction: t= O(log n+loglog(1/ )) + log(1/ ) u=O(loglog n +loglog(1/ )), m=2 ¢ u
17
The construction SE m=100(log k+loglog(1/ )) s’=O(m+log n) D TUZ t=10s’+log(1/ ) x |X| ¸ 2 k u=O(log t) (t,1/2)-disperser(t,1/2)-disperser
18
Combinatorial interpretation We built a bipartite graph G on (V,W), | V | = | W | =2 n Each edge is associated with a list of log 5 n colors, out of a rainbow of size log 10 n. If X µ V and Y µ W have size | X | = | Y | =n 20, then E(X,Y) contains a complete rainbow.
19
Open questions Show a strong (k, ) disperser D : { 0,1 } n x { 0,1 } s ! { 0,1 } with Preferrably s n =log n + O(1). The End
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.