{[ promptMessage ]}

Bookmark it

{[ promptMessage ]}

# l02 - 6.896 Sublinear Time Algorithms February 8 2007...

This preview shows pages 1–2. Sign up to view the full content.

6.896 Sublinear Time Algorithms February 8, 2007 Lecture 2 Lecturer: Ronitt Rubinfeld Scribe: Adriana Lopez In this lecture, we will prove a theorem from last lecture that states that there exists a closeness tester in L 2 . We will also give an algorithm for L 1 -distance testing. Recall the following definitions and facts from last lecture: Definition 1 p x = Pr[ p outputs x ] Definition 2 (collision probability of p and q ) CP ( p, q ) = Pr[ sample from p equals sample from q ] = x p x q x Definition 3 (self-collision probability of p ) SCP ( p ) = Pr[ 2 samples from p are equal ] = x p 2 x = p 2 2 1 L 2 -Distance Testing Theorem 4 For all , p, q , there exists an O ( 4 ) sample tester T such that If p q 2 < 2 then Pr[ T accepts ] 2 3 . If p q 2 > then Pr[ T rejects ] 2 3 . We will prove that tester T from last lecture (reproduced below) satisfies the conditions stated in Theorem 4. T ( p, q, ): m O ( 4 ) S p m samples from p S q m samples from q r p number of self-collisions in S p r q number of self-collisions in S q Q p m (new) samples from p Q q m (new) samples from q r pq number of pairs i, j s.t. i th sample in Q p is equal to j th sample from Q q r 2 m m 1 ( r p + r q ) s 2 r pq if r s > m 2 2 2 then reject else accept Also recall the following fact from last lecture: E [ r s ] = m 2 p q 2 2 Lemma 5 Let b = max x p x , q x . Then Var[ r s ] O ( m 3 b 2 + m 2 b ) .

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}