Algorithms Reading Group Spring 2006

Venue : B-13
Time : 11:30-12:30
Days : Thursdays.

If you would like to speak, please send e-mail to rraman at cs dot uiowa dot edu.

Date
Speaker
Topic
1/31/2006
Rajiv Raman
Min-wise Independent Permutations.
2/9/2006 Kasturi Varadarajan
Online algorithms for metric minimum bipartite matching    
2/16/2006
Imran Pirwani
Virtual Coordinates for adhoc sensor networks
2/23/2006
Ben McCune
An application of game theoretic techniques to cryptography
3/2/2006
Rajiv Raman Distributions on level-sets with applications to approximation algorithms
3/9/2006
Sriram Pemmaraju Labeling schemes for tree representations
3/23/2006
Yan Liu  
3/30/2006
Matt Gibson
4/6/2006
Wenli He
4/13/2006
Saurav Pandit
4/20/2006
Gaurav Kanade
4/27/2006
Kasturi Varadarajan

5/4/2006
Tao He


Seminars from previous semesters:  Fall 2005, Spring 2005, Fall 2004, Spring 2004, Fall 2003.

Abstracts


Rajiv Raman (1/31/2006)
Min-wise independent permutations.

Given a set [n] = {0,1,...,n-1}, a set of permutations F of [n] is said to be min-wise independent if for any subset X of [n], and a permutation chosen randomly from F, each element of X has equal probability of becoming the minimum element under this permutation. I will prove some upper and lower bound results on the size of min-wise independent families and for some relaxations of the min-wise independence concept. I will also describe applications of min-wise independent permutations in document clustering and in finding dense subgraphs of large graphs.
1. "Min-wise independent permutations", A.Z.Broder, M.Charikar, A.M.Frieze and M.Mitzenmacher, JCSS (60) (2000).
2. "Discovering large dense subgraphs in massive graphs", D.Gibson, R.Kumar. A.Tomkins, VLDB 2005.

Kasturi Varadarajan (2/9/2006)
Online algorithms for Metric Minimum Bipartite Matching.

In this talk, we will discuss the online metric minimum bipartite matching problem, following the paper `Randomized Online Algorithms for Minimum Metric Bipartite Matching' by Meyerson, Nanavati, and Poplawski in SODA 06. Suppose there is a metric space in which k points are specified as red in advance. Blue points arrive one by one, and as soon as a blue point arrives, it has to be matched to a red point. No red point can serve more than one blue point. After k blue points have arrived, we compare the cost of the perfect matching computed by our algorithm to the cost of the optimal perfect matching (For instance, this can be computed by an algorithm that knows all the red and blue points before computing the matching). We will discuss how well some matching algorithms perform under this measure.
The paper can be accessed here.
Imran Pirwani(2/16/2006)
Virtual Coordinates for adhoc sensor networks.

In many applications of wireless ad hoc and sensor networks, position- awareness is of great importance. Often, as in the case of geometric routing,it is sufficient to have virtual coordinates, rather than real coordinates. In this paper, we address the problem of obtaining virtual coordinates based on connectivity information. In particular, we propose the first approximation algorithm for this problem and discuss implementational aspects. In the talk, I will be focusing on finding distance information that is subsequently used to arrive at an embedding in a plane.
The paper can be accessed here.
Ben McCune (2/23/2006)
An Application of Game Theoretic Techniques to Cryptography

This paper provides an application of game theoretic techniques to the analysis of a class of multiparty cryptographic protocols for secret bit exchange.
The paper can be accessed here.
Rajiv Raman(3/2/2006)
Distributions on level-sets with applications to approximation algorithms

For vectors x in {0,1}^t, the weight of x is the number of 1's in it. Let Let W_k(t) denote the set of elements from {0,1}^t with weight equal to k and let [s] denote the set {1,2, .., , s}. Consider any sequence P = (p_1, ..., p_t) of t reals such that each p_i \in [0,1] and such that \sum p_i is an integer. Each P defines a distribution D(t;P) defined to be any distribution on {0,1}^t such that if l= \sum p_i and (X_1, ..., X_t) denotes a vector sampled from D(t;P) then the following hold : 1. for all i, Pr[X_i = 1] = p_i 2. Pr[|{i : X_i = 1}| = l] = 1 3. some negative correlation properties hold for all S \subset [t]. Pr[ X_i = 0] <= Pr[X_i = 0] for all X_i in S, and Pr[ X_i = 1] <= Pr[X_i = 1] for all X_i in S. I'll present proof of existance of such distributions and a linear-time algorithm to sample from these distributions and describe some applications to approximation algorithms.
The paper can be accessed here.
Sriram Pemmaraju(3/9/2006)
Labeling schemes for tree representations

We are given a graph G = (V, E) and a labeling of edges incident on each vertex. Specifically, for each vertex v, the edges incident on v are labeled 0, 1,..., degree(v)-1 in some order. The problem is to compute a spanning tree T of G with compact labeling. There are different ways of defining the labeling cost of a spanning tree. In this talk I will define a couple of measures of labeling cost and present bounds on labeling cost and algorithms for computing spanning trees with minimum labeling cost.
This paper is from the paper Labeling Schemes for Tree Representations, R. Cohen, P. Fraigniaud, D. Ilcinkas, A. Korman, and D. Peleg, IWDC 2005.

P.S. For one of the labeling cost measures, the authors conjecture that for any n-vertex graph and any given edge labeling, there is a spanning tree with labeling cost O(n). The authors have shown the existence of a spanning tree with labeling cost O(nloglogn). My hope is that some of the students in the audience will be able to close this gap.