Matroids, Secretary Problems, and Online Mechanisms Nicole Immorlica, Microsoft Research Joint work with Robert Kleinberg and Moshe Babaioff.

Slides:



Advertisements
Similar presentations
On allocations that maximize fairness Uriel Feige Microsoft Research and Weizmann Institute.
Advertisements

A threshold of ln(n) for approximating set cover By Uriel Feige Lecturer: Ariel Procaccia.
Optical networks: Basics of WDM
Approximation Algorithms Chapter 14: Rounding Applied to Set Cover.
1 EE5900 Advanced Embedded System For Smart Infrastructure Static Scheduling.
1 Maximal Independent Set. 2 Independent Set (IS): In a graph G=(V,E), |V|=n, |E|=m, any set of nodes that are not adjacent.
Multicut Lower Bounds via Network Coding Anna Blasiak Cornell University.
Dynamic Wavelength Allocation in All-optical Ring Networks Ori Gerstel and Shay Kutten Proceedings of ICC'97.
Combinatorial Algorithms
The Probabilistic Method Shmuel Wimer Engineering Faculty Bar-Ilan University.
Absorbing Random walks Coverage
The number of edge-disjoint transitive triples in a tournament.
Greedy Algorithms for Matroids Andreas Klappenecker.
Item Pricing for Revenue Maximization in Combinatorial Auctions Maria-Florina Balcan, Carnegie Mellon University Joint with Avrim Blum and Yishay Mansour.
Approximation Algorithm: Iterative Rounding Lecture 15: March 9.
A general approximation technique for constrained forest problems Michael X. Goemans & David P. Williamson Presented by: Yonatan Elhanani & Yuval Cohen.
Matroids, Secretary Problems, and Online Mechanisms Nicole Immorlica, Microsoft Research Joint work with Robert Kleinberg and Moshe Babaioff.
Michael Bender - SUNY Stony Brook Dana Ron - Tel Aviv University Testing Acyclicity of Directed Graphs in Sublinear Time.
Job Scheduling Lecture 19: March 19. Job Scheduling: Unrelated Multiple Machines There are n jobs, each job has: a processing time p(i,j) (the time to.
1 Introduction to Approximation Algorithms Lecture 15: Mar 5.
A General Approach to Online Network Optimization Problems Seffi Naor Computer Science Dept. Technion Haifa, Israel Joint work: Noga Alon, Yossi Azar,
Packing Element-Disjoint Steiner Trees Mohammad R. Salavatipour Department of Computing Science University of Alberta Joint with Joseph Cheriyan Department.
1 Joint work with Shmuel Safra. 2 Motivation 3 Motivation.
1 Distributed Computing Optical networks: switching cost and traffic grooming Shmuel Zaks ©
Minimal Spanning Trees What is a minimal spanning tree (MST) and how to find one.
The Theory of NP-Completeness 1. Nondeterministic algorithms A nondeterminstic algorithm consists of phase 1: guessing phase 2: checking If the checking.
Called as the Interval Scheduling Problem. A simpler version of a class of scheduling problems. – Can add weights. – Can add multiple resources – Can ask.
Competitive On-Line Admission Control and Routing By: Gabi Kliot Presentation version.
Yossi Azar Tel Aviv University Joint work with Ilan Cohen Serving in the Dark 1.
APPROXIMATION ALGORITHMS VERTEX COVER – MAX CUT PROBLEMS
1 Introduction to Approximation Algorithms. 2 NP-completeness Do your best then.
© 2009 IBM Corporation 1 Improving Consolidation of Virtual Machines with Risk-aware Bandwidth Oversubscription in Compute Clouds Amir Epstein Joint work.
1 Greedy algorithm 叶德仕 2 Greedy algorithm’s paradigm Algorithm is greedy if it builds up a solution in small steps it chooses a decision.
Market Design and Analysis Lecture 5 Lecturer: Ning Chen ( 陈宁 )
Greedy Algorithms and Matroids Andreas Klappenecker.
First lecture in Mathematics
Maximizing the Spread of Influence through a Social Network Authors: David Kempe, Jon Kleinberg, É va Tardos KDD 2003.
Testing the independence number of hypergraphs
CSE 589 Part VI. Reading Skiena, Sections 5.5 and 6.8 CLR, chapter 37.
CS223 Advanced Data Structures and Algorithms 1 Maximum Flow Neil Tang 3/30/2010.
Implicit Hitting Set Problems Richard M. Karp Erick Moreno Centeno DIMACS 20 th Anniversary.
Improved Competitive Ratios for Submodular Secretary Problems ? Moran Feldman Roy SchwartzJoseph (Seffi) Naor Technion – Israel Institute of Technology.
© The McGraw-Hill Companies, Inc., Chapter 12 On-Line Algorithms.
Unique Games Approximation Amit Weinstein Complexity Seminar, Fall 2006 Based on: “Near Optimal Algorithms for Unique Games" by M. Charikar, K. Makarychev,
A Unified Continuous Greedy Algorithm for Submodular Maximization Moran Feldman Roy SchwartzJoseph (Seffi) Naor Technion – Israel Institute of Technology.
Correlation Clustering Nikhil Bansal Joint Work with Avrim Blum and Shuchi Chawla.
11 -1 Chapter 12 On-Line Algorithms On-Line Algorithms On-line algorithms are used to solve on-line problems. The disk scheduling problem The requests.
TU/e Algorithms (2IL15) – Lecture 12 1 Linear Programming.
Non-LP-Based Approximation Algorithms Fabrizio Grandoni IDSIA
COSC 3101A - Design and Analysis of Algorithms 14 NP-Completeness.
Approximation Algorithms based on linear programming.
Greedy Algorithms. p2. Activity-selection problem: Problem : Want to schedule as many compatible activities as possible., n activities. Activity i, start.
New Algorithms for Disjoint Paths Problems Sanjeev Khanna University of Pennsylvania Joint work with Chandra Chekuri Bruce Shepherd.
CHAPTER SIX T HE P ROBABILISTIC M ETHOD M1 Zhang Cong 2011/Nov/28.
Theory of Computational Complexity Probability and Computing Chapter Hikaru Inada Iwama and Ito lab M1.
Greedy Algorithms.
Independent Cascade Model and Linear Threshold Model
Chapter 5. Greedy Algorithms
The Price of information in combinatorial optimization
Distributed Submodular Maximization in Massive Datasets
Independent Cascade Model and Linear Threshold Model
Framework for the Secretary Problem on the Intersection of Matroids
Discrete Mathematics for Computer Science
Dynamic and Online Algorithms for Set Cover
3.5 Minimum Cuts in Undirected Graphs
Coverage Approximation Algorithms
The Byzantine Secretary Problem
Independent Cascade Model and Linear Threshold Model
Submodular Maximization with Cardinality Constraints
Presentation transcript:

Matroids, Secretary Problems, and Online Mechanisms Nicole Immorlica, Microsoft Research Joint work with Robert Kleinberg and Moshe Babaioff

The Secretary Problem Company wants to hire a secretary  There are n secretaries available, each of whom will accept any offer they receive  Each secretary i has an inherent value v i  Secretaries interview in a random order, revealing their value at the interview  Hiring decision must be made at the interview Question: Can the company design an interviewing procedure to guarantee that it hires the (approximately) best secretary?

The Secretary Algorithm Algorithm:  Observe first n/e elements. Let v=maximum.  Pick the next element whose value is > v. Theorem: Pr(picking max elt. of S) > 1/e.* Proof: Select best elt. if i’th best elt is best in first 1/e elts and best elt is first among best (i-1) elts. Happens with probability (1/e) ¢ (1-1/e) i ¢ (1/i). * Elements come in a random order. Threshold time t = n/etime t = n i’th bestbest 2 nd best through (i-1) st best

Generalized Secretary Problems Input  Set of secretaries {1, …, n}, each has a value v i  Feasible or independent family of subsets of {1, …, n} Secretaries arrive in random order, and alg. must decide online whether to select each secretary Goal is to select maximum weight feasible set Performance measure is competitive ratio: E[weight of selected set]/[weight of max ind. set]

Example Multicast in a network Each node wants an edge-disjoint path to source Value: $8 Value: $10 Value: $7 Value: $12

Special Cases Standard secretary problem: independent sets are all singletons Thm [Dynkin ‘63]: There is an algorithm with competitive ratio (1/e). k-Secretary problem: independent sets are all sets of size at most k Thm [Kleinberg ‘05]: There is an algorithm with competitive ratio 1- Θ (k -1/2 )

Matroid Secretary Problems Defn.: A matroid consists of a universe of elements and a family of distinguished subsets called independent sets which satisfy:  Subsets of independent sets are independent.  Exchange property: If S,T are independent and |S| < |T| then S U {t} is independent for some t in T. A matroid secretary problem is a generalized secretary problem in which the independent sets form a matroid. The standard and k-secretary problems are matroid secretary problems.

More Examples Gammoid Matroids:  Elements (customers) are sources in a graph  Set S of sources is independent if there exist edge-disjoint paths routing each source in S to the sink Graphical Matroids:  Elements are the edges of an undirected graph G = (V;E)  Set of edges is independent if it does not contain a cycle Truncated Partition Matroids of rank k:  Elements (items) are partitioned into m sets  Set of elements is independent if it contains at most one item from each partition and at most k items in total (production constraint)

Open Question Is there a constant-competitive secretary algorithm for all matroids? Intuition:  In matroids, a single mistake can only ruin your chance of picking one element of the best set  If algorithm could discard a previously selected element, matroid properties guarantee the greedy algorithm always selects optimal set. Thm.: If independent sets are allowed to be an arbitrary set system closed under containment, no algorithm can be constant-competitive.

Our Results 1. O(log k)-competitive algorithm for general matroids, where k is the rank competitive algorithm for graphical matroids. 3. 4d-competitive algorithm for transversal matroids, where d is the max size of an agent’s set of desired items. 4. If M has a c-competitive algorithm, then every truncation of M has a 48c-competitive algorithm.

Lower bound for general set systems

Proof of lower bound Suppose the algorithm makes its first selection at time t, and that it chooses an element x in Si. All future selections must be elements of Si. Let Ti be the subset of Si which have not yet been observed at time t. The values of the elements of Ti are independent of all the information observed up to time t; there are less than k elements in Ti and each of them has expected value 1/k, so the expected combined value of all remaining elements is less than 1. The only element selected up to time t is x, whose value is at most 1. This proves that the expected value of the set selected by the algorithm is less than 2. Let j = k/(2 ln(k)). Probability that at least one Si has j elements of value 1 is 1-o(1). Therefore the maximum-weight element of I has weight lower bounded by j=(ln n /2 ln ln n)

O(log k)-competitive algorithm 1. Assume the algorithm knows an integer s between log(k)-1 and log(k).* 2. Sample the first n/2 elements without selecting any of them. Let v* be the maximum value observed so far. Pick random r in {1,…,s}. 3. Set threshold value w = v*/2 r. 4. From then on, select every element independent of previous selections whose value is at least w. * This assumption is not needed. We can estimate s using the rank of the sample.

Single Threshold Algorithms An algorithm which computes a threshold value v and stopping time  and then selects every feasible element after  whose value is at least v  O(log k)-competitive algorithm is single threshold Counterexample: single threshold algorithms are not constant competitive (lower-bound)  Partition matroid with k sets of size n/k  Set i has (k-1) values =1/(c i ) and 1 value =1/i

Greedy Algorithms Algorithm  Observe a constant fraction of the input without selecting any element  Compute a maximum weight basis among elements observed so far  Select any feasible element which can be exchanged with an element in the basis to improve its weight Counterexample: greedy algorithms can not be constant competitive … … 1 Node i Weight n-i Weight i

Open Questions Is there a constant-competitive algorithm for general matroids? If so, is it e-competitive? Relaxations:  Matroid structure known in advance.  Values assigned randomly to the matroid elements. Special cases:  Transversal matroids, gammoid matroids  Is the class of constant-competitive matroids closed under contraction?