1

Operations Research Society of Israel

(ORSIS)

Annual Meeting 2013

May 19-20

Leonardo Hotel, Beer-Sheva

Sponsored by:

Organizing Committee:

Gad Rabinowitz

Arieh Gavious

Ella Segev

Yoav Kerner

2

הכנס השנתי של האגודה הישראלית לחקר ביצועים

0292מאי ב 02 - 91 ,באר שבע, 4הנרייטה סולד , מלון ליאונרדו נגב

Sunday Monday 9:00 Registration 8:45 Registration

10:00-11:15 Parallel Sessions S1

Transportation 1 מצפה

Continuous Optimization תמנע

Scheduling 1 ערד

9:30-10:45 Parallel Sessions M1

Strategic Behavior in Queues מצפה

Multi Agent Optimization תמנע

Combinatorial Optimization ערד

Military OR נגב

11:15-11:30 Opening Session 10:45-11:00 מצפה Prize Award Ceremony מצפה

11:30-12:25 Naor Plenary Lecture: Dorit S. Hochbaum מצפה

11:00-12:00 Plenary Lecture:

Sergiu Hart מצפה

12:25-13:55 Lunch & ORSIS General Assembly ( 13:30) 12:00-12:15 Break

13:55-15:10 Parallel Sessions S2

Queueing 1 מצפה

Scheduling 2 תמנע

Game Theory 1 ערד

Mutli Criteria Optimization ערבה

12:15-13:05 Tutorial T2:

Shlomo Yitzhaki נגב

Parallel Sessions M2

ORSIS and Mehrez prize winners מצפה

Applied Optimization ערד

Supply Chain תמנע

15:10-15:25 Break 13:05-14:10 lunch

15:25-16:15 Tutorials T1:

Nicole Adler מצפה

Iddo Eliazar תמנע

14:10-15:30 Symposium: OR in Public Service (Hebrew) מצפה

16:20-17:35 Parallel Sessions S3 Simulation

(in Memory of Reuven Rubinstein) מצפה

Queuing 2 ערד

Data Envelopment Analysis תמנע

Heavy Scheduling Problems ערבה

15:30-16:15 Tutorials T3:

Ron Holzman מצפה

Avraham Shtub תמנע

17:35-17:45 Break 16:15-16:30 Break

17:45-21:30 Evening Program (bus leaves at 5:50pm)

Tour of Beer-Sheva – Yishai Avital

Wild Plants Agriculture – Yossi Mizrachi

Dinner @ Arabica (downtown)

16:30-17:45 Parallel Session M3

Scheduling 3 מצפה

Game Theory 2 תמנע

Queueing Network in Production ערד

Transportation 2 נגב

17:45-18:00 Closing Session

(. Dקומה ) נגב, (Mקומה ) ערבה, (תיאטרון) Dקומה -ערד Dקומה -תמנע, ( מליאה)Dקומה -מצפה: מיקום החדרים

:(חלו שינויים) מועדי רכבות 00:44, 00:72, 01:44, 01:72 ,9:44 ,9:72 ,8:44 ,8:72 :בבוקרמרכז לבאר שבע ההגע 70:47 ,71:47 ,09:47 ,08:47 ,08:17 ,02:47 ,02:17 , 01:47, 01:17 :בערבמרכז מבאר שבע היציא

3

Detailed Program - Sunday, 19.5

9:00 Registration

10:00-11:15 Parallel Session S1

Transportation 1 - Chair:

Michal Tzur מצפה

Iris Forma, Tal Raviv and Michal

Tzur,

A Clustering Algorithm for the

Static Repositioning Problem

in a Bike-Sharing System

Mor Kaspi, Tal Raviv and

Michal Tzur,

Parking Reservations in

Shared Mobility Systems

Shoshana Anily and Aharona

Pfeffer,

The Uncapacitated Dial-a-Ride

Problem on a Tree

Continuous Optimization -

Chair: Amir Beck תמנע

Marc Teboulle and Amir Beck,

Nonsmooth Convex

Minimization: To Smooth or

Not to Smooth?

Ron Shefi, Marc Teboulle,

Rate of Convergence Analysis

for Proximal-Lagrangian

Methods

Shoham Sabach, Jerome Bolte,

and Marc Teboulle,

A Simple Algorithm for

Nonconvex and Nonsmooth

Minimization Problems

Scheduling 1 - Chair: Gur

Mosheiov ערד

Shlomo Karhi, Dvir Shabtay and

Daniel Oron,

Multipurpose Machine

Scheduling with Rejection and

Identical Job Processing Times

Enrique (Tzvi) Gerstl and Gur

Mosheiov,

The optimal number of used

machines in a two-stage flow

shop scheduling problem

Baruch Mor and Gur Mosheiov,

Polynomial time solutions for

scheduling problems on a

proportionate flowshop with

two competing agents

11:15-11:30 Opening Session מצפה

11:30-12:55 Naor Plenary Lecture מצפה

Chair and opener: Aharon Ben-Tal, Technion, Speaker: Dorit S. Hochbaum, Berkeley

12:25-13:55 Lunch & ORSIS General Assembly ( 13:30)

13:55-15:10 Parallel Sessions S2

Queueing 1 - Chair: Uri Yechiali מצפה

Efrat Perel, Kostia Avrachnkov, Uri Yechiali,

The Traffic Policeman Problem: A Two-Queue

System with Switching Policy Depending on the

Queue which is NOT Being Served

Nir Perel, Uri Yechiali,

The Israeli Queue with Priorities

Shlomi Reuveni, Iddo Eliazar, Uri Yechiali,

Occupation Probabilities and Catalan Numbers in

the Asymmetric Inclusion Process

Scheduling 2 - Chair: Michal Penn תמנע

Ilan Cohen, Yossi Azar, Seny Kamara, Bruce

Shepherd

Tight Bounds for Online Vector Bin Packing

Sagi Halleli, Ephraim Korach and Michal Stern, למציאת מרכזים של קבוצות ושימושים OS-שיפור בשיטת ה

Vladimir Lipets, Eitan Israeli, Zahar Chikishev, Dvora

Lopez, Avraham Har-Paz,

Crew Assignment and Pairing Optimization

System for El Al

Game Theory 1 - Chair: Ella Segev ערד

Oscar Volij, Casilda Lasso de la Vega

Segregation, Informativeness and Lorenz

domination

Ro'i Zultan

Timing of messages and the Aumann conjecture

Shoshana Anily, Moshe Haviv

Regular cooperative games: classification and total

balancedness

Multi Criteria optimization - Chair: Moshe Kaspi

ערבה

Yarden Belsky, Yossi Bukchin, Michael Masin

DMA based multi-objective optimization

David Strimling

Multidisciplinary/Multicriteria Design Space

Exploration and Optimization

Yahel David and Nahum Shimkin

Simple Confidence Bounds Algorithms for Coupled

Multi-armed Bandit Problems

15:10-15:25 Break

4

15:25-16:15 Tutorials T1

Nicole Adler, HUJI

OR Research in the field of Aviation מצפה

Iddo Eliazar, HIT & Morrel Cohen, Princeton & Rutgers

The Econophysics of Wealth תמנע

16:20-17:35 Parallel Sessions S3

Simulation (In memory of Reuven Rubinstein) -

Chair: Joseph Kremer מצפה

Nahum Shimkin

Some Applications of Rubinstein's Cross Entropy

Method

Reuven Rubinstein, Ilya Gertsbakh, Yoseph Shpungin,

Radislav Vaisman

Permutational Methods for Performance Analysis

of Stochastic Flow Networks

Slava Vaisman

Counting vertex covers with Monte Carlo

Queueing 2 - Chair: Yoav Kerner ערד

Yuval Nov, Nafna Blanghaps, Gideon Weiss

Sojourn Time Estimation in an M/G/ Queue with

Partial Information

Yonit Barron

A Fluid EOQ model with Markovian Environment

Data Envelopment Analysis - Chair: Zilla Sinuany-

Stern תמנע

Nicole Adler, Nicola Volta, Gianmaria Martini

DEED: a Directional Economic Environmental

Distance function

Zilla Sinuany-Stern, Nir Shakohi, Osnat Cohen

Efficiency of Hospitals Operating Departments

Heavy Scheduling Problems - Chair: Gabi Pinto

ערבה Offer Reshef

Manufacturing environments and scheduling

problems overview

Avraham Mordoch

Using Exepron Software to Resolve Scheduling

Dilemmas in a Multi Projects Management

Environments

Gaby Pinto, Yariv T. Ben-Dov, Gad Rabinowitz

Formulating and Solving a Multi-Mode Resource-

Collaboration and Constrained Scheduling

Problem (MRCCSP)

17:35-17:45 Break

17:45-21:30 Evening Program (bus leaves at 5:50pm)

Tour of Beer-Sheva – Yishai Avital (Beer-Sheva City Hall)

Wild plants for the benefit of agriculture in Israel in the millennium

- Yossi Mizrachi (BGU)

Dinner @ Arabica (downtown)

5

Detailed Program – Monday, 20.5

8:45 Registration

9:30-10:45 Parallel Sessions M1

Strategic Behavior In Queues - Chair: Moshe Haviv

מצפהLiron Ravner, Moshe Haviv

Arrival Times to a Queue With Order Penalties

Binyamin Oz

Equilibrium Service Length Demand in a Single

Server Queue System

Moshe Haviv

Regulating an M/G/1 queue when customers know

their demand

Multi Agent Optimization – Chair: Roie Zivan תמנע

Sofia Amador, Steven Okamoto, Roie Zivan

Combining Fairness and Efficiency in Dynamic

Task Allocation with Spatial and Temporal

Constraints

Raz Nissim, Ronen I. Brafman,

Cost-Optimal Planning by Self-Interested Agents

Zahy Bnaya, Roni Stern, Ariel Felner, Roie Zivan,

Steven Okamoto

Multi-Agent Path Finding for Self Interested

Agents

Combinatorial Optimization - Chair: Liron

Yedidsion ערד

Mark Last, Guy Danon, Shlomo Biderman, Eli Miron

Optimizing a Batch Manufacturing Process

through Interpretable Data Mining Models

Liron Yedidsion, Moshe Kaspi, Dvir Shabtai

Complexity Analysis Of an Assignment Problem

with Controllable Assignment Costs and its

Applications in Scheduling

Nir Halman, Giacomo Nannicini, James Orlin

An Efficient FPTAS for Convex Stochastic

Dynamic Programs

Military OR - Chair: Eilam Gofer נגב

Ben Levav מודל נפגעים מאש ישירה -ר "פגיעות חיילי חי

Eilam Gofer השפעת עומס המשקל על תפקודם של לוחמים רגליים

Yonatan Grinshpun

Deliver or Search? Transportation Tactics in the

Presence of Improvised Explosive Devices

10:45-11:00 Prize Award Ceremony מצפה

11:30-12:55 Plenary Lecture מצפה

Chair and opener: Shoshana Anily, TAU, Speaker: Sergiu Hart, HUJI

12:15-13:05 Tutorial T2

12:15-13:05 Parallel Sessions M2

ORSIS and Mehrez Prize –

Chair: Moshe Haviv מצפה

ORSIS prize winner: Yoel Drori

and Marc Teboulle

A novel approach for analyzing

the performance of first-order

methods

Mehrez prize winner: Ricky

Roet-Green and Refael Hassin

Equilibrium in a two

dimensional queueing game:

When inspecting the queue is

costly

Applied Optimization - Chair:

Ofer Levi ערד

Yaniv Zaks

The Optimal Asset and

Liability Portfolio for a

Financial Institution with

Multiple Lines of Businesses

Ofer Levi, Shaul Ladany

Optimal Sailing Policy

Supply Chain - Chair: Yael

Perlman תמנע

Yael Perlman, Elad Crispil

Shift scheduling of preventive

maintenance for wafer

fabrication plants

Yael Perlman, Yaacov Oz

Reducing shoplifting by

investment in security

12:00-12:15 Break

Shlomo Yitzhaki, HUJI

Gini's Mean Difference offers a response to Leamer's critique נגב

6

13:05-14:10 lunch

סימפוזיון 14:10-15:30

מצפה חקר ביצועים בשירות הציבורי גדי רבינוביץ :מנחה . שלמה מזרחי, עודד מימון, אב-זיו אמיר, שלמה יצחקי: משתתפי הפאנל

15:30-16:15 Tutorials T3

Ron Holzman, Technion

Combinatorial auctions: selling many items at

once מצפה

Avraham Shtub, Technion

Project Planning Monitoring and Control –

Unsolved Problem and Research Opportunities

תמנע

16:15-16:30 Break

16:30-17:45 Parallel Sessions M3

Scheduling 3 - Chair: Liron Yedidsion מצפה

Danny Heremelin, Reuven Bar-Yehuda and Dror

Rawitz

Admission control in line topologies via rectangle

graphs

Yossef Luzon, Yariv Marmor, Eugene Khmelnitsky

Optimal Control of a Flow-Shop Network

Shay Hakim, Michael Masin, Michal Penn,

Asymptotically Optimal Approximation for the

(High) Multiplicity Multi-Skill Scheduling Problem

Game Theory 2 - Chair: Arieh Gavious תמנע

Itai Arieli, Manuel Mueller-Frank

Ram Orzach, Ezra Einy, Ori Haimanko, Aner Sela

Common-Value All-Pay Auctions with Asymmetric

Information

Amitay Kauffmann, Gal Zahavi

FEER Index – Forecasting Extreme Events Risk

Queueing Networks in Production systems - Chair:

Yoav Kerner ערד

Ruth Shilman, Dean Grosbard, Gadi Rabinowitz,

Israel Tirkel

Queueing Network Models for Manufacturing

Systems with Downtimes

Shai Goren, Gad Rabinowitz and Yoav Kerner

A QBD approach for resources allocation

Dean Grosbard

Rapid feasibility check for small size constrained

capacity assignment problems

Transportation 2 - Chair: Hillel Bar-Gera נגב

Uri Yovel, Asaf Levin

Local Search Algorithms for Multiple-Depot

Vehicle Routing and for Multiple Traveling

Salesman Problems with Proved Performance

Guarantees

Michal Blumberg Nitzani, Hillel Bar-Gera

Representing signalized intersections in analytic

dynamic traffic assignment model of transportation

networks.

Sarit Freund and Hillel Bar-Gera

An Optimization Framework for Travel Pattern

Interpretation of Cellular Data

17:45-18:00 Closing Session מצפה

7

Plenary Lectures

Naor Plenary Lecture

Chair and opener: Aharon Ben-Tal, Technion

Speaker: Dorit S. Hochbaum, Berkeley

Combinatorial algorithms for the Markov Random Fields problem and implications for

ranking, clustering, group decision making and image segmentation

One of the classical optimization models for image segmentation is the well known Markov

Random Fields (MRF) model. The MRF problem involves minimizing pairwise-separation and

singleton-deviation terms. This model is shown here to be powerful in representing classical

problems of ranking, group decision making and clustering. The techniques presented are stronger

than continuous techniques used in image segmentation, such as total variations, denoising, level

sets and some classes of Mumford-Shah functionals. This is manifested both in terms of running

time and in terms of quality of solution for the prescribed optimization problem We present the

first known efficient, and flow-based, algorithms for the convex MRF (the non-convex is shown to

be NP-hard). We then discuss the power of the MRF model and algorithms in the context of

aggregate ranking. The aggregate ranking problem is to obtain a ranking that is fair and

representative of the individual decision makers' rankings. We argue here that using cardinal

pairwise comparisons provides several advantages over score-wise or ordinal models. The

aggregate group ranking problem is formalized as the MRF model and is linked to the inverse

equal paths problem. This combinatorial approach is shown to have advantages over other

pairwise-based methods for ranking, such as PageRank and the principal eigenvector technique.

Plenary Lecture

Chair and opener: Shoshana Anily, TAU

Speaker: Sergiu Hart, HUJI

Two(!) Good To Be True

How to sell goods optimally? While the mechanism-design literature has solved this problem

neatly when there is only one good, the multiple goods case turns out to be extremely difficult,

mathematically and conceptually. Much of what is true for one good does not extend to multiple

goods. We will try to explain the difficulties, show what can go wrong, and then present some

universal approximation results. The talk is essentially self-contained; no background in

mechanism design is necessary.

8

Tutorials

T1 Nicole Adler, HUJI

OR Research in the field of Aviation

In this tutorial I will discuss the application of various operational research techniques to questions that

have arisen in the aviation industry, such as how to design a tender for thin airline markets, how to

encourage the aviation industry to consider the environment in their decision making processes and under

which conditions agreements between airline carriers can also increase consumer surplus.

Socially-oriented Flight Scheduling and Fleet Assignment Model with an Application to Norway One of the direct results of air transport liberalization has been the development of publicly supported links

on thin markets in remote regions through a tender process. In this article, we present a flight scheduling

and fleet assignment optimization model to determine a network configuration that minimizes the social

costs of a subsidized air transport system. With the results of the optimization model, a second stage

welfare analysis of the network is carried out that distinguishes between passenger, airline, airport and

government surpluses. The optimization model and subsequent welfare analysis were applied to the

Norwegian Public Service Obligation network, which is currently the largest in Europe. The results

obtained indicate that improvements over the current network can be achieved for all relevant stakeholders

simultaneously, with savings in the order of $1.2 million daily.

The Directional Economic-Environmental Distance function: The case of the global aviation fleet

We develop a directional economic-environmental distance function (DEED) which accounts for the

production of both desirable and undesirable output and the potential for constrained increases in input

utilization from the production of airline services. This research applies the modeling framework to analyze

the potential to reduce noise and airborne pollutants emitted by aircraft-engine combinations given the

current state of aeronautical technology. The global engine-aircraft market is viewed from the regulatory

perspective in order to compare the single environmental and operational efficient frontier to that of the

airline carriers and environmental objectives. The results of DEED are then applied in order to substitute

the fleets serving Schipol, Amsterdam and Arlanda, Stockholm airports in June 2010 with the benchmark

aircraft. The results highlight the inefficiencies of the current airline fleets and that the IPCC values of

externalities are a magnitude of TEN too low to encourage changes in the global fleet hence the need for

government intervention.

Airline agreements: Who are they good for?

The major objective of this research is to develop a model framework that analyzes airline-to-airline

agreements and their impact on both producers and multiple consumer types thus permitting airlines and

governmental authorities to fully evaluate the effects of specific agreement types. Our focal point will be to

utilize a hybrid competitive-cooperative game theoretic model in order to analyze airline agreements, from

competition (no agreement), through bi-lateral or multi-laterals, free sale, hard block, soft block, royalties

and pooling agreements up to anti-trust immunity. We assume that demand for flights depends on

frequency as well as price (i.e. inventory dependent demand as described in the operations management

literature). From a managerial perspective, alliances maximize profits whereas codesharing agreements in

which airlines still compete in frequency and price rank highest from a consumer surplus and social welfare

perspective, which should be of interest to regulators. However, (partial) mergers appear preferable to no

agreement on `thin' markets, in which both demand and profit margins are relatively low. A numerical

analysis demonstrates under asymmetric and uncertain demand that codesharing on parallel links may be

preferable to competitive outcomes for multiple consumer types.

Iddo Eliazar, HIT and Morrel Cohen, Princeton & Rutgers

The Econophysics of Wealth

In this tutorial we amalgamate ideas and concepts from various scientific disciplines – economics,

mathematics, operations research, physics, probability, and statistics – to explore a topic that was pioneered

by the celebrated Italian scientist Vilfredo Pareto: the distribution of wealth in human societies. The tutorial

is split into two parts which are outlined as follows.

9

Part I. Prolog: Gauss law and “mild” randomness, Pareto tails and “wild” randomness § The distribution of

wealth: Log-Normal bulk and Power-Law tails § The Central Limit Theorem cannot explain the distribution

of wealth § Langevin’s equation: potentials, forces, white noise, and equilibria § Ornstein-Uhlenbeck

dynamics: quadratic potentials, linear forces and Gauss equlibria § Shifting from linear to sigmoidal forces,

and from Gauss to Gauss-Exponential equlibria § Shifting from additive to multiplicative dynamics: the

emergence of a phase transition and of the composite Log-Normal–Power-Law statistical structure § The

linear-force limit and its Log-Normal structure § The Heaviside-force limit and its Log-Laplace structure §

Universality of the composite Log-Normal–Power-Law statistical structure.

Part II. Setting: complex “black-box” systems with multiplicative outputs § Entropy: the notion of entropy,

the maximization of entropy, and the notion temperature § The economic interpretation of entropy and its

shadow prices § The coincidence of the Langevin and entropy approaches: micro-scale and macro-scale,

noise and temperature § Reverse-engineering of “black-boxes”: from observed outputs to intrinsic forces §

The shapes of the forces underlying our economies: physical visualization of Adam Smith’s invisible hand

§ Physical segmentation of socioeconomic classes § Physical interpretation of socioeconomic feedback

mechanisms: rich-get-richer and poor-get-poorer § Model or interpretation? The analogue of the Merton-

Black-Scholes formula for option prices: implied volatilities and market temperatures § Epilog: Mandelbrot

and the classification of “mild” and “wild” randomness.

T2 Shlomo Yitzhaki, HUJI

Gini's Mean Difference offers a response to Leamer's critique

Gini's mean difference has decomposition properties that nest the decomposition of the variance as a special

case. By using it one may reveal the implicit assumptions imposed on the data by using the variance. I

argue that some of those implicit assumptions can be traced to be the causes of Leamer's critique. By

requiring the econometrician to report whether those assumptions are violated by the data, we may be able

to offer a response to Leamer's critique. This will reduce the possibility of supplying "empirical proofs"

which in turn may increase the trust in econometric research.

T3 Ron Holzman, Technion

Combinatorial auctions: selling many items at once

In a combinatorial auction, a number of items are being offered for sale to a group of agents whose

valuations for combinations of items may not be additive (i.e., the utility that an agent derives from owning

two of the items need not be the sum of the utilities derived from owning each of them separately).

Examples include spectrum auctions for wireless services, the sale of airport time slots, etc.

The design of such auctions meets with several challenges: obtaining economic efficiency of the allocation

of items, giving the agents the right strategic incentives, and keeping the communication and computation

burden at a feasible level.

The talk will introduce a formal model of combinatorial auctions, and present a few ideas for achieving

some of these goals, while pointing out the tradeoffs involved.

Avraham Shtub, Technion

Project Planning Monitoring and Control – Unsolved Problem and Research Opportunities

Operations research tools for solving Project Scheduling problems are focusing on the work content of the

project (the project scope). Scheduling of project activities under resource and budget constraints is an issue

that attracted the OR community attention for many years. Deterministic and stochastic models and a large

variety of solution algorithms are discussed in the literature and in OR conferences. In this tutorial we

introduce the systems engineering aspect (the product scope) and the need to develop Project Planning

Monitoring and Control methodologies that integrates the project scope and the product scope. Several

problems and the opportunity to develop new models and solution techniques will be presented along with a

teaching environment developed and tested at the Technion. The teaching environment helps students

understand the complexity of the problems and the need to find good solutions.

10

:סימפוזיון בנושא

חקר ביצועים בשירות הציבורי

:רקענצא מהשגרה של הרצאות אקדמיות כדי ; יתקיים סימפוזיון -לאחר ארוחת הצהריים , למאי 71ביום שני

? לדון בשאלה האם רצוי שחוקרי ביצועים ייטלו חלק פעיל יותר בהשבחת השירות הציבורי בישראל

:משתתפי הפאנל ,טיקהשימש עד לאחרונה כראש הלשכה המרכזית לסטטיס, כלכלן –שלמה יצחקי

,התחבורה משרד של ראשיה מדעןה – אב-זיו אמיר ,שותף פעיל למחקרים בשירות הציבורי, א"ת' אונ, המחלקה להנדסת תעשייה –עודד מימון

. חוקר את ועם השירות הציבורי, ג"ב' אונ, ראש המחלקה למנהל ומדיניות ציבורית –שלמה מזרחי

:ינחה את הדיון גדי רבינוביץ

: הדיון יתנהל בארבעה סבבי שאלות :להלן רשימת שאלות טנטטיבית

? בישראל הציבורי בשירות בחקר ביצועים השימוש מידת מה. 9

? האם מומחים אחרים נותנים מענה לצורך זהו, צורך לכךהאם קיים •

?בשירות הציבורימופעלים פחות מאחרים חוקרי ביצועיםהאם •

?בארץ עקב כך חקר ביצועיםשטח או/ו, הציבוריהשירות האם וכיצד נפגם •

? בישראל הציבורי בשירות בחקר ביצועים שימוש המעכבים הגורמים, ומהם, ישנם האם. 0

?בשירות הציבורי מתקיימיםהם והאם ,חקר ביצועיםלפרויקטי הדרושים מהם התנאים •

?הצדדים בכיוון זההאם נעשים מאמצים משני ו, האם המערכת תומכת במחקרים מסוג זה •

האם יש לחוקרים את הרצון והיכולתו, מאיים על המזמין אותו חקר ביצועיםהאם פרויקט • ?לבצעו

?בשיטות חקר ביצועיםהאם יש סכנה בשימוש יתר ו ?האם יש מכשולים מנהליים והסכמיים •

?לכך בהקשר לעשות ראוי ומה האם. 2

?לעניין ביצועיםחוקרי האם אין די בשינוי המודעות של •

?בפרויקטים גדולים חקר ביצועיםלחייב בחוק יש האם •

?בישראל לחקר ביצועיםלהקים לשכה מרכזית כדאי האם •

?בישראל לחקר ביצועים בשירות הציבוריהאם לייחד קרן לאומית •

?יישומי חקר ביצועיםלהרחיב את ההכשרה בכיוון יש האם •

שאלות מהקהל. 4

11

Book of abstracts (grouped by sessions)

Sunday, 19.5

S1

Transportation 1 - Chair: Michal Tzur

Iris Forma, Tal Raviv and Michal Tzur

A Clustering Algorithm for the Static Repositioning Problem in a Bike-Sharing System

These last years a new mode of transportation called Bike-Sharing was implemented in a large number of big cities in

the world. This young kind of mobility deals with a lot of difficulties, as bicycles' maintenance and stations' capacities.

But one major problem of bike-sharing systems is their ability to response to the fluctuating demands for bicycles and

for vacant lockers at each station, which influences directly the service level. Raviv et al. (2012) concentrated their

research on the static repositioning operation, which consists of transferring bicycles from stations with high return

rates to stations with high demand rates using a dedicated fleet. The service level is determined by the inventory at the

stations that is set by the repositioning operation, which consists of decisions on routes that the trucks should follow,

and of decisions of the number of bicycles that should be removed or placed in each station at each visit of the truck.

In this paper, we propose a three-step math heuristic solution to the existing problematic model proposed by Raviv et

al.(2012): first, stations are clustered according to geographic and inventory considerations. Second, the vehicles are

routed through the clusters while inventory decisions are made for each individual station separately. Finally, the

inventory routing problem is solved for all stations, but traversal of the vehicles is allowed only between stations of

the same cluster, or stations that belong to two consecutive clusters, according to the decision made in the previous

step. The first step is solved by a specialized saving heuristic and the last two steps are formulated as Mixed Integer

Programs and solved by a commercial solver. The method is tested on instances of up to 200 stations and three

repositioning vehicles. It is shown to perform better than previously introduced heuristics that are based on mixed

integer programming.

Mor Kaspi, Tal Raviv and Michal Tzur

Parking Reservations in Shared Mobility Systems

Shared mobility systems are increasingly deployed in cities around the world. Such systems allow users to rent a

vehicle (either a car or a bicycle) for a short period and return it in any of the system's stations scattered throughout the

city. There are many advantages to shared mobility systems. Among others, they reduce city traffic congestion and

improve utilization of city land resources, as the need for parking spaces is decreased. Financially, the sharing of

resources results in lower costs per user compared to owning a private vehicle. In addition, users are not troubled

about securing the vehicle when not in use and are not troubled about repairs

The main challenge of shared mobility system operators is to satisfy demands for vehicles (on rent) and for vacant

parking spaces (on return). The demands are typically stochastic, non-stationary and asymmetric processes.

Occasionally, users may find that some stations are empty (no available vehicles) and some are full (no available

parking spaces). In this study we measure the performance of the system by the total excess time spent by users due to

unfulfilled requests for vehicles or parking spaces.

For the first time, we propose using reservation policies in shared-mobility systems and examine their effect. In

particular, we focus on reservations of the parking spaces. In this study we investigate a policy which we refer to as

the Complete-Parking-Reservation (CPR) policy. According to this policy, when a user rents a vehicle he declares his

destination station and a vacant parking space in that station is reserved for him, if one is available. This assures that

when the user reaches his destination he will be able to return his vehicle. If there is no available parking at the

destination the transaction is denied and the user can either abandon the system or try to reserve a parking space in a

different station.

The CPR policy is compared to the base policy, entitled No-Reservation (NR). We prove that if the demand rate is not

too high, the CPR policy always performs better than the NR policy. Furthermore, we demonstrate analytically that the

12

CPR policy outperforms the NR policy whenever the capacity of the system, relative to workload, is large enough to

provide adequate service level. This claim is also verified via extensive simulation study of large realistic systems.

Shoshana Anily, Aharona Pfeffer

The Uncapacitated Dial-a-Ride Problem on a Tree

This research considers the Dial-a-Ride Problem (DaRP) on a tree with a set of objects of different types: A single

vehicle of unlimited capacity is assumed to be positioned at the root of the tree. Each vertex of the tree is associated

with a pair of object types, the object type that is currently held by the vertex and the object type that is needed by the

vertex. We assume that each vertex owns a single unit of its object type and it needs a single unit of an object type.

The problem is assumed to be balanced, i.e., the total supply of each object type equals the total demand of this object

type. The objective is to design a minimum length feasible route that starts and ends at the root of the tree, so that the

vehicle while following the route loads and unloads objects at the vertices in order to satisfy the requirements of all the

vertices.

We start by proving that this version of DaRP is NP-hard. Then tight lower and upper bounds on the optimal solution

are presented. The lower bound is based on a lower bound that we developed for the Minimum Weighted Feedback

Vertex Set Problem. Next, two algorithms for solving the problem are presented. The first is an exact algorithm based

on a dynamic programming formulation. For this sake, we present some structural properties that can be assumed

without loss of generality to be satisfied by an optimal route. The second is a heuristic that determines a feasible route

while following a depth-first scanning of the tree. The heuristic is shown to give near optimal results, via a

computational study.

Continuous Optimization - Chair: Amir Beck

Marc Teboulle Amir Beck

Nonsmooth Convex Minimization: To Smooth or Not to Smooth?

Smoothing is a well-known approach to tackle nonsmooth optimization problems via adequate algorithms applied to

their smoothed counterparts. Within this approach the resulting schemes only provide a $\varepsilon$-optimal solution

to the approximated smoothed problem, which regretfully also depends on a smoothing parameter, but do not find a

$\varepsilon$-optimal solution to the original nonsmooth problem.

In this talk, we prove that independently of the structure of the convex nonsmooth function to be minimized and of a

given fast first order iterative scheme, by solving an adequately smoothed approximation counterpart, the original

nonsmooth problem can be solved with an $O(\varepsilon^{-1})$ efficiency estimate, thus improving by a square root

factor standard nonsmooth algorithms.

Our approach relies on the notion of smoothable functions that we introduce combined with a natural extension of the

Moreau infimal convolution technique and other related convex analytical tools. This allows for clarification and

unification of several issues on the design, analysis, and the potential applications of smoothing methods when

combined with fast first order algorithms, and eventually answer to the question posed in the title!

Ron Shefi, Marc Teboulle

Rate of Convergence Analysis for Proximal-Lagrangian Methods

Augmented Lagrangians based methods have attracted renewed interest recently due to their relevance for solving

large scale convex structured minimization problems arising in many applications. This talk presents two generic

classes of Proximal Lagrangian Methods (PLM) and focuses on their theoretical efficiency. We first show that the

PLM framework is a natural vehicle to build and analyze novel schemes, and is at the root of many past and recent

algorithmic variants suggested in the literature. We then prove various types of global convergence rate results for the

two proposed generic classes. Our approach relies on elementary convex analytic arguments and allows revisiting

seemingly different algorithms for which new and refined rate of convergence results are established within a unifying

framework.

13

Shoham Sabach , Jerome Bolte, and Marc Teboulle

A Simple Algorithm for Nonconvex and Nonsmooth Minimization Problems

We introduce a new algorithm for a broad class of nonconvex and nonsmooth problems. It relies on an elementary

mixture of first order methods and data information. We outline a self-contained convergence analysis framework

describing the main tools and methodology to prove that the sequence generated by the proposed scheme globally

converge to a critical point. A by-product of our framework also shows that our results are new even in the convex

setting. The resulting scheme involves elementary iterations and is particularly adequate for solving many problems

arising in a wide variety of fundamental applications.

Scheduling 1 - Chair: Gur Mosheiov

Shlomo Karhi, Dvir Shabtay and Daniel Oron

Multipurpose Machine Scheduling with Rejection and Identical Job Processing Times

We study a scheduling problem where there are n jobs and m uniform machines. All jobs have the same processing

time and each job can either be rejected or be processed on one machine among a predefined subset of the m

machines. A solution to our problem is given by partitioning the jobs into two subsets, A and A, which are the set of

accepted and the set of rejected jobs, respectively, and by scheduling the jobs in set A on the m machines. We evaluate

the quality of a solution by two criteria. The first, F₁ , can be any regular scheduling criterion and the second, F₂ , is

the total rejection cost. We consider two possible types of regular scheduling criterion, the first is a maximization

criterion and the second is a summation criterion. For each criterion type we study four different problem variations.

We prove that all four variations are solvable in polynomial time for any maximization type of a regular scheduling

criterion while for a summation type of a regular scheduling criterion only one of the four problem variations is

solvable in polynomial time. We provide a pseudo-polynomial time algorithm that solves various variations of the NP-

hard problems as well as a polynomial time algorithm that solves various other special cases.

Enrique (Tzvi) Gerstl, Gur Mosheiov

The optimal number of used machines in a two-stage flow shop scheduling problem

We study practical scheduling problems with a major decision referring to the number of machines to be used. We

focus on a two-stage flowshop, where each job is processed on the first (critical) machine, and then continues to one of

the second-stage machines. Jobs are assumed to have identical processing times, and are processed in batches. A setup

time is required when starting a new batch. We consider two objective functions: minimum makespan, and minimum

flowtime. In both cases, a closed form expression for the optimal number of machines to be used is introduced, and a

unique and unusual sequence of decreasing batch sizes is shown to be optimal.

Baruch Mor, Gur Mosheiov

Polynomial time solutions for scheduling problems on a proportionate flowshop with two competing agents

In scheduling problems with two competing agents, each one of the agents has his own set of jobs and his own

objective function, but both share the same processor. The goal is to minimize the value of the objective function of

one agent, subject to an upper bound on the value of the objective function of the second agent. In this paper we study

two-agent scheduling problems on a proportionate flowshop. Three objective functions of the first agent are

considered: minimum maximum cost of all the jobs, minimum total completion time, and minimum number of tardy

jobs. For the second agent, an upper bound on the maximum allowable cost is assumed. We introduce efficient

polynomial time solution algorithms for all cases.

14

S2

Queueing 1 - Chair: Uri Yechiali

Efrat Perel, Kostia Avrachnkov, Uri Yechiali

The Traffic Policeman Problem: A Two-Queue System with Switching Policy Depending on the Queue which is

NOT Being Served

A policeman regulates two streams of traffic crossing an intersection. When one direction has the right-of-way and the

accumulating number of cars in the other direction reaches a threshold, the right-of-way is transferred to the latter

direction, and vice versa. Consequently, we study a polling-type system of two non-identical and separate finite-buffer

Markovian queues served by a single server whose switching policy is based on a threshold determined by the state of

the queue which is not being served. We provide a probabilistic analysis of this system and investigate the effect of

buffer sizes and arrival rates, as well as service rates, on the system's performance.

Nir Perel, Uri Yechiali

The Israeli Queue with Priorities

The 'Israeli Queue' model was first introduced [1] when studying an N-queue, single server polling system with

unlimited-size batch-service governed by the following server's visit-order rule: After completion of a visit at a queue,

the next queue to be served is the one where its first customer in line has been waiting for the longest time. The model

was extended in [2].

In this presentation we consider a 2-class, single-server, preemptive priority queueing model in which the high priority

customers form a classical M/M/1 queue, while the low priority customers form the Israeli Queue with at most N

different groups. We present a probabilistic analysis of the model using both Probability Generating Functions and

Matrix Geometric methods and calculate various performance measures. Special cases are analyzed and numerical

results are presented.'

[1] O. Boxma, Jan v.d. Wal and U. Yechiali, Polling with Batch Service. Stochastic Models 24, 604-625 (2008).

[2] N. Perel and U. Yechiali, The Israeli Queue with Finite and Infinite Number of Families. Submitted

Shlomi Reuveni, Iddo Eliazar, Uri Yechiali

Occupation Probabilities and Catalan Numbers in the Asymmetric Inclusion Process

The Asymmetric Inclusion Process (ASIP) is a system of n Markovian queues in tandem, each with unbounded

capacity and batch service [1]. That is, when service is completed at queue k, all particles present at that queue move

simultaneously to queue k+1, or out of the system when k=n. Numerical simulations demonstrated that the probability

that site k is occupied (by one or more particles) is inversely proportional to the square root of k [2]. Consequently,

limit laws for various variables (e.g. busy period, draining time, etc.) have been established [3].

In this presentation we show that occupation probabilities in the ASIP obey a discrete two-dimensional boundary

value problem. Solving this problem we find explicit expressions for the probability that site k is occupied by l

particles (l=0,1,2,…). Catalan's numbers are shown to naturally arise in the context of these occupation probabilities.

[1] S. Reuveni, I. Eliazar and U. Yechiali, Asymmetric Inclusion Process. Physical Review E 84, 041101, (2011).

[2] S. Reuveni, I. Eliazar and U. Yechiali, The Asymmetric Inclusion Process: A Showcase of Complexity. Physical

Review Letters 109, 020603, (2012).

[3] S. Reuveni, I. Eliazar and U. Yechiali, Limit Laws for the Asymmetric Inclusion Process. Physical Review E 86,

061133 (2012).

15

Scheduling 2 - Chair: Michal Penn

Ilan Cohen, Yossi Azar, Seny Kamara, Bruce Shepherd

Tight Bounds for Online Vector Bin Packing

Recently the usage of cloud computing is increasingly wide spreading. Therefore, we study the problem of assigning

tasks with d-dimensional demand (CPU, Memory, GPU, etc.) to identical machines (computers) without overloading

in any machine d-dimensional capacity constraint. Our goal is to minimize the number of machines opened by the

algorithm. This problem reduced to the vector bin packing problem (VBP).

We study the online VBP, where the tasks arrive in an online fashion and need to be assigned immediately. We

present a lower bound of $\Omega(d^{\frac{1}{B}-\epsilon}$ where $B$ is the minimum ratio between the

capacity of machine and any task demand (in any coordinate). Additionally, we discuss almost-matching upper bound

results for general values of $B$. We show an upper bound whose exponent is additively ``shifted by 1" from the

lower bound exponent.

Sagi Halleli, Ephraim Korach and Michal Stern

למציאת מרכזים של קבוצות ושימושים OS-שיפור בשיטת ה

הינה נתון היפרגרף עם קבוצת קודקודים ואוסף של תתי קבוצות ( OS - בקיצור) Optimal Stars – clustering tree -בעיית המקבוצת הקודקודים ונתון גרף מלא על כל הקודקודים ומשקל על כל צלע רוצים למצוא עץ פורש כך שכל קבוצת קודקודים תשרה

ג חסם קומבינטורי להסתברות פיזיבילי לבעייה ונציג אלגוריתם פתרון ונצי ,בעץ כוכב וכך שסך כל המשקל של העץ יהיה מינימום .קיום

כדי להגדיל את ההסתברות למציאת פתרון פיזיבילי נרחיב את השיטה לשיטה שמתגברת על הבעיתיות של יצירת מעגלים בשיטה א "אנו משתמשים בשיטה למציאת מרכזיות של חברות בבורסה במדד ת. הקודמת ובכך נגדיל את ההסתברות לקיום פתרון פיזיבילי

011.

Vladimir Lipets, Eitan Israeli, Zahar Chikishev, Dvora Lopez, Avraham Har-Paz

Crew Assignment and Pairing Optimization System for El Al

CAPOS (Crew Assignment and Pairing Optimization System) is a robust, powerful, user-friendly system for crew

pairing and assignment optimization, designed to meet the rigorous and dynamic requirements of commercial airlines.

CAPOS is the result of intensive collaborative research and development between IBM Israel Software Labs and El Al

Israel Airlines Ltd.. Finding optimal assignments is especially complicated, since demands for different activities and

availability of human resources vary over time. Added to the complexity are such factors as distinct types of crew

members, night and weekend flights, high rates of attrition, and changes in skill availability.

CAPOS allows the user to handle crew preferences and availability constraints, and make sure business rules such as

fairness are maintained when shifts are assigned – and all this dynamically, without compromising business

objectives, labor regulations, or operational demands. CAPOS provides optimal results quickly and easily - solving the

problems of optimal pairing creation and crew member assignments. With a set of hard constraints, the system

balances hundreds of dynamic rules and criteria defined by the user, thereby minimizing costs for management and

operation, and improving satisfaction.

Game Theory 1 - Chair: Ella Segev

Oscar Volij, Casilda Lasso de la Vega

Segregation, Informativeness and Lorenz domination

A city consists of a population of two ethnic groups distributed across neighborhoods. The more different the two

groups are distributed across the various neighborhoods, the more segregated the city is. For any city, one can build its

Lorenz segregation curve and then partially order the set of cities accordingly. Similarly, for any city one can build an

associated experiment and partially order the cities according to their informativeness. We show that the partial orders

based on Lorenz segregation curves and on informativeness of the experiment coincide.

16

Ro'i Zultan

Timing of messages and the Aumann conjecture

The Aumann (1990) conjecture states that cheap-talk messages do not necessarily help to coordinate on efficient Nash

equilibria. In an experimental test of Aumann’s conjecture, Charness (2000) found that cheap-talk messages facilitate

coordination when they precede the action, but not when they follow the action. Standard game-theoretical modeling

abstracts from this timing effect, and therefore cannot account for it. To allow for a formal analysis of the timing

effect, I study the sequential equilibria of the signaling game in which the sender is modeled as comprising two selves:

an acting self and a signaling self. I interpret Aumann’s argument in this context to imply that all of the equilibria in

this game are ‘babbling’ equilibria, in which the message conveys no information and does not affect the behavior of

the receiver. Using this framework, I show that a fully communicative equilibrium exists—only if the message

precedes the action but not when the message follows the action. In the latter case, no information is transmitted in any

equilibrium. This result provides a game-theoretical explanation for the puzzling experimental results obtained by

Charness (2000). I discuss other explanations for this timing-of-message effect and their relationship to the current

analysis.

Shoshana Anily, Moshe Haviv

Regular cooperative games: classification and total balancedness

The research on cooperative games with transferrable utility in supply chains and service systems is growing as a

result of increasing collaboration among partners in the association, and the need for fair cost allocation schemes. In

this talk we present a new framework that fits many cooperative games in operations management, where each player

is fully characterized by a vector of quantitative properties, representing his/her contribution to any coalition he/she

may join, and the cost of a coalition is a closed algebraic expression of the vectors of properties of the coalition's

members. In such games the characteristic function values of different coalitions are strongly related, as all are

obtained by applying the same algebraic expression on the input of vectors of properties associated with each

coalition. We call such games regular games. We discuss some new theoretical results that hold within the proposed

framework, and we provide new conditions for verification of total balancedness of such games. A few examples in

the areas of retailing, service management and facility location will be presented.

Multi Criteria optimization - Chair: Moshe Kaspi

Yarden Belsky, Yossi Bukchin, Michael Masin

DMA based multi-objective optimization

In Multi-Objective Optimization multiple conflicting objective functions are optimized simultaneously subject to a set

of constraints. In this research, we first provide a formal problem definition where the Pareto-optimal solutions and

efficient frontier approach replace the notion of optimal solution in single-objective optimization. Then, we investigate

an interaction with the decision maker (DM), and provide a method for finding a suitable solution for the decision

maker, based on DMA (a method for finding a diverse representation of non-dominated solutions on the efficient

frontier). Different types of decision makers are considered including decision maker's risk attitude, feedback

inaccuracies and resolution.

In the proposed approach, decision makers are required to rank a limited number of diversified non-dominated

solutions, and the algorithm, in return, provides estimated parameters of the unknown utility function, and additional

potential solutions, based on the above ranking. Finally, a wide empirical experiment verifies the performance of the

proposed algorithm for different types of decision makers and problem types. In all experiments the algorithm finds a

solution that is just few percent far from the optimal solution of the actual decision maker's utility function, while on

average the distance is less than 1%..

David Strimling

Multidisciplinary/Multicriteria Design Space Exploration and Optimization

Multidisciplinary models are system level representation of products that require multiple engineering

disciplines for design. These multiple engineering disciplines could include Finite Element Analysis (FEA),

Computational Fluid Dynamics (CFD) and other Computer Aided Engineering (CAE) or Computer Aided Design

(CAD) tools to represent the system functionality. A multidisciplinary model precludes sub-optimal designs by

17

integrating these multidisciplinary engineering analysis tools into a system level model that allows integrated analysis

at the system rather than subsystem level. Analysis of multidisciplinary systems uses a multidisciplinary, system

level model to understand the design space and seek optimum solutions that satisfy customer requirements.

This multidisciplinary system level model enables system level design space exploration and optimization to exploit

all of the interactions between the disparate engineering disciplines required to represent the details of the product

being designed –

1) Design space exploration will isolate critical design parameters and feasible regions of the system

level design space.

2) Multicriteria optimization will generate an efficient frontier of solutions at the system level focusing

on the critical design parameters and feasible regions of interest discovered during the design space

exploration.

This paper will provide an introduction to the concepts and processes of multidisciplinary/multicriteria design

space exploration and optimization. The paper will introduce the MEGA process;

Yahel David and Nahum Shimkin

Simple Confidence Bounds Algorithms for Coupled Multi-armed Bandit Problems

The classical framework of the statistical multi-armed bandit problem assumes independent parameterization of the

mean rewards of different arms. We consider here the extended model, where the parameters of the different arms may

be coupled, so that the obtained rewards at one arm can provide meaningful statistical information on the expected

reward at other arms. We propose simple low-regret control laws for this model, extending the UCB1 algorithm of

Auer et al. (2002) to this case. We provide logarithmic upper bounds on the regret, and show that the effective use of

cross-arm coupling can reduce the logarithmic coefficient of increase of the regret for certain arms, while reducing

logarithmically-increasing regret to a finite one for other arms.

S3

Simulation (In memory of Reuven Rubinstein) - Chair: Joseph Kremer

Nahum Shimkin

Some Applications of Rubinstein's Cross Entropy Method

The celebrated Cross Entropy method has been introduced by Reuven Rubinstein in the late 1990's, first for rare event

simulation, and then extended as a powerful heuristic for combinatorial optimization. It has since found numerous

applications in a variety of challenging optimization problems. In this talk I will briefly survey some applications in

which I had part, from multiple-target tracking to reinforcement learning and basis function adaptation.

Reuven Rubinstein, Ilya Gertsbakh, Yoseph Shpungin, Radislav Vaisman

Permutational Methods for Performance Analysis of Stochastic Flow Networks

In this paper we show how the permutation Monte Carlo method, originally developed for reliability networks, can be

successfully adapted for stochastic flow networks, and in particular for estimation of the probability that the maximal

flow in such a network is above some fixed level, called the threshold. A stochastic flow network is defined as one,

where the edges are subject to random failures. A failed edge is assumed to be erased (broken) and, thus, not able to

deliver any flow. We consider two models; one where the edges fail with the same failure probability and another

where they fail with different failure probabilities. For each model we construct a different algorithm for estimation of

the desired probability; in the former case it is based on the well known notion of the D-spectrum and in the later one -

on the permutational Monte Carlo.

Slava Vaisman

Counting vertex covers with Monte Carlo

In graph theory, a vertex cover of a graph is a set of vertices such that each edge of the graph is incident to at least one

vertex of the set. We are interested in counting all vertex covers in a graph. The area of counting, and in particular the

18

definition of #P complete class introduced by Valiant in 1979 received much attention in the Computer Science

community. For example, Karp and Lubby introduced a FPRAS (A fully polynomial randomized approximation

scheme) for counting the solutions of DNF satisfiability formula. Similar results were obtained for the Knapsack and

Permanent problems. On the other hand, there are many 'negative' results. Dyer, Frieze and Jerrum showed that there

is no FPRAS for #IS (Independent Set) if the maximum degree of the graph is 25 unless RP = NP. Counting the

number of vertex covers remains hard even when restricted to planar bipartite graphs of bounded degree, or regular

graphs of constant degree.

We propose a Sequential Importance Sampling procedure for counting the number of vertex covers in a graph. In spite

of the fact that counting algorithms based on Importance Sampling seems to perform very well in practice, it is also

known that their performance depends heavily on the closeness of the proposal distribution to the uniform one. Our

algorithm introduces probabilistic relaxation technique combined with Dynamic Programming, in order to obtain

efficient estimate of this distribution. At each step, the procedure decides whether to include the given vertex in the

cover set or not. The decision is based on the approximated number of covers in each case. During the algorithm

execution, we guarantee that the sequential procedure always produces a valid vertex cover while saving computation

effort. Moreover, the algorithm can supply a probabilistic lower bound that is calculated online.

Queueing 2 - Chair: Yoav Kerner

Yuval Nov, Nafna Blanghaps, Gideon Weiss

Sojourn Time Estimation in an M/G/ Queue with Partial Information

We propose an estimator for the CDF G of the sojourn time in a steady-state M/G/ queueing system, when the

available data consists of the arrival and departure epochs alone, without knowing which arrival corresponds to which

departure. The estimator is a generalization of an estimator proposed by Brown in 1970, and is based on a functional

relationship between G and the distribution of the time between a departure and the rth latest arrival preceding it. The

estimator is shown to outperform Brown's estimator, especially when the system is heavily loaded.

Yonit Barron

A Fluid EOQ model with Markovian Environment

We consider a single bu¤er fluid system in which the rate of change of the fluid is determined by the current state of a

background stochastic process called environment. When the fluid level hits zero, it can undergo an instantaneous

transition to state i 2 f1; :; ng and jumps to a predetermined positive level qi. The costs (purchasing and holding costs)

are modulated by the state at the order epoch time. Using matrix analytic approach, fluid flow analysis and martingales

we develop methods to compute the expected discounted order and holding costs and the long run average costs.

Data Envelopment Analysis - Chair: Zilla Sinuany-Stern

Nicole Adler, Nicola Volta, Gianmaria Martini

DEED: a Directional Economic Environmental Distance function

We propose a new data envelopment analysis model in order to compute efficiency from an eco-environmental point

of view. The directional, eco-efficient distance (DEED) function model includes undesirable as well as standard

variables hence is suitable for multiple industrial applications. Moreover, the model designs a single frontier which

permits a constrained increase in inputs where relevant. In addition, given the prices of inputs, the objective function

estimates the maximum monetary savings potential for inefficient decision making units. We show that the DEED

function improves on existing models in the literature by substantially reducing negative externalities within a budget

constraint and utilizing only efficient decision making units as benchmarks.

Zilla Sinuany-Stern, Nir Shakohi, Osnat Cohen

Efficiency of Hospitals Operating Departments

In this study we measured the efficiency of 24 operating department in Soroka Hospital, in Beer-Sheva, Israel. Data

Envelopment Analysis (DEA) was used with 5 inputs and 3 outputs over 2008 and 2009. The 5 inputs used are: 1.the

average time between consecutive operations, 2. the no. of hours allocated to the department, 3. the number of

19

spillover hours beyond the daily shift, 4. the no. of repeated operations, and 5. the average hour when the first

operation starts in the daily shift. The 3 outputs used are: 1. income, 2. average hour of ending the daily operations,

and 3. total no. of operating hours. There are three operating sites in the hospital serving 24 operating departments.

There was no significant difference among the three sites efficiency. There was no difference between the efficiency

of the two years considered (2008 and 2009). Looking at the operating rooms utilization (the old measure of

effectiveness of operating rooms, we found that it differs significantly from the DEA rating. In the input oriented

model we used input no. 1 required the highest percentage of reduction (43%), input no 3 was the next (23%). All the

other inputs required 5% reduction.

Heavy Scheduling Problems - Chair: Gabi Pinto

Offer Reshef

Manufacturing environments and scheduling problems overview

This lecture provides an overview to some industry manufacturing environments scheduling problems. Starting with a

brief presentation of OPTUM-Dispatcher, OPTUM-IES scheduling and dispatching platform followed by a

presentation f a few manufacturing environments with their unique problems, issues & goals.

For each environment the lecture will review:

Initial dispatching methods in use and management planning capabilities.

Main characteristics, scheduling/dispatching issues.

Facility goals and management focus.

Mathematical problem overview

Implementation highlights

Improvements achieved and the new scheduling and dispatching method.

We will finalize with the view of the most critical information required as an outcome of the scheduling to the

planners & production manager

Keywords: dispatching; scheduling; manufacturing environment.

Avraham Mordoch

Using Exepron Software to Resolve Scheduling Dilemmas in a Multi Projects Management Environments

Every project manager is well aware of the dilemma he many times has in prioritization between tasks when the

resource needed is the execution of these tasks is the same resource. The very popular method of the Critical Path

cannot provide the project managers with the right direction in resolving this dilemma since Critical Path has an

almost hidden assumption that the availability of the resources is unlimited. Unlike the Critical Path, the Critical Chain

methodology, which is the derivative of the Theory of Constraints (TOC), identifies the chain of tasks with sequence

of both precedence- and resource-dependent tasks that prevent a project from being completed in a shorter time, given

finite resources. The Exepron software, which is only about a year in the market, was developed to support the Critical

Chain methodology.

Exepron is a cloud based, real time, planning and execution tool that provides the right planning based on the Critical

Chain methodology not only during the planning phase of the project but, not less important, during the execution

phase also. Every day that passes, the software calculates the penetration into the different buffers of the project and

provides the project manager with the sense of urgency of the different, still uncompleted, tasks and as a result the

project managers have the right priority for all tasks of the project.

The problem is, of course, getting more complicated in the very common multi-projects environment, when the same

resources are shared between projects in the same group (division or department). In this case the first issue is how to

significantly decrease the high probability that delays in one project, not necessarily a strategic one, are causing delays

in many of the other projects in the same group, sharing the same resources. The method to avoid these delays, which

is a part of the Critical Chain, is called Drum and is based on the following steps:

a) Strategic prioritization between the projects

b) Identification the most loaded resource (the Drum) across all participating projects in the horizon of time

between today and the end of the last task of all the projects

c) Synchronization between the projects taking into account the priority (step A), the load on the Drum (step B)

and a capacity buffer of the Drum resource

20

The result is a planned delay of some of the projects but with realistic planned timetable for all projects vs. unrealistic

planning with no delays in the planning phase but by far more delays in the execution phase of the participating

projects.

Gaby Pinto, Yariv T. Ben-Dov, Gad Rabinowitz

Formulating and Solving a Multi-Mode Resource-Collaboration and Constrained Scheduling Problem

(MRCCSP)

The main motivation of this study is to provide for the first time a formulation and solution for a class of production

scheduling problems (as in cluster tools) characterized mainly by resource collaboration to perform an operation,

while allowing batches and considering alternative production methods. We develop a formulation for the new

problem and term it a multiple mode per operation, resource collaboration, and constrained scheduling problem

(MRCCSP). Some of the important new characteristics we consider are: multiple products (families); Multiple orders

(jobs) per family; Precedence restrictions among the operations that constitute a job; Alternative modes for the

performance of an operation (each of which needs a set of collaborating resources) may be defined; Complementary

and exclusive restrictions between operation-modes; Batch production is allowed; and setup times may depend on

sequence and batch-size. The objective of the MRCCSP is to minimize makespan. We formulate the MRCCSP as a

mixed integer linear programming model, and acknowledging the considerable size of the monolithic formulation

required, we prescribe a specific method to achieve size reduction. A customized branch and bound (B&B) algorithm

for optimally solving this problem is proposed and examined experimentally. Finally, we developed a genetic

algorithm (GA) and compared the effectiveness and runtime of the GA versus a B&B algorithm. The results

demonstrate that the GA reaches optimality in most cases, and its runtime was much less sensitive to the size of the

problem instance.

Keywords: resource-sharing and scheduling problem; resource constrained scheduling problem; mixed integer linear

programming; branch and bound algorithm.

21

Monday 20.5

M1

Strategic Behavior In Queues - Chair: Moshe Haviv

Liron Ravner, Moshe Haviv

Arrival Times to a Queue With Order Penalties

Customers are often faced with a choice of when to arrive to a congested queue with some desired service at the end.

Suppose the server operates for a certain time interval , and customers are served according to their arrival order. We

study a model where the customers incur not only congestion (waiting) costs but also penalties for their index of

arrival. Arriving before other customers is desirable when the value of service decreases with every admitted

customer. Examples of such scenarios are arriving at a concert or a bus with unmarked seats and going to lunch in a

cafeteria. We provide game theoretic analysis of such queueing systems, specifically we characterize the arrival

process which constitutes a symmetric Nash equilibrium.

Binyamin Oz

Equilibrium Service Length Demand in a Single Server Queue System

In this talk we propose a model of an unobservable M/?/1 queue where costumers chooses their service length

strategically in order the maximize their own welfare. We assume a quasilinear utility function with increasing and

concave utility from the service length and linear waiting costs. We assume that the costumers differ from each other

by their evaluation of the service, which is random and private information. We show that in a symmetric equilibrium,

the joining strategy will be based on threshold of the service evaluation, and the service length chosen by the

costumers will be increasing function of their service evaluation. Next, we find the socially optimal strategies and

discuss various ways of regulation.

Moshe Haviv

Regulating an M/G/1 queue when customers know their demand

Selfish customers do not necessarily join a queue at a socially optimal rate. Hence, queueing systems may call for

regulation. For customers in an M/G/1 unobservable (not necessarily FCFS) M/G/1 queue and homogeneous with

respect to waiting costs and service rewards, we show how queueing systems can be regulated by imposing entry,

holding, or service fees in the case where customers know their service requirements. We start with a unified

approach, assuming minimal assumptions on the waiting functions, and state the socially optimal fees.

We show that customers are always worse off under a flat entry free in comparison with holding and service fees.

As for holding vs. service fees, the answer depends on the queueing regime and/or the service length. For example,

under FCFS, service fees are preferred by all. Details are given on some common service regimes. For the sake of

completeness we also review the case where customers know only their distribution, but not its requirement.

Multi Agent Optimization – Chair: Roie Zivan

Sofia Amador, Steven Okamoto, Roie Zivan

Combining Fairness and Efficiency in Dynamic Task Allocation with Spatial and Temporal Constraints

Realistic multiagent team applications often feature distributed dynamic environments with soft deadlines that

penalize late execution of tasks. This puts a premium on quickly allocating tasks to agents, but finding the optimal

allocation is NP-hard due to temporal and spatial constraints that require tasks to be executed sequentially by agents.

We propose a novel task allocation algorithm that allows tasks to be easily sequenced to yield high quality solutions

by finding allocations that are fair (envyfree), balancing the load and sharing important tasks between agents, and

efficient (Pareto optimal).We compute such allocations in polynomial time using a Fisher market with agents as

buyers and tasks as goods, then sequence the allocations by maximizing utility at each step. We empirically compare

22

our algorithm to two state-of-the-art incomplete methods on synthetic problems and on realistic law enforcement

problems inspired by real police logs. The results show a clear advantage for our algorithm in measures commonly

used by law enforcement authorities.

Raz Nissim, Ronen I. Brafman,

Cost-Optimal Planning by Self-Interested Agents

As our world becomes better connected and autonomous agents no longer appear to be science fiction, a natural need

arises for enabling groups of selfish agents to cooperate in generating plans for diverse tasks that none of them can

perform alone in a cost-effective manner. While most work on planning for/by selfish agents revolves around finding

stable solutions (e.g., Nash Equilibrium), this work combines techniques from mechanism design with a recently

introduced method for distributed planning, in order to find cost optimal (and, thus, social welfare maximizing)

solutions. Based on the Vickrey-Clarke-Groves mechanisms, we present both a centralized, and a privacy-preserving

distributed mechanism.

Zahy Bnaya, Roni Stern, Ariel Felner, Roie Zivan, Steven Okamoto

Multi-Agent Path Finding for Self Interested Agents

Multi-agent pathfinding (MAPF) deals with planning paths for individual agents such that a global goal function (e.g.,

the sum of costs) is minimized, while avoiding collisions between agents. Previous work proposed centralized or fully

cooperative decentralized algorithms assuming that agents follow paths assigned to them. However, when agents are

self-interested, they are not expected to follow paths unless they consider them to be their most beneficial option. In

his paper we propose the use of taxation schemes to implicitly coordinate self-interested agents in MAPF. We propose

several taxation schemes and compare them experimentally. We show that intelligent taxation schemes can result in a

lower total cost than the non coordinated scheme even if we take into consideration both the costs of the original graph

and the taxes paid by agents.

Combinatorial Optimization - Chair: Liron Yedidsion

Mark Last, Guy Danon, Shlomo Biderman, Eli Miron

Optimizing a Batch Manufacturing Process through Interpretable Data Mining Models

In this talk, we present a data mining based methodology for optimizing the outcome of a batch manufacturing

process. Predictive data mining techniques are applied to a multi-year set of manufacturing data with the purpose of

reducing the variation of a crystal manufacturing process, which suffers from frequent fluctuations of the average

outgoing yield. Our study is focused on specific defects that are the most common causes for scraping a manufactured

crystal. A set of probabilistic rules explaining the likelihood of each defect as a function of interaction between the

controllable variables are induced using the Single-Target and the Multi-Target Information Network algorithms. The

rules clearly define the worst and the best conditions for the manufacturing process, also providing a complete

explanation of all major fluctuations in the outgoing quality observed over the recent years. In addition, we show that

an early detection of nearly the same predictive model was possible almost two years before the end of the data

collection period, which could save many of the flawed crystals. The paper provides a detailed description of the

optimization process, including the decisions taken at various stages and their outcomes. Conclusions applicable to

similar engineering tasks are also outlined.

The talk is based on the paper published in the Journal of Intelligent Manufacturing.

Liron Yedidsion, Moshe Kaspi, Dvir Shabtai

Complexity Analysis Of an Assignment Problem with Controllable Assignment Costs and its Applications in

Scheduling

We extend the classical linear assignment problem to the case where the cost of assigning agent j to task i is a

multiplication of task i's cost parameter by a cost function of agent j. The cost function of agent j is a linear function of

the amount of resource allocated to the agent. A solution for our assignment problem is defined by the assignment of

agents to tasks and by a resource allocation to each agent. The quality of a solution is measured by two criteria. The

first criterion is the total assignment cost and the second is the total weighted resource consumption. We address these

criteria via four different problem variations. We prove that our assignment problem is NP-hard for three of the four

23

variations even if all the resource consumption weights are equal. However, and somewhat surprisingly, we find that

the fourth variation is solvable in polynomial time. In addition, we find that our assignment problem is equivalent to a

large set of important scheduling problems whose complexity has heretofore been an open question for three of the

four variations.

Nir Halman, Giacomo Nannicini, James Orlin

An Efficient FPTAS for Convex Stochastic Dynamic Programs

We propose an efficient Fully Polynomial-Time Approximation Scheme (FPTAS) for stochastic convex dynamic

programs using the technique of K-approximation sets and functions introduced by Halman et al. This paper deals

with the convex case only, and it has the following contributions: First, we improve on the worst-case running time

given by Halman et al. Second, we design an FPTAS with excellent practical performance, and show that it is faster

than an exact algorithm even for small problems instances and small approximation factors, becoming orders of

magnitude faster as the problem size increases. Third, we show that with careful algorithm design, the errors

introduced by floating point computations can be bounded, so that we can provide a guarantee on the approximation

factor over an exact infinite-precision solution. Our computational evaluation is based on randomly generated problem

instances coming from applications in supply chain management and finance.

Military OR - Chair: Eilam Gofer

Ben Levav

מודל נפגעים מאש ישירה -ר "ת חיילי חיפגיעו

הראשונה היא אפקט הדיכוי אשר מנחית את יכולות הלוחמים : האש בקרב היבשה משפיעה על הכוחות הלוחמים בשתי דרכים שונות, העתידיתנפגעים בכוח לוחם משפיעים על יכולתו לעמוד במשימתו הנוכחית או . לפרקי זמן שונים והשני הוא הנפגעים שסופג הכוח

.מחייבים תשומת לב והקצאת כוחות לשם פינוי וטיפול ופוגעים במורל המסגרת

לרוב זו אש , המצויה בקשר עין עם היורה, אש הנורית לעבר מטרה -ואש ישירה ( ארטילרית)אנו מבדילים בין שני סוגי אש עקיפה אש ישירה . ל ומקלעים"נק -כזו הנורית מנשק קליעי ר האש הישירה הרלוונטית היא"בהקשר של היפגעות כוח חי. שטוחת מסלול

. עם זאת היא שכיחה יותר בשלבי הלחימה אל היעד ובלחימה על היעד. ר בכל שלבי ההתקפה"יכולה להיות מופעלת כנגד כוח חי

אש עקיפה מוגבלת מכיוון שיכולת התגובה שלו כנגד , זאת. כוח שתוקף רגלית יגיב באופן שונה לאש ישירה מאשר כלפי אש עקיפהלרוב . ר עם אש ישירה גדול יותר"מרחב אפשרויות ההתמודדות של כוח חי, לעומת זאת. מאוד וכוללת בעיקר צעדי מניעה והתמגנות

. אל עבר מקור הירי על מנת לגרום לשיתוקו ולפגיעה בו, תוך כדי הפעלת אש ישירה משל עצמו, ישאף הכוח התוקף להתקדם .צדדי-בעוד תרחיש הפעלת אש ישירה מוגדר כדו, צדדי-ר תוקף כחד"גדיר תרחיש הפעלת אש עקיפה כנגד כוח חיניתן לה, מתודולוגית

הוא מתאר את פעולותיהם של הכוחות . מודל הנפגעים מאש ישירה הוא מודל מתמטי הסתברותי הממומש באמצעות סימולציה.על התפלגות מספר הנפגעים בקרב השונים בשלבי הקרב השונים באמצעות כלים מתמטיים ומסיק מכך

Eilam Gofer

השפעת עומס המשקל על תפקודם של לוחמים רגליים

הנדרש לא רק להפעיל , ר הוא האדם הלוחם"המוקד בחי. אינו מזוהה עם מערכת נשק עיקרית, בניגוד לחילות אחרים, חיל הרגליםיכולת הלוחמים לנוע רגלית עם משקל . מזון ומים, ציוד אישי, תחמושת, מגוון גדול של מערכות נשק אלא גם לשאת אמצעי לחימה

לנסח ולקבוע אילוצי , לזהות, ל בפרט"ובצה, לפיכך נעשה מאמץ רב בצבאות העולם. ר"של החימהווה את אחד המרכיבים העיקריים אולם הניסיון מראה כי קשה מאוד לעמוד באילוצי המשקל . ר"משקל מחמירים שבכפוף להם יתבצעו תהליכי בניין הכוח עבור החי

היתר ושל אופן השפעתו על היכולות -ה של משמעויות משקלמכאן החשיבות לפתח מּודעות והבנ. הללו והם מופרים באופן כמעט קבוע .ר"המבצעיות של לוחם החי

על ידי ניתוח הקשר שבין משך פעילות הלוחם בתנועה רגלית לבין המשקל שהוא ר"העבודה שתוצג מעריכה את יכולותיו של לוחם החיח וההצטיידות בו כלים גמישים יותר בבואם לבצע את "ניתוח כזה ייתן למקבלי ההחלטות בתחום פיתוח האמל, לדעתנו. נושא

.משימתםהמשימה המוטלת עליו והסביבה , שונים של הלוחםמתמטי הקושר בין פרמטרים -לצורך הניתוח האמור פותח בעבודה מודל פיזיולוגי

.שבה הוא פועל לבין משך הפעילות שהוא מצליח לבצע

Yonatan Grinshpun

Deliver or Search? Transportation Tactics in the Presence of Improvised Explosive Devices

Many military conflicts in recent years are asymmetric, in the sense that a relatively large, organized, trained and well

equipped government force is confronted by a relatively small, loosely organized and poor equipped insurgency. The

insurgents, realizing that direct confrontation with the government force may lead to the insurgency demise, use

guerrilla tactics to attack the government forces. As demonstrated in Southern Lebanon, Iraq and Afghanistan, one of

the most popular weapons used by insurgents and most lethal and damaging to the government forces, is improvised

explosive devices, planted along roads with the objective to hit and damage vehicles traveling on that road.

24

In this research, we present Markov-based probability models for an important problem related to current combat

situations, Improvised Explosive Devices (IED) attacks on convoys transporting supply and troops. We develop

continuous and discrete time Markov models to capture the key aspects that describe the situation and implement them

computationally to obtain some tactical insights regarding the effect of various operational parameters on the outcome

of convoy transportation missions in the presence of IED.

M2

ORSIS and Mehrez Prize Winners – Chair: Moshe Haviv

ORSIS prize winner: Yoel Drori and Marc Teboulle

A novel approach for analyzing the performance of first-order methods

We introduce a novel approach for analyzing the performance of first-order black-box optimization methods.

Following the seminal work of Nemirovski and Yudin (1983) in the complexity analysis of convex optimization

methods, we measure the computational cost based on the oracle model of optimization. Building on this model, our

approach relies on the observation that by definition, the worst case behavior of a black-box optimization method is by

itself an optimization problem, which we call the Performance Estimation Problem (PEP). We analyze the properties

of the resulting PEP for various black-box first order schemes. This allows us to prove a new tight analytical bound for

the classical gradient method, as well as to derive numerical bounds that can be efficiently computed for a broad class

of first order schemes. Moreover, we derive an efficient procedure for finding step sizes which produces a first-order

black-box method that achieves best performance.

Mehrez prize winner: Ricky Roet-Green and Refael Hassin

Equilibrium in a two dimensional queueing game: When inspecting the queue is costly

The classical model of customers' decision making in an unobservable queue assumes that it is too costly to acquire

the queue length information. However, due to recent advances in communication technology, various services make

that kind of information accessible to customers at a reasonable cost. Our model reflects this new opportunity. In our

model, customers of an unobservable queue can choose from three options: join the queue, balk, or acquire

information about the queue length and then decide whether or not to join it. When customers' decision problem

includes only two possible actions, it is intuitively expected that the action taken by an individual who tries to avoid

congestion will be inversely correlated with the actions taken by the others. In a model of more than two actions, the

strategy of the customers in equilibrium is not straightforward. We compute the equilibrium in this model and prove

its existence and uniqueness. We also show that the monopoly firm, that wishes to maximize the number of customers

who enter queue, can benefit from offering this third possibility of obtaining queue length information. Comparing to

the classical queue policies, the observable and unobservable queue models, we show that in many cases, the number

of customers that enter the queue will be larger with the queue policy that is presented in our model.

Applied Optimization - Chair: Ofer Levi

Yaniv Zaks

The Optimal Asset and Liability Portfolio for a Financial Institution with Multiple Lines of Businesses

In this paper we present an optimization framework to deal with the asset-liability portfolio selection problem. We

consider a financial institution that has multiple lines of business. The capital allocation is obtained by minimizing the

sum of the expected squared differences between the liability in each line of business and the value of the

corresponding investment portfolio. We show that in certain circumstances the bottom-up approach is consistent with

the top-down approach, where the optimal capital is determined for the whole portfolio rather than its individual

components. Such a case happens for example if the same weight function is used for all lines of business in the two

approaches. Finally, we obtain investment portfolios under some limitations on short sales.

25

Ofer Levi, Shaul Ladany

Optimal Sailing Policy

A dynamic programming model has been formulated to derive the minimal sailing route for a vessel to sail from an

origin to a destination, while passing n different segments in which deterministic wind conditions prevail. Turning

times and crossing wind-lines were considered. Complicated discrete computer solution procedure was programmed

which had to cope with ranges of inadmissibility of the continuous decision variables caused by inability to sail at

certain angles around a facing wind. Binary search method and iterative refinement schemes were applied to reduce

calculation times to the order of magnitude of ne minute, while obtaining sufficient accuracy. Numerical examples are

provided. Sensitivity analysis has indicated robust behavior of the optimal solutions with respect to changes in the

values of the different parameters

Supply Chain - Chair: Yael Perlman

Yael Perlman, Elad Crispil

Shift scheduling of preventive maintenance for wafer fabrication plants

Preventive maintenance (PM) scheduling is crucial to the performance of wafer fabrication plants (fabs). This research

proposes a short-term scheduling model that seeks to minimize the cost incurred during a fabrication shift. The

schedule needs to fulfill the following conditions: (i) the shift’s required output, measured by work-in-process (WIP)

levels, must be achieved; (ii) PM must be carried out in compliance with the manufacturer’s recommended PM policy;

and (iii) the equipment must undergo cleaning operations, as recommended by the manufacturer, in a way that reduces

the production of blank wafers. A linear programming model is formulated, and a practical procedure is provided to

solve the PM scheduling problem. Data from a real fab are used in a case study to determine which equipment requires

maintenance in the upcoming shift and to provide a work plan that enables the remaining equipment to attain the target

production quantities. Results of what-if analysis demonstrate how fabrication managers can use the proposed model

to analyze different scheduling alternatives.

Keywords: Semiconductor manufacturing, Scheduling, Preventive Maintenance.

Yael Perlman, Yaacov Oz

Reducing shoplifting by investment in security

We consider a single retailer with a given potential revenue, who sells a product that is subject to shoplifting. In order

to decrease losses due to shoplifting and to maximize his profit, the retailer can invest in security measures. In

particular, we assume that the retailer purchases security services from a single security supplier. The security supplier

decides which price to charge the retailer for these services, with the purpose of maximizing his own profit, and the

retailer decides on the quantity of security services to purchase. We address this problem using a game theoretic

approach, where the retailer competes with the supplier—the leader—who specifies first the service price. The retailer

responds by deciding how much to invest in security. We study the conditions under which both players are profitable

and the extent to which double marginalization affects the supply chain performance.

Keywords: Supply Chain; Shoplifting; Security Pricing, Game Theory

M3

Scheduling 3 - Chair: Liron Yedidsion

Danny Heremelin, Reuven Bar-Yehuda and Dror Rawitz

Admission control in line topologies via rectangle graphs

In this talk we consider the following scenario: You want to schedule connection requests in a network which has the

line topology. Each connection request is between a pair of nodes in the network, and is for a specific time interval.

The assumption is that only a single pair of computers can use a connection line in the network, and so in a feasible

schedule no two requests should conflict on the network in the same time. Our goal is to reject as little requests as

possible so as to obtain a feasible schedule.

26

We model this problem as the vertex cover problem in intersection graphs of axis-parallel rectangles on the plane. We

present two algorithms: The first is an EPTAS for non-crossing rectangle families, rectangle families where in every

pair of intersecting rectangles one rectangle contains a corner of the other. This algorithm extends to intersection

graphs of pseudo-disks. The second algorithm achieves a factor of (1.5 + epsilon) in general rectangle families, for any

fixed epsilon > 0, and works also for the weighted variant of the problem. Both algorithms exploit the plane properties

of axis-parallel rectangles in a novel way, and can be efficiently implemented.

Yossef Luzon, Yariv Marmor, Eugene Khmelnitsky

Optimal Control of a Flow-Shop Network

In this paper we suggest a new, intuitive and easy to use method for scheduling customers in a two server flow-shop

network with the maximum utilization objective. Multiple types of customers arrive to the network over a finite time

interval with corresponding time-varying arrival rates and constant service times. An analogue fluid network is

proposed and its optimal fluid control policy is determined. We make use of this optimal control policy to suggest a

new method for scheduling customers in the original discrete flow-shop network. Numerical simulations are used to

evaluate the performance of the suggested method and show that it performs optimally in almost all simulated

instances. Some additional properties of the network are discussed and illustrated.

Shay Hakim, Michael Masin, Michal Penn,

Asymptotically Optimal Approximation for the (High) Multiplicity Multi-Skill Scheduling Problem

The Multi-Skill Scheduling problem (MSP) is a generalization of job shop scheduling and resource-constrained

project scheduling (RCSPS) problems in which each activity has its known processing time and is processed by a

single resource from its subset of the capacitated resources. Each activity has its own subset of resources that can

execute it. The aim is to assign activities to resources and to order the activities on the resources so that the process

will terminate in the smallest amount of time under the precedence relations and the resource availabilities constraints.

MSP is NP-hard since RCSPS is known to be NP-hard and has many practical applications in the industry, service

applications and others.

We present the LP Synchronization Algorithm (LPSA) which is an LP-based asymptotically optimal approximation

scheme for the high multiplicity non-preemptive version of the problem. Our LPSA is a dispatching rule for the MSP

that imitates the fractional solution of the linear relaxation of a Mixed Integer Programing formulation of the problem.

Some numerical experiments will be presented as well.

Game Theory 2 - Chair: Arieh Gavious

Itai Arieli, Manuel Mueller-Frank

This paper considers single agent decision making under uncertainty and addresses the question to which degree the

private information of an agent is revealed through his optimal action. We show that if the set of actions is rich

(formally, a perfect set), then the agents optimal action reveals his posterior distribution for all but a negligible subset

(formally, a meager subset) of the space of continuous utility functions. If the action set is uncountable (not

necessarily perfect), then there exists a continuous utility function such that actions reveal beliefs. On the basis of the

single agent belief revelation result we establish information aggregation results in the sequential social learning

model and in the framework of repeated interaction in social networks

Ram Orzach, Ezra Einy, Ori Haimanko, Aner Sela

Common-Value All-Pay Auctions with Asymmetric Information

We study two-player common-value all-pay auctions with asymmetric information, assuming that one of the players

has an information advantage over his opponent. We characterize the unique equilibrium of this contest, and examine

the role of information on the players’ expected efforts, probabilities of winning, and expected payoffs. In particular,

we show that the players always have the same probability of winning the contest, their expected efforts are the same,

but they do not have the same expected payoffs. We also show that budget constraints may have non-trivial effect on

the players’ expected payoffs.

27

Amitay Kauffmann, Gal Zahavi

FEER Index – Forecasting Extreme Events Risk

In 2008, the S&P500 aggregated a loss of 30.16% during three selected days. Unfortunately, benchmark risk measures

didn't forecast these hazards. Consequently, we witness a growing interest in coherent risk measures, sensitive to high

moments and heavy tail risk. Such measures were proposed by Aumann-Serrano (2007) and Foster-Hart (2008). As a

generalization of these measures, we construct the FEER Index, coherent down-side risk measure "Forecasting

Extreme Events Risk", sensitive to heavy tail risk. We present closed-form solution as a function of the return

moments, VaR and CVaR. Furthermore, the FEER Index is dynamically calibrated to the market, becoming a live

seismograph for market catastrophes.

Queueing Networks in Production systems - Chair: Yoav Kerner

Ruth Shilman, Dean Grosbard, Gadi Rabinowitz, Israel Tirkel

Queueing Network Models for Manufacturing Systems with Downtimes

Queuing networks are frequently used to model the performance of complex manufacturing systems. Yet, existing

queuing network approximations cannot cope with some of the manufacturing lines’ complexities which frequently

generate non-renewable and highly variable internal flows. Scenarios of manufacturing systems with downtimes (e.g.

set-up, maintenance, and breakdown) present notable challenges. In this research, decomposition methods are used to

improve queuing network approximations suitable for such scenarios, with investigating the down-times' impact on

bottleneck queues.

We consider an open and tandem multi-class queuing network with class dependent service and routing. Each queue

has a single server, and an unlimited waiting line. Since an approximation of a multi-class system can be reduced to

the two-class case, where one class is the class of interest and the other class is an aggregated class of the rest of the

classes, two-class systems are considered. We assume that the aggregated class represents the down-time period.

Arrivals from the same class are processed following first-come-first-serve discipline, prioritizing the aggregated

class. We allow general distributions for inter-arrival and service times for the class of interest, but set deterministic

distributions for the inter-arrival and service times of the aggregated class. All distributions are assumed independent.

Our analysis illustrates that the departure variability can be expressed as the sum of two components. The first

component reflects the within-class effect while the second reflects the between-class interference effect,

demonstrating a novel approximation approach. The proposed development relies in addition on modifying the

variability functions for multi-class systems. The within-class effect variability function is class-dependent, while the

between-class variability function is a result of non-renewal processes. The proposed approximation has been

examined via simulation, for scenarios of multi-class systems with deterministic downtimes. For such scenarios, the

results demonstrate a relative error five times smaller than the best relevant existing procedures in the literature.

Shai Goren, Gad Rabinowitz and Yoav Kerner

A QBD approach for resources allocation

Allocation of limited resources in a stochastic production environment is a complicated task. Variation resources may

include production units’ inter-arrival times, production duration and quality, support resources availability, pace and

reliability, and more. This complexity drives practitioners to utilize static and simulation models as their main decision

support tools. The limitation of such model is being either naïve (static models) or high calculation resources

consuming (simulation models).

We suggest an analytical modeling approach to the resources allocation in a stochastic production environment. The

proposed model utilizes an integrative evaluation of the yield and flow time performance measures, as elements of an

objective function. Enabling yield evaluation and analysis is a significant needed extension of queueing theory.

We define and solve a typical production stage, which will become a node of a queueing network in future studies.

The stage consists of a single processing machine. The process deteriorates randomly from in-control to out-of-control

(with lower yield), and returns back to in-control only by repair. Repair is triggered via inspection of a recently

produced item. Inspection and repair resources are allocated to the stage according to an operational policy.

Production, inspection and repair are assumed to be non-perfect (i.e. not always successful), and their durations are

randomly distributed.

Since solving the general case is very difficult, we start with defining a relaxed model in which all of the durations

follow the exponential distribution, and the arrival process is Poissonian. This enables us to define the single stage as a

28

Quasi Birth and Death (QBD) process, and derive its steady state solution via matrix-geometric techniques. Yield and

FT are calculated as integrative outcomes of the inspection and repair resources allocation. The results of the

analytical model are validated against ones obtained through a simulation model. We also demonstrate how the single

stage model may be used as a building block for defining and solving multi-stage topologies.

Dean Grosbard

Rapid feasibility check for small size constrained capacity assignment problems

The problem of assigning server capacity to tasks under a given set of constraints is solvable using dynamic

programming. However, in production environment it is often the actual feasibility of the assignment problem under

the given constraints that is of more importance than the actual solution space. In this work we present a simple and

rapid method of checking the feasibility of a constrained capacity assignment problem. A set of sufficient and

necessary conditions for feasibility are presented and proven, we then go on to describe an algorithm for checking the

proposed conditions for small sized problems (N<20) followed by a network search algorithm for feasibility checks on

larger problems. Finally we discuss the importance of such feasibility checks in modern manufacturing systems and

present examples for their use.

Transportation 2 - Chair: Hillel Bar-Gera

Uri Yovel, Asaf Levin

Local Search Algorithms for Multiple-Depot Vehicle Routing and for Multiple Traveling Salesman Problems

with Proved Performance Guarantees

We consider two related problems: the Multiple-Depot Vehicle Routing Problem (MDVRP) and the Multiple

Traveling Salesman Problem (mTSP). In both of them, given is the complete graph on n vertices G=(V,E) with

nonnegative edge lengths that form a metric on V. Also given is a positive integer k. In typical applications, V

represents locations of customers and k represents the number of available vehicles. In MDVRP, we are also given a

set of k depots, and the goal is to find a minimum-length cycle cover of G of size k, that is, a collection of k (possibly

empty) cycles such that each vertex is in exactly one cycle, and each cycle in the cover contains exactly one depot. In

mTSP, no depots are given, so the goal is to find (any) minimum-length cycle cover of G of size k. We present local

search algorithms for both problems, and we prove that their approximation ratio is 2.

Michal Blumberg Nitzani, Hillel Bar-Gera

Representing signalized intersections in analytic dynamic traffic assignment model of transportation networks.

Traffic delays and queues at signalized intersections influence the flow of vehicles through the entire transportation

network and are responsible for main dynamic phenomena such as spillback, which is a state where a queue that spills

over to more upstream road sections. The estimation of signal delay time and the resulting queues are an essential part

of any dynamic traffic assignment model that aims to provide realistic traffic pattern predictions.

The analysis of traffic signal delay may focus on one of two states: over saturation - where the focus is on the dynamic

accumulation and dissipation of queues, and under saturation – where steady state conditions can be expected. In the

steady state analysis it is common to distinguish between a ’deterministic’ component of the delay, which represents

the delay caused to vehicles during the red phase of their arrival, and a ’random’ component of the delay which is due

to the variation in the number of vehicles arriving in each cycle. This variation leads to occasional overflow queues

(cycle failures) even in situations of under saturated intersections. To summarize, traffic signal delays have three

components: 1. steady-state deterministic component; 2. steady-state random component; and 3. dynamic queue

accumulation component.

We present an integrated framework to treat all three delay components within an analytic model that incorporates

Markov chain evaluation of the length of the queue at isolated fixed-cycle signalized intersections. Numerical

examples illustrate that overall the model behaves properly, and captures interesting impacts of random queuing traffic

signal delays on route choice and network level solutions.

Keywords: Solution stability; Signalized intersections; Dynamic traffic assignment; Dynamic network loading;

Embedded Markov chains.

29

Sarit Freund and Hillel Bar-Gera

An Optimization Framework for Travel Pattern Interpretation of Cellular Data

Collection of travel data by traditional survey methods is costly and time consuming, thus limiting the amount of data

being collected, as well as collection frequency and coverage.

Recent technologies offer new types of data collection options. In particular, cellular systems generate substantial

amounts of data, including records regarding the connection between handsets (phones) and base stations (antenna).

These records, collected by cellular service providers for various internal purposes, may provide an excellent source of

information regarding travel, with several critical advantages relative to traditional travel surveys: low cost, large

sample, long duration, and high response rate. Cellular data has some limitations too, particularly with respect to

accuracy and traveler identity; therefore, such data cannot provide a complete replacement for traditional surveys, but

it can complement and enhance them.

This study explores methods for identifying travel patterns from cellular data. The dataset used in this study includes

information regarding 9454 user cellular handsets, containing 3.7 M records, collected over 132 hours (nearly one

week). In a weekly perspective it is possible to observe substantial movements, when the handset user was probably

travelling, and periods of minor base station location changes, when the handset user was probably stationary,

participating in a certain activity. A primary challenge in this research is to provide an interpretation of the raw data

that distinguishes between activity durations and travel durations. A novel framework is proposed for this purpose,

based on a grading scheme for candidate interpretations of the raw data. A genetic algorithm is used to find

interpretations with high grades, which are considered as the most reasonable ones.

30

List of authors and participants (Session is indicated for active participants only)

Last Name Name Email Affiliation

Adler Nicole msnic@mscc.huji.ac.il Hebrew U. of Jerusalem T1 Tutorial

Adler Nicole msnic@mscc.huji.ac.il Hebrew U. of Jerusalem S3 Data Envelopment Analysis

Amador Sofia sofiamador@gmail.com Ben-Gurion U. of the Negev M1 Multi Agent Optimization

Anily Shoshana anily@post.tau.ac.il Tel Aviv U. S2 Game Theory 1

Anily Shoshana anily@post.tau.ac.il Tel Aviv U. P2 Plannery Lecture (Chair)

Arieli Itai iarieli@tx.technion.ac.il The Technion M3 Game Theory 2

Bar-Gera Hillel bargera@bgumail.bgu.ac.il Ben-Gurion U. of the Negev M3 Transportation 2

Baron Yonit ybarron@univ.haifa.ac.il U. of Haifa S3 Queueing 2

Beck Amir becka@tx.technion.ac.il The Technion S1 Continuous Optimization

Belsky Yarden yarden.belsky@gmail.com Tel Aviv U. S2 Multi Criteria optimization

Ben-Tal Aharon abental@ie.technion.ac.il The Technion P1 Naor Plenary Lecture (Chair)

Blumberg Nitzani Michal michal.nitzani@gmail.com Ben-Gurion U. of the Negev M3 Transportation 2

Bnaya Zahy zahy@bgu.ac.il Ben-Gurion U. of the Negev M1 Multi Agent Optimization

Bukchin Yossi bukchin@tau.ac.il Tel Aviv U.

Cohen Ilan ilanrcohen@gmail.com Tel Aviv U. S2 Scheduling 2

Cohen Tamar tamar.cg@gmail.com The Technion

Cohen Yaarit yaaritc@tx.technion.ac.il The Technion

Crispil Elad elad.crispil@gmail.com Intel M2 Supply Chain

David Yahel yahel83@gmail.com The Technion S2 Multi Criteria optimization

Drori Yoel dyoel@post.tau.ac.il Tel Aviv U. M2 Prize w inners

Eliazar Iddo eliazar@post.tau.ac.il Holon Institute of Technology T1 Tutorial

Forma Iris irisforma@gmail.com Tel Aviv U. S1 Transportation 1

Freund Sarit sarit.freund@gmail.com Ben-Gurion U. of the Negev M3 Transportation 2

Fruchter Gila Gila.Fruchter@biu.ac.il Bar Ilan U.

Gati Elad gatie@bgu.ac.il Ben-Gurion U. of the Negev

Gavious Arieh ariehg@bgu.ac.il Ben-Gurion U. of the Negev M3 Game Theory 2

Gerstl Tzvi tgerstl@nds.com Hebrew U. of Jerusalem S1 Scheduling 1

Gertz Tali tali.gertz@gmail.com The Technion

Gilenson Miri gilenson@tx.technion.ac.il The Technion

Gofer Eilam eylamg@rafael.co.il Rafael M1 Military OR

Goren Shai shaigoren100@gmail.com Ben-Gurion U. of the Negev M3 Queueing Netw orks in Production

Grinshpun Yonatan yoni_grinshpun@yahoo.com Rafael M1 Military OR

Grosbard Dean dean.grosbard11@gmail.com Intel M3 Queueing Netw orks in Production

Hakim Shay shayhakim@gmail.com The Technion M3 Scheduling 3

Halleli Sagi sagihilleli@gmail.com Ben-Gurion U. of the Negev S2 Scheduling 2

Halman Nir halman@mit.edu Hebrew U. of Jerusalem M1 Combinatorical Optimization

Hanany Eran hananye@post.tau.ac.il Tel Aviv U.

Hart Sergiu hart@huji.ac.il Hebrew U. of Jerusalem P2 Plannery Lecture (Speaker)

Haviv Moshe moshe.haviv@gmail.com Hebrew U. of Jerusalem M1 Strategic Behaviour in Queues

Heremelin Danny hermelin@bgu.ac.il Ben-Gurion U. of the Negev M3 Scheduling 3

Hochbaum Dorit hochbaum@ieor.berkeley.edu Berkeley U. P1 Naor Plenary Lecture (Speaker)

Holzman Ron holzman@tx.technion.ac.il The Technion T3 Tutorial

Karhi Shlomo shlomok@bgu.ac.il Ben-Gurion U. of the Negev S1 Scheduling 1

Kaspi Mor morkaspi@post.tau.ac.il Tel Aviv U. S1 Transportation 1

Kaspi Moshe moshe@bgumail.bgu.ac.il Ben-Gurion U. of the Negev S2 Multi Criteria optimization

Kauffman Amitai Zahavi.gal@gmail.com The Technion M3 Game Theory 2

Kerner Yoav kerneryo@bgu.ac.il Ben-Gurion U. of the Negev S3 Queueing 2

Kerner Yoav kerneryo@bgu.ac.il Ben-Gurion U. of the Negev M3 Queueing Netw orks in Production

Korach Efraim korachephraim@gmail.com Ben-Gurion U. of the Negev

Kremer Joseph kremer@bgumail.bgu.ac.il Ben-Gurion U. of the Negev S3 Simulation

Last Mark mlast@bgu.ac.il Ben-Gurion U. of the Negev M1 Combinatorical Optimization

Lev Varda varda@iec.co.il The Israel Electric Corp.

Levav Ben benlevav@gmail.com Rafael M1 Military OR

Levi Ofer levio123@gmail.com Ben-Gurion U. of the Negev M2 Applied Optimization

Lipets Vladimir lipets@il.ibm.com IBM S2 Scheduling 2

Luzon Yossi Yossi.luzon@gmail.com The Technion M3 Scheduling 3

Maimon Oded maimon@eng.tau.ac.il Tel Aviv U. Symposium

Session

31

Last Name Name Email Affiliation

Maman Shimrit shimritma@gmail.com Rafael

Mizrahi Shlomo mizrahi@bgu.ac.il Ben-Gurion U. of the Negev Symposium

Mor Baruch baruchm@ariel.ac.il Ariel U. S1 Scheduling 1

Mordoch Avraham mordoch@orange.net.il TOC Solutions S3 Heavy Scheduling Problems

Mosheiov Gur msomer@mscc.huji.ac.il Hebrew U. of Jerusalem S1 Scheduling 1

Nissim Raz raznissim@gmail.com Ben-Gurion U. of the Negev M1 Multi Agent Optimization

Nov Yuval yuval@stat.haifa.ac.il U. of Haifa S3 Queueing 2

Onn Shmuel onn@ie.technion.ac.il The Technion

Orzach Ram Orzach@oakland.edu Oakland U. M3 Game Theory 2

Oz Binyamin Binyamin.oz@gmail.com Hebrew U. of Jerusalem M1 Strategic Behaviour in Queues

Penn Michal mpenn@ie.technion.ac.il The Technion S2 Scheduling 2

Perel Efrat fgnaamati@gmail.com Tel Aviv U. S2 Queueing 1

Perel Nir perelnir@post.tau.ac.il Tel Aviv U. S2 Queueing 1

Perlman Yael yael.perlman@biu.ac.il Bar Ilan U. M2 Supply Chain

Pfeffer Aharona rona.pfeffer@gmail.com Tel Aviv U. S1 Transportation 1

Pinker Edieal edieal.pinker@yale.edu Yale U.

Pinto Gabi pintog@post.bgu.ac.il Azrieli college of eng. jerusalem S3 Heavy Scheduling Problems

Rabinowitz Gad rgadi@bgu.ac.il Ben-Gurion U. of the Negev Symposium

Raviv Tal talraviv@eng.tau.ac.il Tel Aviv U.

Ravner Liron lravner@gmail.com Hebrew U. of Jerusalem M1 Strategic Behaviour in Queues

Reif Barak Yair baraky1@post.tau.ac.il Tel Aviv U.

Reshef Offer offer@optum-ies.com Optum-Ies S3 Heavy Scheduling Problems

Roet-Green Ricky rgricky@gmail.com Tel Aviv U. M2 Prize w inners

Sabach Shoham ssabach@gmail.com Tel Aviv U. S1 Continuous Optimization

Sadeh Arik sadeh@hit.ac.il Holon Institute of Technology

Sarid Adi adi@sarid-ins.co.il Tel Aviv U.

Schechtman Edna ednas@bgu.ac.il Ben-Gurion U. of the Negev

Segev Ella ellasgv@bgu.ac.il Ben-Gurion U. of the Negev S2 Game Theory 1

Shaki Yair yair_sh@shoresh.org.il Ariel U.

Shefi Ron ronshefi@post.tau.ac.il Tel Aviv U. S1 Continuous Optimization

Sher Mali mali_sher@yahoo.com Israel Police

Shertzer Eliran scherzter@gmail.com Ben-Gurion U. of the Negev

Shilman Ruth ruthishilman@gmail.com Ben-Gurion U. of the Negev M3 Queueing Netw orks in Production

Shimkin Nahum Shimkin@ee.technion.ac.il The Technion S3 Simulation

Shpungin Yoseph yosefs@sce.ac.il Sami Shamun S3 Simulation

Shtern Shimrit shimrits@tx.technion.ac.il The Technion

Shtub Avraham shtub@ie.technion.ac.il The Technion T3 Tutorial

Sinuany-Stern Zilla zilla@bgu.ac.il Ben-Gurion U. of the Negev S3 Data Envelopment Analysis

Strimling David david.strimling@decisionaiding.com Decision Aiding S2 Multi Criteria optimization

Teboulle Mark teboulle@post.tau.ac.il Tel Aviv U. S1 Continuous Optimization

Tirkel Israel tirkel@bgu.ac.il Ben-Gurion U. of the Negev

Tzur Michal tzur@eng.tau.ac.il Tel Aviv U. S1 Transportation 1

Vaisbourd Yakov yakov.vaisbourd@gmail.com The Technion

Vaisman Slava slava@tx.technion.ac.il The Technion S3 Simulation

Volij Oscar ovolij@exchange.bgu.ac.il Ben-Gurion U. of the Negev S2 Game Theory 1

Yechiali Uri uriy@post.tau.ac.il Tel Aviv U. S2 Queueing 1

Yedidsion Liron yedidsion.liron@gmail.com The Technion M3 Scheduling 3

Yedidsion Liron yedidsion.liron@gmail.com The Technion M1 Combinatorial Optimization

Yitzhaki Shlomo shlomo.yitzhaki@huji.ac.il Hebrew U. of Jerusalem T2 Tutorial

Yitzhaki Shlomo shlomo.yitzhaki@huji.ac.il Hebrew U. of Jerusalem Symposium

Yovel Uri uyovel@hotmail.com The Technion M3 Transportation 2

Zahavi Jacob jacobz@tauex.tau.ac.il Tel Aviv U.

Zaks Yaniv zaksya@math.biu.ac.il Bar Ilan U. M2 Applied Optimization

Zivan Roie zivan.roie@gmail.com Ben-Gurion U. of the Negev M1 Multi Agent Optimization

Ziv-Av Amir zivava@mot.gov.il Ministry of Transportation Symposium

Zultan Ro'i zultan@bgu.ac.il Ben-Gurion U. of the Negev S2 Game Theory 1

Session

32

33

הכנס הרחובות בסמוך למלון

לתל אביב

לעיר העתיקה

מלון מלון

ליאונרדו ליאונרדו

נגבנגב

רכבת רכבת

אגדאגד מרכזמרכז

קניוןקניון

דוארדואר

קניוןקניון

טיילתטיילת

הקמפוס הקמפוס

הישן של הישן של

..גג..בב..אא

חניית חניית

מלון מלון הה

כספומטכספומט

34

האגודה הישראלית לחקר ביצועים

( ב"איל )

3102הכינוס השנתי

למאי 02 – 91

באר שבע, מלון ליאונרדו נגב

:בתמיכת

:הוועדה המארגנת

רבינוביץגדי יוסבאריה ג

אלה שגב יואב קרנר