Department of Computer Science and Electronics,

Politecnico di Milano

 

APPLICATIONS OF DISCRETE OPTIMIZATION:

Routing problems and some general ideas for approximation algorithms

 

Summer Semester 2006

(2.5 credits for Master, 5 credits for PHD)

 

Lecturers: Luca Maria Gambardella and Monaldo Mastrolilli

IDSIA, Istituto Dalle Molle di studi sull’Intelligenza Artificiale,

www.idsia.ch/~luca, www.idsia.ch/~monaldo

 

Exam: Wednsday 28th June

Master students: 16:30 - 18:00, room D.0.1

PhD students: 14:30-18:30, room D.2.2

 

 

Aim and Scope

 

The course is divided in two parts. The first part concerns with routing problems while the second part is dedicated to approximation algorithms.

The first part, lectured by Luca Maria Gambardella, is based on the consideration that many practical distribution problems requires to find a set of routes minimising the number of traveled kilometers, the number of used vehicles, the cost of the fleet while satisfying customer demand and constraints. There are different stages and different models for these distribution problems starting from the traditional traveling salesman problem, moving to problems with precedence constraints and finally to vehicle routing problems. There are three major considerations that are important when we try to solve these problems. First of all, they are inherently combinatorial, and exact algorithms fail when the dimension of the problem (number of customers and orders) reaches a reasonable size. Secondly, the problem can be extended and made more complex in many ways, for instance, adding more than one depot, considering more than one vehicle type, accounting for stochastic customer demand (the exact requested quantity is known only at delivery time), considering time windows during which the customers must be served, taking into account vehicle accessibility restrictions (some customers cannot be served by some vehicles). Finally, the problem can become very different when we consider on-line distribution, that is, we accept delivery orders for lorries which are en route. There, geolocation of customers and vehicles, online data transfer among lorries and the base station, have an impact as great as the solution strategy. At the end there are real industrial applications where all these constraints are present in an operative environment and human planners use optimization software to solve distribution problems.

 

The second part, lectured by Monaldo Mastrolilli, considers approximation algorithms for NP-hard optimization problems. An r-approximation algorithm is a polynomial time algorithm that has been proven to return a solution within r-times the optimal value. The job of a potential algorithm designer, faced with a NP-hard problem, could certainly be simplified if one could give a standard set of recipes for coming up with the algorithmic idea. The focus of this part of the course is on methodology to design approximation algorithms. In particular some fairly general techniques will be presented: namely, we first present the Local Ratio technique and then the main tools and the standard approaches for the construction of Polynomial Time Approximation Schemes.

 

Contents

 

The first part of the course resents solution techniques for traveling salesman problems, sequential ordering problems and vehicle routing problems. We present mathematical formulation, branch-and-bound and branch-and-cut techniques associated to different approach to compute lower bound for these problems, some based on the related assignment problem and other based on tree relaxation. Next constructive methods like nearest neighbor, double ended, multiple fragment, Christofides, Clarke-Wright, space filling curve and composite methods (routing + clustering) are investigated and explained. Next part is dedicated to deeply investigate meta heuristics algorithms like simulated annealing, tabu search, genetic algorithms, ant colony optimization in relation with local search techniques such as 2opt, 3opt, Lin-Kernighan, sop-3-exchange, customers relocation, crossover, customers exchange, path exchange. At the end we present some industrial applications where these techniques have been successfully applied to daily solve distribution problems.

 

The second part of the course introduces Local Ratio technique as a methodology for the design and analysis of algorithms for a broad range of optimization problems. The technique is remarkably simple and elegant, and yet can be applied to several classical and fundamental problems (including covering problems, packing problems, and scheduling problems). The local ratio technique uses elementary math and requires combinatorial insight into the structure and properties of the problem at hand. Many applications of these ideas are provided during the course together with the analysis of the associated approximation algorithms.

Polynomial Time Approximation Schemes (PTAS) are approximation algorithms that return in polynomial time solutions that can be arbitrarily close to the optimum. We present a tutorial that provides an introduction into the area of PTAS. Several approaches are explained in great detail and illustrated in many examples and exercises.

 

Some applications of the presented techniques:

 

  • Vehicle routing
  • Fleet management
  • Covering problems (set cover, hitting set, vertex cover, etc…);
  • Forest problems (shortest path, spanning tree, Steiner tree and forest, etc…)
  • Packing and scheduling problems (independent set, interval scheduling, scheduling with release times and deadlines, bandwidth allocation, general caching problem, etc…)

 

References

 

Course Materials

  1. Slides on Routing Problems (download)
  2. Slides on Local Ratio (download: pdf or ppt)
  3. Slides on Approximation Schemes (download: pdf or ppt)
     

Timetable

Date Time Room Lecturer
Tuesday, 14th March 16:30 - 18:15 D.2.6 Gambardella
Wednesday, 15th March 16:30 - 18:15 D.2.1 Gambardella
Tuesday, 21sh March 16:30 - 18:15 D.2.6 Gambardella
Wednesday, 22nd March 16:30 - 18:15 D.2.1 Gambardella
Tuesday, 28th March 16:30 - 18:15 D.2.6 Gambardella
Wednesday, 29th March 16:30 - 18:15 D.2.1 Gambardella
Tuesday, 4th April 16:30 - 18:15 D.2.6 Mastrolilli
Wednesday, 5th April 16:30 - 18:15 D.2.1 Mastrolilli
Tuesday, 11th April 16:30 - 18:15 D.2.6 Mastrolilli
Wednesday, 12th April 16:30 - 18:15 D.2.1 Mastrolilli
Wednesday, 19th April 16:30 - 18:15 D.2.1 Mastrolilli

Lecturers

Luca Maria Gambardella is Director at IDSIA, Istituto Dalle Molle di Studi sull’Intelligenza Artificiale, Lugano, Professor of Algorithms at SUPSI, lecturer of Application of Discrete Optimization at Milan Polytechnics and lecturer of the Artificial Intelligence module in the Knowledge Management course at Informatics Faculty at the University of Lugano. IDSIA is a research institute of both SUPSI (University of Applied Science of Southern Switzerland, Department of Technology and Innovation) and USI (University of Lugano).
His experience and his background cover different aspects, from management to research, including the preparation and the negotiation of many research projects, the direction of a research institute, the involvement in AntOptima the spin-off company of IDSIA other than the realization of successful and effective algorithms:
His current research interest and his major publications include Ant Colony Optimization and meta-heuristics for global optimization, learning and planning for complex collective systems, simulation and scheduling for logistic and routing problems. He has invented several meta-heuristics algorithms that can be applied to a wide class of combinatorial optimization problems. In some of these domains, such as the quadratic assignment [Gambardella, Taillard, Dorigo,1999], the sequential ordering [Gambardella & Dorigo 2000], the vehicle routing with time window [Gambardella, Taillard, Agazzi 1999] and the flexible job shop scheduling problems [Mastrolilli & Gambardella 2000] these approximation algorithms are among the best currently worldwide available and for many benchmark instances new best known solutions have been computed.

Monaldo Mastrolilli is a senior researcher at IDSIA and lecturer of Operations Research at SUPSI. In 1997 he graduated with honours in Computer Science Engineering at Politecnico di Milano (Italy), and won the Third prize of the general category in the Whizzkids'97 competition on combinatorial optimization, designed by the Technical University of Eindhoven. He joined IDSIA as a PhD student in 1998. In 2002 he received a summa cum laude Ph.D. degree in Computer Science at University of Kiel (Germany). Since 2004 he holds a permanent position as senior researcher at IDSIA. His main scientific interest is in approximation algorithms for NP-hard problems. He is the author of more than 30 papers published in leading international scientific journals (like Journal of Algorithms, Discrete Applied Mathematics, European Journal of Operational Research, etc…) and conferences (like ESA, AAAI, LATIN, ISAAC, SWAT, etc…). He is the Head of Optimization Department of AntOptima for which he supervised and worked on several industrial projects.