Completed Research Projects | AC Group | TU Wien

Balancing Bicycle Sharing Systems

Mon, 01 Jan 2024 00:00:00 +0000

Christian Kloimüllner
Petrina Papazek
Bin Hu
Günther Raidl

Public bicycle sharing systems are booming worldwide in many major
cities. Such systems augment public transport very well, are green
alternatives to motorized individual traffic, may decrease traffic jams
and parking problems in cities to a certain extent, and last but not
least are an incentive for sports and a significant contribution to
public health. Modern systems have automated rental stations
distributed over the (inner) districts of the city, and users can easily
rent bicycles and bring them back at any other stations.

For user acceptance it is essential to provide enough bicycles as well
as parking slots for returning them at any station at almost all time.
As the usage pattern typically is not symmetric, e.g., people tend to
rent bikes at topographically higher stations and return them to lower
stations, an active maintenance by regularly transporting bikes from
some stations to others is crucial.

Project partners:

Problem Definition

The balancing bike sharing systems (BBSS) problem consists of finding an optimal rebalancing plan for the operator that consists of

a route for each vehicle on duty and
loading instructions for each station on the route

so that the system is in a balanced condition afterwards, i.e., no station is overly full or empty.

Two basic problem variants are considered:

Static case: The rebalancing process is done when the system is not used (e.g., overnight). Initial and target station fill levels are given as constants.
Dynamic case: The system is rebalanced while in use. Station fill levels are time-dependent and estimated by empirical data.

Approaches

Exact approaches based on mixed integer programming:

Hop index model
Time index model

Construction heuristics:

Greedy heuristic
PILOT heuristic

Metaheuristic approaches:

Variable neighborhood search approach
GRASP approach
Hybrid of GRASP with Path Relinking

Problem instances which we are using for our work can be found here.

Complete Solution Archives for Evolutionary Combinatorial Optimization

Mon, 01 Jan 2024 00:00:00 +0000

Benjamin Biesinger
Bin Hu
Günther Raidl

A common drawback of metaheuristic approaches, especially evolutionary algorithms (EAs) is that they usually do not keep track of the search history, and already evaluated solutions are often revisited. This leads to premature convergence, wasting resources for re-evaluations, and other malicious effects.

The solution archive (SA) is a concept for dealing with this problem. By attaching it to a classical EA, it enables detecting already evaluated candidate solutions and efficiently transforming them into similar but yet unvisited solutions, i.e., performing an “intelligent mutation”. The figure above illustrates the cooperation between the EA and the archive.

Application

Since this concept causes additional overhead for inserting and transforming solutions in the SA data structure, following conditions should hold in order to maximize the efficiency:

A compact solution representation keeps memory consumption for the archive low.
Problems with expensive solution evaluations benefit most from avoiding re-evaluations.
Problems with few feasibility constraints so that the transformation procedure can be kept simple.

Following combinatorial optimization problems were particularly successful with this approach:

Royal road functions
NK landscapes
Generalized minimum spanning tree problem
(r|p)-centroid problem and other competitive facility location problem

Cutting and Packing Problems

Mon, 01 Jan 2024 00:00:00 +0000

Frederico Dusberger
Günther Raidl

Cutting and packing problems occur in a multitude of industrial or other real-world applications such as

glass, paper and steel cutting,
container or pallet loading,
VLSI design,
and many other applications.

Usually items have to be cut from raw material and waste should be minimized, or the number of bins/containers needed to pack given items has to be minimized. In our research we dealt (until now) with two-dimensional bin packing problems where only guillotineable cutting/packing patterns are allowed.

Another kind of packing problems are the so called Knapsack Problems. Where items with an associated profit have to be packed into one or more knapsacks, and the profit has to be maximized. There are several variants of knapsack problems, such as the classical knapsack problem, multiconstrained knapsack problems, multiple knapsack problems, and many others. In our research we mainly deal with multiconstrained knapsack problems.

Solving cutting and packing problems can be done in various ways, we developed metaheuristics and exact algorithms for dealing with those problems. Especially the combination of evolutionary algorithms with problem specific heuristics, local and global optimization techniques are considered. Please see also our page on metaheuristics and evolutionary computation and the page on hybrid optimization techniques.

Cycles on Graphs and Properties of Graphs with Special Cycle Structure

Mon, 01 Jan 2024 00:00:00 +0000

Exploiting New Types of Structure for Fixed Parameter Tractability

Mon, 01 Jan 2024 00:00:00 +0000

Funding Organisation: The Austrian Science Funds, FWF
Project Number: FWF P26696

Project Team

Eduard Eiben
Robert Ganian
Ramanujan Sridharan
Stefan Szeider (Principal Investigator)

Topic

Many important computational problems are intractable on general instances. However, realistic problem instances often contain a certain hidden structure that can be exploited to solve the problem efficiently. Such instances form a tractable class (or an island of tractability). Previous research has identified a large number of tractable classes for various NP-hard problems.

The aim of this project is to develop a rigorous framework for exploiting new kinds of structure on a wide range of problem inputs; in particular, this framework will be based on the established notion of a modulator to a tractable class of instances, which applies to problem instances that may be placed in a tractable class by a small number of local changes. The type of structure exploited by algorithms developed within the proposed research is fundamentally different from the structure exploited by known state-of-the-art approaches. This will allow us to handle natural instances where all presently known state-of-the-art approaches remain unfeasible.

Human-centered Algorithm Engineering: Graph and Map Visualization

Mon, 01 Jan 2024 00:00:00 +0000

Funding Organisation: The Austrian Science Fund, FWF
Project Number: FWF P 31119
Duration: 11/2018 - 04/2023

Project Team

Sujoy Bhore
Jules Wulms
Guangping Li
Anaïs Villedieu
Martin Nöllenburg (Principal Investigator)

Topic

Human-centered algorithm engineering is an unconventional new paradigm in algorithmics that puts, for the first time, human factors into the focus of algorithm engineering, a methodology that is based on a cycle of design, analysis, implementation, and experimental evaluation of algorithms. Unlike other fields in computer science that investigate the interplay of humans and computers, our focus is on the symbiosis of human expert users and computers on the fundamental level of algorithms. We combine the mathematical accuracy and computational power of formal algorithmics with the creative power of the human mind in order to provide more effective and efficient algorithms for currently insufficiently solved and ill-defined algorithmic problems. Human-centered algorithms aim to produce solutions of better quality with faster overall performance, and higher user satisfaction. The project has two main goals: (i) establishing theoretical foundations, models of computation, and suitable evaluation methods and (ii) showing the practicality and benefits of the new paradigm for several prime examples of human-centered algorithmic problems that lack satisfying traditional algorithmic solutions in graph drawing and computational cartography.

Large-Scale Radio Therapy Scheduling

Mon, 01 Jan 2024 00:00:00 +0000

One of the most advanced medical centers for Ion Beam Therapy and Research in Europe is currently being built in Wiener Neustadt: MedAustron. The radiation therapies applied will use protons and carbon ions for cancer treatment. The first patient treatments are planned in 2016. In the full operational phase up to 1,200 people per year are expected to benefit from the international first class medicine offered by MedAustron.

Scheduling patient treatments with their large number of individual tasks and specific requirements in order to best utilize the available resources and treat patients as soon and smoothly as possible is a huge challenge. The Algorithms and Complexity Group collaborates with MedAustron in the development of advanced scheduling algorithms and their software implementation to fulfill these needs and to give as many cancer patients as possible a new opportunity.

Within the Algorithms and Complexity Group, Günther Raidl is leading this research effort.

Matheuristics: Hybrid Optimization Algorithms for Transportation Problems with Mutltiple Visits

Mon, 01 Jan 2024 00:00:00 +0000

Multi-Criteria Optimization of FTTx Networks

Mon, 01 Jan 2024 00:00:00 +0000

New Frontiers for Parameterized Complexity

Mon, 01 Jan 2024 00:00:00 +0000

Funding Organisation: The Austrian Science Fund, FWF
Duration: 1 September 2018 – 31 August 2022
Project Number: FWF P 31336

Project Team

Robert Ganian (Principal Investigator)
Thekla Hamm
Kirill Simonov
Viktoriia Korchemna

Research Statement

Many important computational problems are intractable on general instances. However, realistic problem instances often contain a certain hidden structure that can be exploited to solve the problem efficiently. The parameterized complexity paradigm provides the perfect tools and techniques to formalize, quantify and exploit various forms of structure in order to efficiently solve a plethora of NP-hard problems. This paradigm has been applied in many fundamental areas of computer science with great success, including graph algorithms, computational logic, boolean satisfiability, constraint satisfaction and others. However, there are prominent areas of computer science where we still lack a thorough understanding of the parameterized complexity of key computational problems.

This project aims at investigating the parameterized complexity of fundamental problems in two such high-impact areas: integer linear programming and machine learning. The obtained results will allow the design of efficient algorithms capable of exploiting various natural forms of structure in order to find exact solutions for NP-hard problems in these important areas of computer science. A secondary aim of the project is to apply newly obtained insights and advancements in other, including more traditional, frontiers of parameterized complexity research.

Optimization Challenges in the Future Federated Internet

Mon, 01 Jan 2024 00:00:00 +0000

Parameterized Compilation

Mon, 01 Jan 2024 00:00:00 +0000

Funding Organisation: The Austrian Science Funds, FWF
Project Number: FWF P26200

Project Team

Simone Bova
Ronald de Haan
Neha Lodha
Stefan Szeider (Principal Investigator)

Topic

Knowledge compilation offers a compelling approach to coping with computationally hard problems. This line of research was initiated in the early 1990s, and since then has become a very active and progressing field. Knowledge compilation proceeds in two phases: In the first phase the input data is compiled into a new representation, which is then used in a second phase to execute a number of individual tasks efficiently. Ideally, one aims at a compilation that causes only a polynomial increase in space, and the classical theory of compilation offers theoretical tools to decide whether such a compilation is possible or not. However, systematic research shows that most relevant problems are not compilable with a polynomial increase in space. Hence, the classical theory cannot provide reasonable guarantees for these problems.

The goal of this project is to overcome this limit of classical knowledge compilation by utilizing structural aspects of problem inputs (such as as tree-likeness, degree of cyclicity, or backdoor size). As the key to this goal we propose to study knowledge compilation within the framework of parameterized complexity, which has become an important and successful research direction in algorithms and complexity. Parameterized complexity provides powerful methods and tools for exploiting structural aspects of problems and is therefore ideally suited for this purpose. Using parameters we can exploit structural aspects of the input in order to support the compilation. Hence we aim at positive results in terms of upper bounds for compilation space and compilation time for problems that are not compilable in the classical sense.

Events

Symposium on New Frontiers in Knowledge Compilation

This symposium was organised by Pierre Marquis and Stefan Szeider and held in Vienna, Austria, June 4-6, 2015. The aim of this symposium was to bring together researchers who work on knowledge compilation from various angles, including knowledge representation, constraints, theory of algorithms, complexity, machine learning, and databases, as well as researchers from related areas. Lectures and discussions have put all these different approaches into context and stimulated a fruitful exchange of ideas between researchers from different fields. The symposium had over 30 participants and featured several invited and contributed talks.

More information can be found on the symposium homepage.

Dagstuhl Seminar on Recent Trends in Knowledge Compilation

Stimulated by the success of the Symposium in Vienna, another larger meeting on this topic was organized by Adnan Darwiche (UCLA, US), Pierre Marquis (Artois University – Lens, FR), Dan Suciu (University of Washington – Seattle, US, and Stefan Szeider (TU Wien, AT). The seminar was held at Schloss Dagstuhl, Wadern, Germany, September 17-22, 2017,

The seminar had over 40 participants and featured several invited and contributed talks as well as a demo session and and open problems session.

Among others, the seminar was an excellent occasion for disseminating results of the project, in terms of several talks.

A report on the seminar has been published as Dagstuhl Report (Volume 7, Issue 9, 2018).

More information can be found on the seminar homepage.

SLIM: SAT Based Local Improvement

Mon, 01 Jan 2024 00:00:00 +0000

Funding organization: Austrian Science Fund (FWF)

Project number: (FWF P 32441)

Project Team

Stefan Szeider (PI)
André Schidler
P.R. Vaidyanathan

Topic

Many problems can be efficiently translated into the propositional satisfiability problem (SAT) and then solved with a powerful SAT solver or variants such as MaxSAT or QBF-SAT. In some cases, the translation causes a significant blow-up in size, so this one-shot translation is not feasible. In this project, we overcome this limitation by repeatedly translating small local parts of the problem instance to SAT, which allows us to scale the SAT solver’s power to large instances. We could successfully implement this approach for various critical computational problems, including learning the structure of a Bayesian network, inducing an interpretable decision tree, or minimizing a Boolean circuit.

The Parameterized Complexity of Reasoning Problems

Mon, 01 Jan 2024 00:00:00 +0000

Reasoning, to derive conclusions from facts, is a fundamental task in Artificial Intelligence that arises in a wide range of applications from Robotics to Expert Systems. Different applications require different forms of reasoning such as Nonmonotonic Reasoning (e.g., reasoning under the presence of default as- sumptions), Constraint-Based Reasoning (reasoning with forbidden configurations), and Bayesian Reasoning (reasoning with uncertain data). All these forms of reasoning give rise to computational problems that can be solved algorithmically. The efficiency of algorithms has a direct impact on applications; for example, improved algorithms for Bayesian inference yield more accurate computer assisted medical diagnosis.

The aim of this project is to devise new efficient algorithms for reasoning problems and to gain new theoretical insights into the question of what makes a reasoning problem hard, and what makes it easy. We study reasoning problems within the framework of Parameterized Complexity, a new and rapidly emerging field of Algorithms and Complexity. Parameterized Complexity takes structural aspects of problem instances into account which are most significant for empirically observed problem-hardness. Most of the considered reasoning problems are intractable in general, but the real-world context of their origin provides structural information that can be made accessible to algorithms in form of parameters. This makes Parameterized Complexity an ideal setting for the analysis and efficient solution of these problems that we want to explore.

Two inputs for a reasoning problem: random (left) and realistic (right).

An Illustration: the picture above shows the visualizations of two inputs for a reasoning problem. The input to the left is a random input. The input to the right is a real-world input (from multiplier verification). Both inputs have approximately the same size in bits, however, evidently the real-world instance is somehow “structured”, whereas the random instance is not. The parameterized complexity approach allows to utilize this kind of structure to solve the problem efficiently.

Project Record

Project Title: The Parameterized Complexity of Reasoning Problems
Principle Investigator: Stefan Szeider
Funding Organization: European Research Council (ERC)
Project Type: ERC Starting Independent Researcher Grant
Project Volume: EUR 1.4 Mio
Project Duration: Jan 2010 - Dec 2014
Host Institution: Vienna University of Technology

Project Team

Simone Bova
Johannes Fichte
Robert Ganian
Ronald de Haan
Eva Nedoma
Friedrich Slivovsky
Stefan Szeider, project leader
Yue Chen
Serge Gaspers (University of NSW and NICTA, Australia)
Sebastian Ordyniak

de Haan, Fichte, Slivovsky, Gaspers, Ordyniak, Szeider, Bova, Ganian (from left to right)

Variable Dependencies of Quantified Boolean Formulas

Mon, 01 Jan 2024 00:00:00 +0000

Funding Organisation: The Austrian Science Fund, FWF
Project Number: FWF P 27721

Project Team

Friedrich Slivovsky
Tomas Peitl
Stefan Szeider (Professor, Principal Investigator)

Topic

The satisfiability problem of Quantified Boolean Formulas (QBF) offers succinct encodings for hard problems arising in areas such as formal verification and planning. This research project explores new ways to leverage independence of variables for QBF. The nesting of existential and universal quantifiers in Quantified Boolean Formulas can generate variable dependencies that are a serious obstacle to lifting successful techniques from propositional satisfiability to QBF. In some cases one can identify a variable dependency as spurious and conclude that the corresponding variables are in fact independent. This information can be used to significantly improve the performance of decision procedures, but deciding whether variables are independent is a highly intractable problem in itself.

The aim of this project is three-fold: (A) to advance the state of the art in detecting variable independence by developing new theory and improved algorithms for currently used methods; (B) to find new approaches to detecting and harnessing variable independence; (C) to extend the scope of successful techniques from QBF to more general problems.

Software

Research under Theme (B) has lead to the development of Qute (GitHub), a dependency learning QBF solver. Qute placed 3rd in the PCNF track and 4th in the Prenex non-CNF track of QBFEVAL'17.