Current Research Projects | AC Group | TU Wien

FWF Cluster of Excellence: Bilateral Artificial Intelligence

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

🌐 Visit the official FWF Project Radar Page

The project Bilateral AI aims at lifting artificial intelligence (AI) to the next level. Current AI systems are in a sense narrow. They center on a specific application or task such as object or speech recognition. Our project will combine two of the most important types of AI which have been developed separately so far: symbolic and sub-symbolic AI. While symbolic AI works with clearly defined logical rules, sub-symbolic AI (such as ChatGPT) is based on training a machine with the help of large datasets to create intelligent behavior. This integration, resulting in a Broad AI, is intended to mirror something that humans do naturally: the simultaneous use of cognition and reasoning skills. But what exactly is Broad AI? As opposed to Narrow AI, which is characterized by task specific skills, Broad AI aims at solving a wide array of problems, rather than being limited to a single task or domain. By combining sub-symbolic AI (machine learning, ML) with symbolic AI (knowledge representation and reasoning, KRR), Bilateral AI provides the means to develop the foundations of the capabilities and skill acquisition for problem solving by a Broad AI. Harnessing the full potential of both symbolic and sub-symbolic approaches can open new avenues for AI that are better at solving new problems, adapting to a wide variety of environments, having better reasoning skills, and being more efficient in terms of both computation and data use. These key features allow for a vast range of use cases for Broad AI, starting with drug development and medicine, over planning and scheduling, to autonomous traffic management and recommendation systems. With fairness, transparency, and explainability as top priorities, developing Broad AI is also essential for addressing ethical concerns and ensuring a positive impact on our society. These concerns play a central role as cross-cutting aspect in our project. The Broad AI resulting from the bilateral AI approach would use its own sensory perceptions to perform abstractions and engage in a logical thinking process. The AI could then, for example, organize a trip, minimize carbon emissions, or renovate a house as cost- effectively and ecologically as possible. In other words, AI could perform complex planning taking all aspects into account. Sepp Hochreiter, Director of Research: Broad AI could potentially improve our everyday lives as well as system-relevant aspects and processes - such as energy, transportation and healthcare - by becoming more environmentally sustainable, efficient and resource-friendly.

Parameterized Graph Drawing

Sun, 01 Jan 2023 00:00:00 +0000

Funding organization: Vienna Science and Technology Fund, WWTF
Project number: ICT22-029 (Information and Communication Technology 2022)

Project Team

Robert Ganian (Principal Investigator)

Martin Nöllenburg (Principal Investigator)

Simon Dominik Fink (Postdoctoral Researcher)

Thomas Depian (PhD Student)

Alexander Firbas (PhD Student)

Research Statement

The project is centered around two well-established fields of information and communication technology: (1) graph drawing and visualization, which deals with the construction and analysis of geometric representations of graphs and networks subject to specific layout conventions, and (2) parameterized complexity analysis, which offers the tools to design efficient algorithms as well as lower bounds custom-tailored to the specific structural properties of relevant inputs. Recent advances have highlighted the huge potential for the application of parameterized techniques on graph drawing and visualization problems. The two PIs of this proposal – Robert Ganian and Martin Nöllenburg – have already spearheaded an initial push to bring the two fields closer together, and this proposal will allow them to bring these efforts into fruition by targeting and resolving some of the most prominent questions in this intersection.

The project focuses on developing the tools and frameworks that will facilitate the parameterized analysis of central problems in graph drawing and visualization. The work is split into four fundamental themes, covering Extension Problems, Linear and Layered Layouts, Geometric Graph Representations and Bridges to Network Visualization. The output of each theme will include not only new algorithms but also tight lower bounds and, where relevant, implementations, significantly advancing the state of the art in these increasingly prominent fields of research.

ASK-SAT: Alternating Symmetry-Breaking Combinatorial Search with SAT

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

Project Acronym: ASK-SAT (Alternating Symmetry-Breaking Combinatorial Search with SAT)

Funding organization: Austrian Science Fund (FWF)

Project number: P 36688

Grant DOI: 10.55776/P36688

Project Team

Topic

Many unsolved problems in discrete mathematics and extremal combinatorics can be stated as whether a combinatorial object with a particular property and size exists.

The project focuses on developing novel methods for answering such questions using the innovative Satisfiability Modulo Symmetries (SMS) technique. This approach departs from traditional exhaustive search strategies by dynamically identifying and excluding redundant sub-configurations, thus streamlining the search process while utilizing the power of solvers for the propositional satisfiability problem (SAT).

The project aims to extend the capabilities of SMS to effectively tackle the existence of objects whose defining property requires alternating quantifiers, which present unique challenges beyond the scope of conventional SAT methods. This involves integrating advanced computational tools, including quantified Boolean formulas and symmetry-reasoning technologies.

Through this research, the project aspires to advance the fields of automated reasoning and discrete mathematics.

Software

The software tool developed through the project is available via GitHub, the documentation via Read the Docs.

Cooperative Optimization Approaches for Distributing Service Points

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

A joined research project from the Algorithms and Complexity Group, TU Wien, Austria, and Honda Research Institute, Germany

Project Team

Topic

For many business models an optimal distribution of service points in a customer community is needed. Examples are charging/battery swapping stations of electric vehicles, bicycle/car sharing stations, and repair stations. When planning such systems, estimating under which conditions which customer demand can be fulfilled is fundamental in order to design and evaluate possible solutions. To this end, demographic data is usually interlinked with geographic information, data on public transport, the street network, knowledge on manifold special locations etc. Usually, the whole process is challenging and error-prone. Customer demand information determined in such ways typically is vague, and not uncommonly a system built on such assumptions is not as effective as originally hoped for due to major deviations in reality.

To possibly improve this situation, we propose a cooperative optimization approach that incorporates potential users on a large scale and more tightly into the data acquisition as well as the optimization process. We confront the potential customers with certain location scenarios and ask them how these would suit his needs and how possibly these scenarios can be improved to fulfill more of his demand. This feedback is used to incrementally gain more knowledge about how much demand may be fulfilled under which conditions. New, more promising candidate location scenarios can then be derived and again be presented to the users. The process is iterated on a large scale with many potential users and many rounds until a satisfactory solution is reached.

The figure below depicts the components of the framework of our cooperative optimization approach and their interaction. The general framework consists of the following components: an evaluation component (EC), an optimization component (OC), a feedback component (FC), and a solution management component (SMC).

Cooperative Personnel Scheduling

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

a joined research project from the Algorithms and Complexity Group, TU Wien, Austria, and Honda Research Institute, Germany

Project Team

Topic

In classical personnel scheduling, the employees of a company are assigned their schedules, which determine at which times they work to fulfill some kind of demand. Our setting is slightly different. Each employee also referred to as a user, has some jobs to complete that need a specific resource. Only a limited amount of resources is available and only one job can use a resource at a time. Each job has to be assigned to a resource and a timespan in which the job is carried out. Furthermore, users will have preferences on when their jobs should be executed. These preferences have to be collected in some way. One possibility would be to let the employees specify their full preferences in the beginning. But this is tedious for the employees, error-prone and only allows for preferences that were explicitly considered when designing the algorithm.

In this project, we will develop and investigate a framework that learns these preferences over time through a limited amount of user interactions. The framework aims to construct schedules that respect employees’ preferences and are feasible with the given restrictions. While constructing the schedules it is allowed to ask the employees a limited number of questions regarding their preferences. The responses to these questions are used to model the employees’ preferences. The framework may also consider fairness, e.g. by balancing the number of user interactions among the employees. Research questions to answer include which algorithms work well for the occurring problems, how to model the employees’ preferences, and which kind of user interaction gives sufficient information on the preferences, without overwhelming the employees.

Doctoral College Logical Methods in Computer Science

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

Doctoral College Vienna Graduate School on Computational Optimization

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

Engineering Linear Ordering Algorithms for Optimizing Data Visualizations

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

Funding Organisation: Vienna Science and Technology Fund, WWTF
Project Number: ICT19-035
Duration: 07/2020 - 06/2024

Project Team

Alexander Dobler
Markus Wallinger
Jules Wulms
Martin Nöllenburg (Professor, Principal Investigator)

Alexander Dobler, Michael Stampfer, Martin Nöllenburg at the 20th anniversary of WWTF

Topic

Optimizing linear orderings of objects is a fundamental problem for many types of data visualizations ranging from graph layouts over geospatial data to abstract sets and sequences or time-series data. Yet a systematic investigation of algorithms for solving novel constrained and application-specific ordering problems that go beyond well-studied and NP-hard classic ordering problems is missing. Practical work in visualization often resorts to heuristics without rigorous performance and quality guarantees for solving these algorithmic problems. In this project we will take an algorithmic perspective on several different ordering problems in data visualization. In the algorithm engineering sense we want to cross the gap between fundamental theory and practical applications and aim to bring the benefit of rigorous formal methods into practically relevant implementations and at the same time define new algorithmic challenges inspired by recent visualization problems. On the one hand, we will investigate the complexity of new problem settings with special input configurations, structural constraints on feasible orderings, and dependencies between multiple objects and orderings. On the other hand, we will design and implement new and sufficiently scalable algorithms with formally proven performance guarantees to compute optimal and approximate solutions. We thoroughly evaluate the improvements over existing state-of-the-art heuristics in computational experiments and user studies.

Learning to Solve Dynamic Vehicle Routing Problems

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

a joined research project from the Algorithms and Complexity Group, TU Wien, Austria, and Honda Research Institute, Germany

Project Team

Topic

Vehicle routing - i.e., planning an optimal set of routes for a fleet of vehicles - is an intensely studied research area with enormous practical and raising relevance due to increasing mobility and transportation demand and new challenges coming, e.g., from increasing interest in shared mobility services and electric vehicles. The dial-a-ride problem (DARP), for example, is the problem of finding optimal tours of vehicles through different pickup and drop-off locations in order to serve a number of transportation requests, allowing different customers to share a vehicle. The electric autonomous dial-a-ride problem (E-ADARP) represents a challenging and practically relevant extension to the DARP, where electric autonomous vehicles are employed and their charging requirements have to be taken into consideration. Furthermore, not only classical objectives like total travel time have to be optimized, but user inconvenience also plays an important role. Thus, factors such as user excess ride time, which is due to detours because of the ride-sharing, have to be taken into account.

For such problems, heuristic optimization approaches are considered to be the means of choice due to a better scalability compared to exact approaches. In this project, we propose a heuristic framework based on large neighborhood search (LNS) to solve the E-ADARP, and we plan to tackle the issue of scalability by automatically designing, i.e., learning, efficient heuristics that either guide or possibly replace classical optimization techniques. We will investigate the usage of reinforcement learning with different machine learning models to dynamically select operators for the LNS from a set of possible operators during the optimization process. We intend to experimentally compare this approach to other learning techniques such as classical supervised learning, imitation learning, and Q learning.

Learning to Solve Quantified Boolean Formulas

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

Funding Organisation: Vienna Science and Technology Fund, WWTF
Project Number: ICT19-060
Duration: 05/2020 - 04/2023

Project Team

Franz Xaver Reichl
Leroy Chew
Friedrich Slivovsky
Stefan Szeider

Topic

Quantified Boolean Formulas (QBFs) can succinctly encode hard problems arising in planning, verification,
and synthesis, and the development of efficient procedures for evaluating QBFs (so-called QBF solvers)
will drive advances in all of these areas.

These applications require that QBF solvers not only decide whether a formula is true or false but also
output a witnessing strategy. For instance, when using QBF to solve a program synthesis problem, we
expect solvers to return the synthesized program. Our work on solver design and proof theory has
convinced us that strategies are at the heart of QBF research. However, the prevailing theoretical models of
QBF solvers as decision procedures or proof systems do not capture the fact that they need to compute
strategies. This project aims to develop a new, richer model of QBF solvers as strategy learning algorithms.

Software

Unique: A preprocessor for (D)QBF that computes unique Skolem and Herbrand functions (GitHub).
Pedant: A certifying DQBF solver based on definition extraction (GitHub).

MSCA COFUND Doctoral Programme LogiCS@TUWien

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

Overcoming Intractability in the Knowledge Compilation Map

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

Parameterized Analysis in Artificial Intelligence

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

Funding organization: The Austrian Science Funds, FWF
Project number: Y 1329 START-Programm (ParAI)
Grant DOI: 10.55776/Y1329

Project Team

Picture credits: Soeren Nickel, 2020

Research Statement

Parameterized complexity theory is a well-established paradigm used for the fine-grained analysis of computational problems. Such analysis can provide efficient algorithms for these problems by exploiting subtle structural properties of relevant inputs, as well as powerful lower bounds that rule out efficient algorithms even for severely restricted instances. Parameterized complexity analysis has found great success across numerous fields of computer science, with notable examples including graph algorithms, computational geometry, database theory, computational logic and constraint satisfaction. In the highly prominent fields of artificial intelligence (AI) and machine learning (ML) – areas which have become an ubiquitous part of today’s society – we see a distinct lack of foundational research targeting the fine-grained, parameterized complexity of fundamental problems. The goal of this six-year project is to change this.

A Parameterized Toolbox for Problems in AI and ML

One main objective of this project is the development of new innovative tools and machinery that allows us to apply the parameterized complexity framework in this setting. Indeed, most of the existing tools developed in parameterized complexity theory are designed to work in the setting of discrete problems on graphs. On the other hand, many problems of interest in AI and ML do not admit straightforward graph representations and/or contain non-discrete components. The development of the required tools will then go hand in hand with obtaining new algorithms and matching lower bounds for the studied problems.

QBF Proofs and Certificates

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

Funding organization: Austrian Science Fund (FWF)

Project number: (ESP 197 ESPRIT-Programm)

Grant DOI: 10.55776/ESP197

Project Team

Leroy Chew
Stefan Szeider
Friedrich Slivovsky

Topic

Quantified Boolean Formulas (QBF) extends propositional logic. Solving a QBF is theoretically more difficult than solving a propositional SAT (satisfiability) problem. QBFs are canonically PSPACE complete, meaning solving them could be used to solve any PSPACE problem by proxy. In the last decade progress in QBF solving is being undertaken. Proof complexity is the main theoretical framework used to understand SAT and QBF solving. Proof systems use the same logical rules as sound solvers, the proof size can often be a lower bound of running time between proof systems and related solvers. Proof systems can also be used to certify the correctness of solvers. However certification of QBF solvers is not yet a common occurrence.

We wish to further study the landscape of QBF proof complexity, studying the existing QBF proof systems that have emerged to capturing the solvers over the last decade. We also wish to use knowledge of soundness and complexity to make minor and major modifications to QBF proof systems, that could be beneficial to theory and practice. Our hope is that such knowledge will be of use to the QBF solving community.

QBF proof systems arranged via p-simulation

REVEAL-AI: Revealing and Utilizing the Hidden Structure for Solving Hard Problems in AI

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

Funding organization: Vienna Science and Technology Fund (WWTF)
Call: Information and Communication Technologies (ICT- 19)
Project number: ICT19-065 (Reveal-AI)

Project Team

Wolfgang Dvorak
Markus Hecher
Matthias König
Vaidyanathan P. R.
Anna Rapberger
André Schidler
Stefan Szeider (PI)
Stefan Woltran (co-PI)

Topic

At the core of several critical areas of AI and reasoning are hard-to-solve computational problems, such as the processing of constraints, carrying out sound reasoning tasks, and the verification of the correctness of procedures and protocols. All these problems are in general intractable and pose a challenge to algorithm design, even more as today’s applications demand more extensive problem inputs be solved. This requires new and more robust algorithms to facilitate further progress in technological innovation. Fortunately, typical problem inputs tend to contain some form of hidden structure, as the problem data is usually the product of a process. This research project is about revealing this hidden structure and utilizing it for an efficient solution.

STRIDES: Structure Identification with SAT

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

Project Acronym: STRIDES (Structure Identification with SAT)

Funding organization: Austrian Science Fund (FWF)

Project number: P 36420

Grant DOI: 10.55776/P36420

Project Team

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

Topic

SAT is the famous propositional satisfiability problem. It asks to assign the variables of a propositional formula with truth values 0 and 1 such that the entire formula becomes true. SAT is generally considered intractable, but over the last twenty years, computer programs (SAT solvers) have been engineered that can solve the problem surprisingly fast. Numerous other hard problems can be translated to SAT and solved via SAT solvers. However, the translation to SAT often causes a significant increase in size, which limits the application of SAT solvers to small problem inputs.

The project aims at scaling the use of SAT solvers to large problem inputs by utilizing the recently introduced SAT-based Local Improvement Method (SLIM). It starts with an initial heuristic solution and repeatedly applies a SAT solver to small local parts of the input, overcoming the size limitation. The project will investigate using SLIM for problems that ask to find a specific hard-to-find structure in given data. Such structure identification problems arise in text data, large graphs and networks, and logical circuits. It will focus on methods for making the SLIM approach more efficient and finding general insights into its workings. The research is expected to lead to new theoretical and practical results.