Talks (Archive)
Showing 1  50 of 181 Results

DateTitle

11/05/2021
15:00  15:30Bachelorkolloquium Martin Krüßel 
16/04/2021
14:00  14:30Bachelorkolloquium Jonas Groven 
22/03/2021
10:30  11:15Bachelorkolloquium Jonathan Schneider 
01/03/2021
10:00  10:45Bachelorvortrag Alexander Cushicondor
Title: "Homomorphism Indistinguishability over Structures of Bounded Tree Depth" Abstract: It will be proved that two structures G, G' satisfy the same sentences of quantier rank at most k of first order logic with counting, if and only if they are homomorphismindistinguishable over the class of structures of tree depth at most k. 
22/02/2021
10:00  10:30Nondeterministic Automata with XORacceptance
Jagadish Singh
https://rwth.zoom.us/j/94253016445?pwd=VVdScGUzOFNJamZ5MCtRclYwK2xDdz09 
18/02/2021
10:15  11:15Oberseminar Anton Pirogov: Determinization and Ambiguity of Classical and Probabilistic Büchi Automata
Anton Pirogov
Büchi automata can be seen as a straightforward generalization of
ordinary NFA,
adapted to handle infinite words. While they were originally introduced for
applications in decidability of logics, they became a popular tool for
practical
applications, e.g. in automatabased approaches to model checking and
synthesis
problems. Being arguably both the simplest and most wellknown variant
in the
zoo of socalled omegaautomata that are considered in this setting, they
serve as an intermediate representation of omegaregular specifications of
verification or synthesis requirements that are usually expressed in a more
declarative fashion, e.g. using linear temporal logic.
Unfortunately, nondeterministic automata are not directly suitable for
certain
applications, whereas deterministic Büchi automata are less expressive. This
problem is usually solved by either constructing deterministic automata of a
different kind, or by restricting their ambiguity, i.e., the maximal
number of
accepting runs on some word. In both cases, the transformation is expensive,
yielding an exponential blowup of the state space in the worst case.
Therefore,
optimized constructions and heuristics for common special cases are
useful and
in demand for actual practical applications.
In this talk we present a new general construction for determinization from
nondeterministic Büchi to deterministic parity automata that unifies the
dominant branches of previous approaches based on the Safra construction
and the
MullerSchupp construction. Additionally, we sketch a number of new
heuristics
for determinization, some of which exploit properties of our unified
construction.
Apart from the classical nondeterministic and deterministic variants of
automata, it is natural to consider probabilistic automata, i.e.,
automata that
use a probability distribution on the states instead of nondeterminism
to decide
which state to go to next. It is known that in general, such automata
are more
expressive than classical automata and many decision problems are
undecidable.
We present subclasses of probabilistic automata that correspond to certain
ambiguity classes and are not more expressive than classical automata. 
08/01/2021
10:00  11:00ATIDeep Weisfeiler Leman
Daniel Wiebking
We introduce the framework of Deep Weisfeiler Leman algorithms (DeepWL), which allows the design of purely combinatorial graph isomorphism tests that are more powerful than the wellknown WeisfeilerLeman algorithm.
We prove that, as an abstract computational model, polynomialtime DeepWLalgorithms have exactly the same expressiveness as the logic Choiceless Polynomial Time (with counting) introduced by Blass, Gurevich, and Shelah (Ann. Pure Appl. Logic., 1999).
It is a wellknown open question whether the existence of a polynomialtime graph isomorphism test implies the existence of a polynomialtime canonisation algorithm. Our main technical result states that for each class of graphs (satisfying some mild closure condition), if there is a polynomialtime DeepWL isomorphism test, then there is a polynomialtime canonisation algorithm for this class. This implies that there is also a logic capturing polynomial time on this class. 
18/12/2020
10:30  11:00BachelorVortrag: The Logical Structure of Probabilistic Databases
Jonas Lindner
As probabilistic databases (PDBs) may be very large, compact representation systems for these frameworks have been studied. But because complete representation systems (like pctables) can lead to very complex probability spaces, considering simpler and more intuitive noncomplete representation systems, e.g. TItables or BIDtables, is of interest. These fundamental ’building blocks’ are especially interesting as they can be extended with views from the relational calculus to be more expressive. Similar to work done by Das Sarma et al. on representation systems for incomplete databases, this thesis introduces logical properties in order to separate the classes of PDBs that are representable with noncomplete representation systems. The properties studied take a PDB’s underlying incomplete database as well as its probability distribu tion into account and therefore help to grasp its logical structure. Having conducted this examination, convex combinations of probabilistic databases are studied, as they offer some advantages over the commonly considered representation systems regarding expressiveness. 
12/11/2020
11:00  11:30Connectivity and Routing using Graph Neural Networks
Frorian Frantzen
Graph Neural Networks form a deep learning architecture for machine learning tasks on graphs. Recently, they have been used to obtain heuristics for hard combinatorial problems such as TSP, SAT, or constraint satisfaction problems. The purpose of this thesis is to investigate how such approaches perform on computationally much easier shortest path problems. We will present lower bounds regarding the depth and width of a message passing neural network, which must be fulfilled for such a model to learn the shortest path problem. Then we will evaluate an existing neural network  originally designed for the traveling salesperson problem  for its applicability to the shortest path problem. While this network will fail with insignificant accuracies, these tests will allow us to derive unique challenges of the shortest path problem and which weaknesses in the architecture we need to fix to achieve better results. Based on these results, we will built our own neural network architecture and train it both supervised and unsupervised. We will evaluate this network in terms accuracy, scalability to larger graphs and how well the network generalizes the shortest path problem. 
30/10/2020
10:00  11:00ATIFlag Algebras in Graph Theory
Tim Seppelt
Flag algebras as introduced by Razborov represent a powerful tool in extremal combinatorics. This talk will provide a gentle introduction to the framework highlighting applications graph theory. Moreover, limitations of flag algebraic techniques are discussed following the work of Hatami and Norine. 
30/10/2020
10:00  11:00ATIFlag Algebras in Graph Theory
Tim Seppelt
Flag algebras as introduced by Razborov represent a powerful tool in extremal combinatorics. This talk will provide a gentle introduction to the framework highlighting applications graph theory. Moreover, limitations of flag algebraic techniques are discussed following the work of Hatami and Norine. 
27/10/2020
14:00  14:30BScThe FineGrained Complexity of Longest Common Subsequence
Benjamin Stutte
[bachelor]We will examine approaches to finegrained analyses of the Longest Common Subsequence (LCS) problem with special attention to recent results that have proven lower bounds for solving LCS under the Strong Exponential Time Hypothesis. We will first showcase two results, one by Abboud et al. (FOCS’15) and the other by Bringmann et al. (FOCS’15), where subquadratic lower bounds have been proven for the general LCS problem. Later, we turn to Bringmann et al.’s (SODA’18) which examined the optimal running time of multivariate LCS which proved lower bounds that coincide with the running times of the fastest known algorithms. 
27/10/2020
14:45  15:15BScStable and Efficient Algorithms for Logarithmically Counting Homomorphisms from Selected Graph Classes
Anton Florey
Graph homomorphisms are mappings between two graphs H and G that preserve all edge relationsof the lefthand side graph. The graph homomorphism counting problem #Hom(H, G) asks for
the number of homomorphisms from one graph H ∈ H to another graph G ∈ G. It is polynomial
time solvable for certain lefthand graph classes H. As part of his recent master thesis, Maximilian
Merz already provided implementations of various homomorphism counting algorithms. Unfor
tunately, they quickly get impracticable for large problem instances, as homomorphism numbers
grow extremely fast. This work revisits most of these algorithms with the novel idea of counting
the logarithm of these numbers. One main contribution is the implementation and evaluation of
these adapted algorithms. For most of the newly implemented functions, a significant improve
ment in efficiency over the exact counting implementations can be measured. Further experiments
also show that errors of this logarithmic approximation stay sufficiently close to the unavoidable
accuracy loss induced by finite machine precision.
https://rwth.zoom.us/j/95988725888?pwd=NUlaM05VdU5Kdkp0NXVTZHplWlFSQT09 
27/10/2020
14:00  14:30BScThe FineGrained Complexity of Longest Common Subsequence
Benjamin Stutte
[bachelor]We will examine approaches to finegrained analyses of the Longest Common Subsequence (LCS) problem with special attention to recent results that have proven lower bounds for solving LCS under the Strong Exponential Time Hypothesis. We will first showcase two results, one by Abboud et al. (FOCS’15) and the other by Bringmann et al. (FOCS’15), where subquadratic lower bounds have been proven for the general LCS problem. Later, we turn to Bringmann et al.’s (SODA’18) which examined the optimal running time of multivariate LCS which proved lower bounds that coincide with the running times of the fastest known algorithms. 
27/10/2020
14:45  15:15BScStable and Efficient Algorithms for Logarithmically Counting Homomorphisms from Selected Graph Classes
Anton Florey
Graph homomorphisms are mappings between two graphs H and G that preserve all edge relationsof the lefthand side graph. The graph homomorphism counting problem #Hom(H, G) asks for
the number of homomorphisms from one graph H ∈ H to another graph G ∈ G. It is polynomial
time solvable for certain lefthand graph classes H. As part of his recent master thesis, Maximilian
Merz already provided implementations of various homomorphism counting algorithms. Unfor
tunately, they quickly get impracticable for large problem instances, as homomorphism numbers
grow extremely fast. This work revisits most of these algorithms with the novel idea of counting
the logarithm of these numbers. One main contribution is the implementation and evaluation of
these adapted algorithms. For most of the newly implemented functions, a significant improve
ment in efficiency over the exact counting implementations can be measured. Further experiments
also show that errors of this logarithmic approximation stay sufficiently close to the unavoidable
accuracy loss induced by finite machine precision.
https://rwth.zoom.us/j/95988725888?pwd=NUlaM05VdU5Kdkp0NXVTZHplWlFSQT09 
23/10/2020
10:30  11:00MScDetecting Substructures with Graph Neural Networks
Michael Scholkemper
[master]We present theoretical results relating the computational expressiveness of graph neural networks (GNNs) to the color refinement algorithm for graph isomorphism (Morris et al., X et al.). These yield that given the wrong initialisation, graph neural networks are not able to detect triangles on regular graphs. We introduce a refinement algorithm for graph isomorphism that outperforms color refinement on regular graphs and propose a GNNarchitecture motivated by this new algorithm. We experimentally investigate the effect of different initialisations on the detection of triangles and extend this to kcycles and kcliques. 
16/10/2020
10:00  11:00ATILearning Concepts Described by Weight Aggregation Logic
Steffen van Bergerem
We consider weighted structures, which extend ordinary relational structures by assigning weights, i.e. elements from a particular group or ring, to tuples present in the structure. We introduce an extension of firstorder logic that allows to aggregate weights of tuples, compare such aggregates, and use them to build more complex formulas. We provide locality properties of fragments of this logic including FefermanVaught decompositions and a Gaifman normal form for a fragment called FOW1, as well as a localisation theorem for a larger fragment called FOWA1. This fragment can express concepts from various machine learning scenarios. Using the locality properties, we show that concepts definable in FOWA1 over a weighted background structure of at most polylogarithmic degree are agnostically PAClearnable in polylogarithmic time after pseudolinear time preprocessing. This is joint work with Nicole Schweikardt. 
13/10/2020
10:00  10:30BScA comparison of translation from LTL to Büchi automata
Anna Maiworm 
01/10/2020
10:00  10:45Bachelor Vortrag JanChristoph Kassing
Vortrag zur Bachelorarbeit: The Recursive Algorithm for Parity Games 
01/10/2020
10:00  10:45Bachelor Vortrag JanChristoph Kassing
Vortrag zur Bachelorarbeit: The Recursive Algorithm for Parity Games 
29/09/2020
10:00  10:45BachelorVortrag Johannes Lehmann
Vortrag zur Bachelorarbeit "Exact Minimization of omegaAutomata" 
29/09/2020
10:00  10:45BachelorVortrag Johannes Lehmann
Vortrag zur Bachelorarbeit "Exact Minimization of omegaAutomata" 
25/09/2020
10:30  11:00MScMasterVortrag: Graph Autoencoder for Chemical Compounds
Stefanie Winkler
https://rwth.zoom.us/j/91919735145?pwd=OWFJMDZ5aTErUjhLWEN0T1AwY0lIdz09 A challenging task in fuel design is to find suitable molecules that optimise desired properties like cetane and octane numbers. Our goal is to automate the generation of molecular graphs whose corresponding molecules serve as components for fuels with a high autoignition quality. We use three different neural network models built on variational autoencoders and a generative adversarial network: JTVAE by Jin et al., MHGVAE by Kajino and MolGAN by De Cao and Kipf [1–3]. Using a subset of QM9 for training, we adapt the models to optimise different combinations of the cetane and octane numbers of new molecules through Bayesian optimisation. Our results show that all models are able to produce valid molecules with high cetane and octane numbers. MHGVAE is able to generate most of the highest scoring molecules in the different experiments. Moreover, it creates a lot more unique molecules than the other two models, many of which are easy to synthesise. 
25/09/2020
11:15  12:00BScLower Bounds for the WeisfeilerLeman Dimension of Graphs of Bounded Tree Width
Lea Schirp
The WeisfeilerLeman (WL) algorithm is an iterative approach used to classify graphs and other relational structures. For every natural number k, there is a kdimensional version, which colors ktuples of vertices, running in polynomial time but there are also nonisomorphic graphs, which the kdim version cannot distinguish. This makes it useful to know the WLdimension of a graph, which is the least natural number k such that the kdim WL distinguishes the graph from all other nonisomorphic graphs. It is known that bounds on the WL dimension of a graph can be determined if its tree width is known. In this talk, we present tools for computing the tree width of a graph with a focus on Tamaki's implementation. We finally discuss the results of an experimental evaluation of the tree width of these graphs with the goal of finding out if the currently best known bounds on the WL dimension of graphs parametrized by their tree width are tight. 
25/09/2020
12:30  13:00BScCounting Homomorphisms via Model Counting and Knowledge Compilation
Patrick Bögel
The homomorphism counting problem #HOM asks how many homomorphisms there are from a graph H to a graph G. The model counting problem #SAT asks how many satisfying assignments a propositional formula has. Both problems are #Pcomplete, but #HOM becomes tractable if the lefthand side graphs are restricted to a class of bounded treewidth. In this talk we consider reductions from #HOM to #SAT with special attention to whether the tractable subproblems of #HOM are mapped to tractable subproblems of #SAT. Specifically we find reductions that map homomorphism counting instances where the lefthand side graph is a tree, to formulas which adhere to structural restrictions under which #SAT becomes tractable. We also utilize the approach of knowledge compilation to create formulas in a representation that allows for the model count to be calculated in linear time of its size. We evaluate the reductions practically with multiple stateoftheart model counters and find the reduction to be inferior to the direct approach. 
18/09/2020
10:00  11:00ATITupleIndependent Representations of Probabilistic Databases
Christoph Standke
https://rwth.zoom.us/j/96525649453?pwd=bHRma0lmV2xuRTNhWUdQbi9ockdNZz09 
03/08/2020
12:00  12:30BScBachelor Kolloquium Meder 
28/07/2020
16:00  17:00Vortrag Wolfgang Gatterbauer 
28/07/2020
16:00  17:00Vortrag Wolfgang Gatterbauer 
26/06/2020
10:00  11:00ATIATI Seminar Jan Böker
Jan Böker 
19/06/2020
10:00  10:45ATIATI Seminar Vortrag Steffen van Bergerem
Steffen van Bergerem 
05/06/2020
10:00  11:00ATIomegaWord Automata with Global Constraints
Patrick Landwehr 
29/05/2020
10:00  10:30ATIGraph isomorphism in quasipolynomial time parameterized by treewidth
Daniel Wiebking 
22/05/2020
10:00  11:00ATIThe Iteration Number of Colour Refinement
Sandra Kiefer
The Colour Refinement procedure and its generalisation to higher dimensions, the WeisfeilerLeman algorithm, are central subroutines in approaches to the graph isomorphism problem. In an iterative fashion, Colour Refinement computes a colouring of the vertices of its input graph. A trivial upper bound on the iteration number of Colour Refinement on graphs of order n is n1. We show that this bound is tight. More precisely, we prove via explicit constructions that there are infinitely many graphs G on which Colour Refinement takes G1 iterations to stabilise. Modifying the infinite families that we present, we show that for every natural number n ≥ 10, there are graphs on n vertices on which Colour Refinement requires at least n2 iterations to reach stabilisation. This is joint work with Brendan McKay. 
27/02/2020
10:00  10:30MScUnsupervised machine learning for constraint satisfaction problems
Jan Tönshoff 
27/02/2020
10:00  10:30MScUnsupervised machine learning for constraint satisfaction problems
Jan Tönshoff 
17/01/2020
10:00  11:00ATIGenerative Datalog with Continuous Distributions
Peter Lindner
Probabilistic Databases (PDBs) are a formal model of uncertainty in relational databases, as might occur in a variety of practical application scenarios such as noisy or unreliable input data, data integration or data cleaning. Quite recently, Bárány et al. (TODS 2017) proposed a language called "Probabilistic Programming Datalog (PPDL)" which uses classic Datalog rules that are extended by random sampling. In a nutshell, PPDL is a declarative probabilistic programming language with very close ties to database applications and can be seen as a tool to specify PDBs. In this talk, we focus on the generative part of the language, "Generative Datalog". While the original language of Bárány et al. only supported discrete probability distributions, we allow using probability density functions and inputs that are already PDBs themselves. We present the formal semantics of the language and discuss various properties and consequences, most notably, the support of PDB inputs and robustness with respect to the order of rule applications. This is joint work with M. Grohe, B. L. Kaminski, J.P. Katoen. 
15/01/2020
10:00  10:30MScAlgorithms for counting homomorphisms from small treewidth graphs
Maximilian Merz 
10/01/2020
10:00  11:00ATIIsomorphism Testing: From Strings to Hypergraphs
Daniel Neuen
We consider the isomorphism problem for hypergraphs taking as input two hypergraphs over the same set of vertices $V$ and a permutation group $Gamma$ over domain $V$, and asking whether there is a permutation $gamma in Gamma$ that proves the two hypergraphs to be isomorphic. We show that for input groups, all of whose composition factors are isomorphic to a subgroup of the symmetric group on $d$ points, this problem can be solved in time $(n+m)^{O((log d)^{c})}$ for some absolute constant $c$ where $n$ denotes the number of vertices and $m$ the number of hyperedges. The previous best algorithm for this problem due to Schweitzer and Wiebking (STOC 2019) runs in time $n^{O(d)}m^{O(1)}$.
As an application of this result, we obtain, for example, an algorithm testing isomorphism of graphs excluding $K_{3,h}$ ($h geq 3$) as a minor in time $n^{O((log h)^{c})}$. In particular, this gives an isomorphism test for graphs of Euler genus at most $g$ running in time $n^{O((log g)^{c})}$. 
13/12/2019
10:00  11:00ATIRUNCSP: Recurrent unsupervised network for CSPs
Martin Ritzert
Constraint satisfaction problems form an important and wide class of combinatorial search and optimization problems with many applications in AI and other areas. We introduce a recurrent neural network architecture RUNCSP (Recurrent Unsupervised Neural Network for Constraint Satisfaction Problems) to train message passing networks solving binary constraint satisfaction problems (CSPs) or their optimization versions (MaxCSP). The architecture is universal in the sense that it works for all binary CSPs: depending on the constraint language, we can automatically design a loss function, which is then used to train generic neural nets. In this paper, we experimentally evaluate our approach for the 3colorability problem (3Col) and its optimization version (Max3Col) and for the maximum 2satisfiability problem (Max2Sat). We also extend the framework to work for related optimization problems such as the maximum independent set problem (MaxIS). Training is unsupervised, we train the network on arbitrary (unlabeled) instances of the problems. Moreover, we experimentally show that it suffices to train on relatively small instances; the resulting message passing network will perform well on much larger instances (at least 10times larger). 
29/11/2019
10:00  11:00ATIFractional Sets, Fractional Decompositions, and Fractional Tangles
Eva Fluck 
29/11/2019
10:00  11:00ATIFractional Sets, Fractional Decompositions, and Fractional Tangles
Eva Fluck 
22/11/2019
10:00  11:00ATIAmbiguity in Probabilistic Büchi Automata
Anton Pirogov
Probabilistic Büchi automata are a natural generalization of PFA to infinite words, but have been studied indepth only rather recently and many interesting questions are still open. PBA are known to accept, in general, a class of languages that goes beyond the regular languages. In this talk I will present new classes of restricted PBA which are still regular, strongly relying on notions concerning ambiguity in classical ωautomata. 
20/11/2019
13:15  13:45MScAlgorithms for Maximum Matching Width
Yassin Bahloul 
20/11/2019
15:00  15:30BScThe FineGrained Complexity of FirstOrder Properties
Louis Härtel
Based on the results by Jiawei Gao, Russell Impagliazzo, Antonina Kolokolova and Ryan Williams, we break down the finegrained complexity of firstorder properties by analyzing their reduction from any modelchecking problem on a (k + 1)quantifier firstorder formula to the sparse version of kOrthogonal Vectors, and summarize their algorithmic improvements and consequences for common finegrained conjectures. 
15/11/2019
10:00  11:00ATIHow Stable are Node Embeddings and does it Matter?
Hinrikus Wolf 
25/10/2019
10:00  11:00ATITree automata with global constraints for infinite trees
Patrick Landwehr
We study an extension of tree automata on infinite trees with global equality and disequality constraints. These constraints can enforce that all subtrees for which in the accepting run a state q is reached (at the root of that subtree) are identical, or that these trees differ from the subtrees at which a state q' is reached. We consider the closure properties of this model, its decision problems and its connection to logic. 
24/10/2019
10:00  10:30BScRegular Sensing for Nested Word Automta
Alina Ibach 
17/10/2019
10:00  10:30BScComparing learning algorithms for regular expressions from positive examples
Konrad Ostrowski 
17/10/2019
10:45  11:15BScA comparison of algorithms for automata learning on sparse data
Caspar Zecha