informatik-vortraege March 2023

informatik-vortraege@lists.rwth-aachen.de

6 participants
6 discussions

UnRAVeL Survey Lecture: Biggest Milestones

by Kassing, Jan-Christoph

Dear all, part of the programme of the research training group UnRAVeL is a series of lectures on the topics of UnRAVeL’s research thrusts algorithms and complexity, verification, logic and languages, and their application scenarios. Each lecture is given by one of the researchers involved in UnRAVeL. This years topic is "Biggest Milestones - Research at Its Peak", UnRAVeL professors will present the most important milestone of their respective research. This is not only about outstanding or astonishing results for the respective community, but also about their very own scientific milestones, for example a paper or result that the presenter particularly likes and is very proud of. We get answers to the following questions: * Which of their papers/results would never be allowed to be deleted if they could keep just one? * Which paper/result had the greatest influence, on science or on themselfs? * Why and in what context is it a milestone? All interested doctoral researchers and master students are invited to attend the UnRAVeL lecture series 2023 and engage in discussions with researchers and doctoral students. All events take place on Thursdays 12:30 to 14:00, Computer Science Center, Building E2, ground floor, B-IT room 5053.2. The schedule is as following: * 06.04. Michael Schaub - Learning from Networks with Unobserved Edges * 13.04. Erika Ábrahám - Building Bridges between Symbolic Computation and Satisfiability Checking * 20.04. Martin Grohe - The Quest for a Logic Capturing PTIME * 27.04. Christina Büsing - Robust Strategic Planning for Mobile Medical Units * 04.05. Sebastian Trimpe - Event-Triggered Learning * 11.05. Nils Nießen - Can Trains Be on Time? * 25.05. Jürgen Giesl - Proving Termination with Dependency Pairs * 15.06. Britta Peis - Ascending Auctions and Matroids * 22.06. Gerhard Lakemeyer - The Situation Calculus as Lingua Franca for Reasoning about Action * 29.06. Christopher Morris - Weisfeiler and Leman Go Machine Learning: Expressivity and Generalization of Graph Neural Networks * 13.07. Joost-Pieter Katoen - Can we meet the deadline? Most probably: yes We are looking forward to seeing you at the lectures. Kind regards, Jan-Christoph for the organisation committee

8 months, 1 week

Einladung: Informatik-Oberseminar Max Lyon

by Max Lyon

+********************************************************************** * * * Einladung * * * * Informatik-Oberseminar * * * +********************************************************************** Zeit: Montag, 17. April 2023, 10.00 Uhr Ort: Gebäude E3, Seminarraum 118, Ahornstr. 55 Referent: Max Lyon M.Sc. Lehrstuhl für Informatik 8 Thema: Quad Mesh Layout Structure Optimization Abstract: Quad meshes are very useful in many applications such as modeling and animation in the entertainment industry, simulation in the construction industry, or as intermediate representation to form easily editable spline surfaces. Some of the interesting properties of quad meshes compared to triangular meshes are natural alignment of edges to the surface's principal curvature directions and their ability to form highly anisotropic elements without the need of excessively small angles. Therefore, the computer aided or fully automatic generation of quad meshes has been a research focus for many years and is probably going to remain as such for the foreseeable future. One set of methods that have proven to be particularly well suited for the generation of high quality quad meshes are centered around the creation of an integer-grid map, i.e. a parametrization of the surface into the 2D domain whose parametric integer iso-lines induce a quad mesh structure on the input mesh. These methods offer a high degree of control over the desired sizing and alignment of the resulting quads and generally achieve high element quality. Unfortunately, they disregard the high level structure, i.e. how the irregular vertices of the resulting quad mesh are connected by edge strips. Ideally, these edges form a coarse quad layout, enabling the use of highly efficient multigrid solvers or the manual editing of a small number of spline patches. In this thesis we give an overview of current state-of-the-art quad meshing algorithms which we then improve upon in different aspects: We show how a valid quad mesh can always be extracted for the given layout via careful re-embedding of a so called motorcycle graph combined with patchwise harmonic parametrizations followed by a global optimization. We extend the motorcycle graph construction in order to lift the constraints of layouts having to exactly follow the boundaries of the mesh leading to meshes with less distortion whose quads can be cut off by the boundary at an arbitrary angle. Further, we present a method that is guaranteed to yield a coarse quad layout that adheres to a minimum quality as defined by the user defined maximal allowed deviation of layout arcs from desired direction. For cases where strict adherence to the input field is not required we show how singularities may be moved to better locations or removed entirely to further improve the resulting layout structure. Finally, we discuss how a quad mesh or layout may be refined in order to achieve a locally increased resolution or anisotropy. Es laden ein: die Dozentinnen und Dozenten der Informatik

8 months, 2 weeks

Einladung Informatik Oberseminar René Röpke

by Lehr- und Forschungsgebiet Informatik 9

+********************************************************************** * * * Einladung * * * * Informatik-Oberseminar * * * +********************************************************************** Zeit: Mittwoch, 12. April 2023, 15.00 Uhr Ort: 9222, E3, Ahornstr. 55 und hybrid via Zoom (https://rwth.zoom.us/j/92145754381?pwd=TmhOOXBDc3RITGl1L0NwdGhLTEVKUT09) Referent: René Röpke M.Sc. Lehr- und Forschungsgebiet Informatik 9 (Lerntechnologien) Thema: Extending Game-Based Anti-Phishing Education using Personalization Abstract: Phishing poses an imminent and wide-ranging threat to Internet users worldwide, in which attackers use methods of deception to lure victims into disclosing information. Recent reports state high numbers of phishing incidents and, so far, technical solutions fail to stop the threat completely. As a complementary approach, user education using anti-phishing learning games has been explored to raise awareness and teach the necessary knowledge and skills to detect and protect against phishing attacks. A common game mechanic used in existing games requires learners to classify URLs as either legitimate or phishing in a binary decision scheme. Here, a problem can occur if learners do not know the service of a given URL and are unable to classify the URL due to a lack of reference. As such, learners may revert to guessing which may weaken the game's potential for practice, since learners cannot relate between correct classifications and the applied knowledge. Furthermore, the possibilities for feedback are limited since the binary decision mechanic does not provide any insights into learners' decision processes and possible misconceptions. In this dissertation, the limitations for feedback as well as the problem with classifying unknown URLs in anti-phishing learning games are addressed as follows: First, a review of existing learning games provides insights into their design and covered learning content. Its results are used in guiding the design and implementation of two new game prototypes. Here, the first game extends the before-mentioned binary decision mechanic and requires learners to sort URLs into one of many categories, depending on which manipulation technique was applied to a distinct part of the URL. The second game requires learners to apply different manipulation techniques and create their own malicious URLs using a puzzle mechanic. Next, the means of personalization for anti-phishing learning games are explored and a personalization pipeline is developed. By considering the learners' familiarity with different services and dynamically creating benign and phishing URLs, the content of anti-phishing learning games can be personalized. To evaluate the new game prototypes as well as the application of the personalization pipeline, two comparative user studies are conducted in a between-group design with pre-, post- and longitudinal testing. In the first user study with 133 participants, both games are evaluated and compared to a baseline implementation. While participants of the new games did not perform significantly better than the control group, results show significant improvements in the participants' performance and confidence between pre- and post-tests for all games, as well as notable differences when classifying URLs of unknown and known services. In the second user study with 49 participants, the personalization pipeline is integrated into one of the games, in order to compare its personalized and non-personalized version. Here, personalization enables the control of service familiarity and allows insights into how URLs of unknown services are handled within the game. While participants of the personalized game did not outperform the participants of its non-personalized version, the evaluation of in-game behavior provides insights into learners' decision processes and possible problems or misconceptions. Furthermore, results of a longitudinal evaluation of all games and versions show that knowledge is retained since the participants perform still significantly better than in the pre-test. In all, this dissertation presents first approaches and research results in the domain of personalized anti-phishing learning games. Future work may entail redesigning anti-phishing learning games to incorporate further means of personalization and to understand how learner characteristics can be utilized in anti-phishing learning games. Es laden ein: die Dozentinnen und Dozenten der Informatik

8 months, 2 weeks

Einladung Informatik-Oberseminar Weiyue Wang, 15. März 2023, 14:00 Uhr

by Stephanie Jansen

********************************************************************** * * * Einladung * * * * Informatik-Oberseminar * * * *********************************************************************** Zeit: Mittwoch, 15. März 2023, 14:00 Uhr Zoom: https://rwth.zoom.us/j/93749104715?pwd=bEFSR0lsR1o1Tzl2enVMSzM0aTE2Zz09 Meeting-ID: 937 4910 4715 Kenncode: 743829 Referent: Weiyue Wang, M.Sc. Thema: Neural Hidden Markov Model for Machine Translation Abstract: Recently, neural machine translation systems have shown promising performance. One of the key components that almost all modern neural machine translation systems contain is the attention mechanism, which helps an encoder-decoder model attend to specific positions on the source side to produce a translation. However, recent studies have found that using attention weights straight out of the box to align words results in poor alignment quality. This inspires us to introduce an explicit alignment model into the neural architecture in order to improve the alignment and thus also the translation quality of the overall system. To this end, we propose a novel neural hidden Markov model consisting of neural network-based lexicon and alignment models trained jointly with the forward-backward algorithm. Various neural network architectures are used to model the lexicon and the alignment probabilities. We start with feedforward neural networks and apply our first model to re-rank n-best lists generated by phrase-based systems and observe significant improvements. In order to build a monolithic neural hidden Markov model, the more powerful recurrent neural networks are applied to the architecture, and a standalone decoder is implemented. By replacing the attention mechanism with an alignment model, we achieve comparable performance to the baseline attention model while significantly improving the alignment quality. We also apply the state-of-the-art transformer architecture to the neural hidden Markov model and the experimental results show that the transformer-based hidden Markov model outperforms the standard self-attentive transformer model in terms of TER scores. In addition to the work on the neural hidden Markov model, we propose two novel metrics for machine translation evaluation, called CHARACTER and EED. These are easy-to-use and perform promisingly in the annual WMT metrics shared tasks. Es laden ein: die Dozentinnen und Dozenten der Informatik -- Stephanie Jansen Faculty of Mathematics, Computer Science and Natural Sciences HLTPR - Human Language Technology and Pattern Recognition RWTH Aachen University Theaterstraße 35-39 D-52062 Aachen Tel: +49 241 80-21601 sek(a)hltpr.rwth-aachen.de www.hltpr.rwth-aachen.de

9 months

Einladung: Informatik-Oberseminar Herr Arnab Chakrabarti

by Maaßen, Leany

+********************************************************************** * * * Einladung * * * * Informatik-Oberseminar * * * +********************************************************************** Zeit: Montag, 13.03.2023, 15:00-16:00 Uhr Der öffentliche Vortrag findet hybrid statt: Raum: Raum 5053.2 (großer B-IT-Hörsaal)/Informatikzentrum, Ahornstraße 55 Zoom: https://rwth.zoom.us/j/94314167045?pwd=VWt3TTUrUjQrbTRQcDUzYm8yMVcydz09 Meeting-ID: 943 1416 7045 Kenncode: 239326 Referent: Herr Arnab Chakrabarti, M. Sc. Lehrstuhl Informatik 5 Thema: Exploratory Pipeline for Analysis and Visualization of Large Information Spaces Abstract: Visual exploration of datasets having excessive data points and features leads to over-plotting and visual clutter, resulting in the loss of interpretability. This problem has been termed as "visual noise" in the field of high-dimensional data exploration. Our work in this thesis centers around the question of reducing visual noise and improving data interpretability in the quest to find interesting patterns within high-dimensional datasets. In this thesis, we present algorithms and workflows to tackle the so-called "curse-of-dimensionality". We proposed a visualization ranking model capable of comparing diverse visual structures within a unified framework. Finally, we present a novel visualization recommendation system capable of the process of generating meaningful visualizations that help users to explore and understand their data more efficiently. To evaluate the effectiveness of our solutions in a high-dimensional setting, we present a data exploration pipeline and test it within the Internet of Production (IoP) settings. Es laden ein: die Dozentinnen und Dozenten der Informatik _______________________________ Leany Maaßen RWTH Aachen University Lehrstuhl Informatik 5, LuFG Informatik 5 Prof. Dr. Stefan Decker, Prof. Dr. Matthias Jarke, Prof. Gerhard Lakemeyer Ph.D., JunProf. Dr. Sandra Geisler Ahornstrasse 55 D-52074 Aachen Tel: 0241-80-21509 Fax: 0241-80-22321 E-Mail: maassen(a)dbis.rwth-aachen.de

9 months

Einladung zum Informatik Oberseminar von Herrn Hendrik Simon, M.Sc.

by Hendrik Simon

+********************************************************************** * * * Einladung * * * * Informatik-Oberseminar * * * +********************************************************************** Zeit: Donnerstag, 4. Mai 2023, 13.30 Uhr Ort: Seminarraum 2202, Hauptbau, Ahornstr. 55 Referent: Hendrik Simon M.Sc. Lehrstuhl Informatik 11 Thema: Automatic Test Case Generation for PLC Software Abstract: Automatic test case generation for the purpose of bug finding or achieving coverage goals has recently evolved to a scalable technique that is nowadays used to find highly critical security bugs, e. g. by Microsoft. However, in the domain of Programmable Logic Controllers (PLCs), applications of this technique are rare and usually rely on tools and mechanisms that were not initially designed for this domain. In fact, a discussion on how to design such techniques with the peculiarities of PLC software in mind, is missing. At the same time, PLC software is typically used in safety critical environments where software errors pose significant threats to the environment or humans and may additionally result in significant financial losses. Mature automatic testing techniques for the PLC domain would, thus, be highly beneficial to further support software quality in this area. PLC software typically follows a cyclic execution scheme that involves a repeated process of reading input values, executing a (often state machine based) control program that relies on local variables and writing computed values to outputs. Although the cyclic execution resembles only a small change in the execution semantics, the impact on automatic testing techniques is significant. This dissertation provides insights and mechanisms to transfer automatic test case generation into the domain of PLC software. We conduct an in-depth discussion on related approaches and point out strengths and weaknesses in order to provide baseline knowledge that can be utilised in future developments in this field of research. Further, we introduce our own automatic test case generation approaches and exemplify their effectiveness on PLC software. We are able to show that the generation of branch coverage tests can be achieved significantly faster than with existing techniques, rendering our approaches more applicable for larger software. The focus of our techniques lies in the exploitation of state-machine based execution behaviour and the preservation of structural information in Sequential Function Chart. For the latter, our presented algorithm can achieve full coverage in a few seconds for programs that could only partly be covered within an hour by related approaches. Es laden ein: die Dozentinnen und Dozenten der Informatik

9 months

2023

2022

2021

2020

2019

2018

informatik-vortraege March 2023