- informatik-vortraege - lists.rwth-aachen.de

Einladung: Informatik-Oberseminar Ali Athar
by Ali Athar 29 Feb '24

29 Feb '24

+********************************************************************** * * * Einladung * * * * Informatik-Oberseminar * * * +********************************************************************** Zeit: Dienstag, 5. März 2024, 10.00 Uhr Ort: Raum 025, Mies-van-der-Rohe Str. 15 (UMIC Gebäude) Referent: Ali Athar, M.Sc. Lehrstuhl Informatik 13 Thema: Segmenting and Tracking Objects in Video Abstract: Research to develop methods that can accurately localize and track objects in video has been ongoing for decades. Approaches capable of accomplishing this are highly sought after for a variety of applications including autonomous robots, self-driving vehicles, sports analytics, video editing, etc. Despite significant progress in recent times, the task is far from solved, in particular for challenging scenarios involving occlusions, motion blur, and camera ego-motion. In this thesis, we present a series of works that advance the state of research in this domain in various ways, as outlined below. Our first work, STEm-Seg, is an end-to-end trainable method for instance segmentation that models the input video as a single 3D space-time volume and relies on clustering per-pixel embeddings to segment and track objects. This differs from existing approaches, which largely follow the tracking-by-detection paradigm. Our novel formulation for these embeddings enables us to cluster the embeddings in an efficient and end-to-end learned fashion. The second work, called HODOR, is aimed at mitigating the need for densely annotated data for training video tracking methods. Specifically, it tackles the task of Video Object Segmentation (VOS) in a weakly supervised manner where it can be trained using static images or sparsely annotated video. To this end, we adopt a novel approach that encodes objects into concise descriptors. This is in contrast to existing approaches that predominantly learn space-time correspondences, which makes it challenging to train them in such a setting. Whereas the two aforementioned works propose network architectures, our third project proposes a dataset and benchmark called BURST that aims to unify the current, fragmented landscape of datasets in video segmentation research. BURST includes a benchmark suite that evaluates multiple tasks related to object segmentation in video with shared data and consistent evaluation metrics. The idea behind this is to facilitate knowledge exchange between the research sub-communities tackling these tasks and also to encourage the development of methods with multi-task capability. Finally, our fourth work, TarViS, can be seen as a logical continuation of the above in that it is a method that can tackle multiple video segmentation tasks. To achieve this, we decouple the task definition from the core network architecture and use a set of dynamic query inputs to specify the task-specific segmentation targets. This formulation enables us to train a single model jointly on a collection of datasets spanning multiple tasks (Video Instance/Object/Panoptic Segmentation). During inference, the model can switch between tasks by simply hot-swapping the input queries accordingly. Es laden ein: die Dozentinnen und Dozenten der Informatik

1 0

Einladung Informatik-Oberseminar Jan Rosendahl, 22. Februar 2024,
by Stephanie Jansen 16 Feb '24

16 Feb '24

********************************************************************** * * * Einladung * * * * Informatik-Oberseminar * * * *********************************************************************** Zeit: Donnerstag, 22. Februar 2024, 14:00 Uhr Ort: Raum 9222, E3, Informatikzentrum Zoom:https://rwth.zoom-x.de/j/67896121061?pwd=RlJTNUw5RFJYU0NwMVRWNEhvQ0EwZz09 Meeting-ID: 678 9612 1061 Kenncode: 493665 Referent: Jan Rosendahl, M.Sc.; Lehrstuhl Informatik 6 Thema: Attention-Based Machine Translation Using Monolingual Data Abstract: Neural networks present a major advance in modeling for statistical machine translation systems. In this dissertation, we focus on two central aspects of neural machine translation systems, namely the training data and the attention layer that connects the encoder and decoder. The parameters of a neural machine translation system are determined by minimizing the cross-entropy loss on a corpus of bilingual training data, i.e. a set of sentence pairs where one is the translation of the other. Since such sentence-aligned bilingual data is a scarce resource and availability depends on the language pair, we investigate using monolingual data to improve the performance of the machine translation system (Methods used: language model integration, monolingual pre-training, and back-translation). Inspired by existing work on alignment models, we also incorporate a first-order dependency in the encoder-decoder attention layer. In contrast with previous machine translation models, the transformer is a pure feed-forward model without any recurrent layers. That means that no information about the previous attention decision is input to the computation of the attention layer. Modeling attention with first-order dependencies allows the attention layer to access previous attention decisions, which is a prerequisite to express, e.g. source coverage. Es laden ein: die Dozentinnen und Dozenten der Informatik -- Stephanie Jansen Faculty of Mathematics, Computer Science and Natural Sciences Chair of Computer Science 6 ML - Machine Learning and Reasoning RWTH Aachen University Theaterstraße 35-39 D-52062 Aachen Tel: +49 241 80-21601 sek(a)ml.rwth-aachen.de www.hltpr.rwth-aachen.de

1 0

Einladung: Informatik-Oberseminar Herr Tim Niemueller
by Maaßen, Leany 20 Jan '24

20 Jan '24

+********************************************************************** * * * Einladung * * * * Informatik-Oberseminar * * * +********************************************************************** Zeit: Montag, 05.02.2024, 16:00-17:00 Uhr Ort: Raum 5053.2 (großer B-IT-Hörsaal)/Informatikzentrum, Geb. E2, Ahornstraße 55 Der Vortrag findet hybrid statt: https://rwth.zoom-x.de/j/67154623701?pwd=NnNIWWl0VnkzV0RiL3RyaWF3ZFlJQT09 Referent: Herr Tim Niemueller, Dipl.-Inform. LuFG Informatik 5 Thema: Planning and Execution for Mobile Robots Using Distributed Persistent Memory Abstract: Robots are expected to help humans in ever more capacities, in household en vironments as well as in industry. This demands increasing autonomy and resilience in order to decrease the need for human intervention and support to a minimum. Robotic systems are composed of a wide range of software components from different areas like perception, self-localization, navigation, decision making, and task execution. This easily leads to compartmentalization of the development, where the whole robot system becomes an afterthought when only best-in-class benchmarks are considered per component. Three particular problems motivated this thesis: First, data in robot systems is most often volatile, it briefly exists, is transferred, processed, used to make some decision, and eventually discarded, all within a short time window. If data is recorded, it is often done component-specific and not easily accessible. In this thesis, we tackle this problem by a document-oriented database that hooks into a robot system's middleware to collect as much data as possible. By using MongoDB as the data store and adding the ability to map middleware message into document structures, we gain query capabilities. We have used the database in several applications such as perception, performance and failure analysis, and action macro extraction. Second, the data exchange and coordination of multi-robot systems usually requires custom approaches. However, with the robot database already in place, using its distributed nature and augmenting it with triggers and computable elements it can fill this role as well, providing a unified robot memory. Third, robot task execution systems typically decouple domain modeling and execution flow specification, that is the specific details about an application the robot needs to know and the way it goes about choosing goals (which are typically few and human-defined). We have developed the CLIPS-based Executive that builds on Goal Reasoning to explicitly model the flow according to goals, of which there may be many and they can be considered and then selectively chosen for expansion, e.g., by invoking task planning, and then executed. Furthermore, by distributing the robot memory it can easily share data among a fleet of robots and aid in its coordination. Demonstrations of these systems in two mobile robot domains - domestic service robotics with a single robot and industrial factory logistics with groups of robots - show the applicability and versatility of the approaches developed. 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

1 0

Einladung: Informatik-Oberseminar Moritz Ibing
by Moritz Ibing 11 Dec '23

11 Dec '23

+********************************************************************** * * * Einladung * * * * Informatik-Oberseminar * * * +********************************************************************** Zeit: Freitag, 22. Dezember 2023, 11.00 Uhr Ort: Gebäude E3, Seminarraum 118, Ahornstr. 55 Referent: Moritz Ibing M.Sc. Lehrstuhl Informatik 8 Thema: Localized Control over the Latent Space of Neural Networks Abstract: Neural networks (NNs) are prevalent today when it comes to analyzing (classifying, segmenting, detecting, etc.) or generating data in all kinds of modalities (text, images, 3D shapes, etc.). They are so useful in these areas, because they have great representation power, while being easy to optimize and generalizing well to unseen data. However, their complexity makes them hard to interpret and modify. Neural networks are usually used to compute a mapping between the data space and a so-called latent space. Often we are interested in local properties of such a mapping. For example, we might want to slightly change the embedding of a data point to achieve a different classification. Such local modifications however are difficult, as NNs usually have globally entangled properties. In this work we will propose ideas how to deal with this problem. Local control is especially of importance for shape representations. It has been shown that NNs are well suited to represent these e.g. as parametric or implicit functions. However, when a global function is used, local supervision is hard to model. We therefore impose additional structure on the latent space of functional representations, making them easier to work with and more expressive. Such a structured representation makes downstream tasks easier, as we are more versatile regarding the shapes we can represent, we can make use of its regularity for the network design, and it allows a compressed encoding that can help to reduce memory consumption. Our focus will be on general shape generation, but we will also present more specific applications like shape completion or super-resolution among others. Our approaches set the state-of-the-art among generative models both in previously used metrics and a newly introduced measure we adapt for this purpose. Es laden ein: die Dozentinnen und Dozenten der Informatik

1 0

Einladung: Informatik-Oberseminar Henri Lotze
by Henri Lotze 05 Dec '23

05 Dec '23

+********************************************************************** * * * Einladung * * * * Informatik-Oberseminar * * * +********************************************************************** Zeit: Montag, 13. November 2023, 09.30 Uhr Ort: Informatikzentrum, E3, 2. Etage, Raum 9222 Referent: Henri Lotze M.Sc. Lehr- und Forschungsgebiet Theoretische Informatik Thema: Going Offline -- Delays, Reservations and Predictions in Online Computation Abstract: The study of classical online computation builds upon a rather simple model. The input to a given problem is not known in advance, decisions have to made immediately upon the arrival of a new element and these decisions are irrevocable. While this model is very well suitable to compute worst case bounds to a broad range of problems, there are some problems which are not well suited for this analysis. Naturally, problems that do not admit a constant approximation ratio in their offline formulation do not admit a constant competitive ratio in the worst case in the online formulation. However, some problems that are approximable still do not admit a constant competitive ratio when analyzed in the online case. For a subset of these problems, such as the Online Simple Knapsack problem and the class of general F-Node- and F-Edge-Deletion problems, the reason for this seems to be that a single, specific bad decision by an algorithm can be arbitrarily punished by an adversary. In this talk, we explore modifications of the classical online model with the aim to find natural models that on the one hand cover a large range of problems and on the other hand work against pathological kinds of instances that restrict an algorithm from obtaining a constant competitive ratio. To this end, we study three modifications of the classical online model. The models that we study are that of "Late Accepts" -- with Advice --, which allows to postpone decisions in online minimization problems until a current partial solution does not uphold a certain property anymore. The second one is that of "Reservations", in which any decision may be postponed for a cost proportional to the gain of an item. Finally, the model of "Bounded Predictions" allows an online algorithm to see the complete instance in advance, with a caveat: There is noise on the instance -- controlled by an adversary -- and each element may actually deviate from its prediction, up to factor that is known to the algorithm. We are able to partially classify the advice complexity of the complete class of both F-Node-Deletion and F-Edge-Deletion problems. We give tight bounds on the competitive ratio for the Online Simple Knapsack problem with Reservations for the whole range of possible reservation costs, i.e. for a factor between 0 and 1. Finally, we give partially tight bounds for the Online Simple Knapsack problem with Bounded Predictions for the whole range of possible noise factors on the size of the items. Es laden ein: die Dozentinnen und Dozenten der Informatik

2 1

[sekretariate] Einladung: Informatik-Oberseminar Jana Berger
by Elke Ohlenforst 04 Dec '23

04 Dec '23

+********************************************************************** * * * Einladung * * * * Informatik-Oberseminar * * * +********************************************************************** Zeit: Mittwoch, 13. Dezember 2023, 16:00 Uhr Ort: Informatikzentrum Ahornstr. 55, Raum 5053.2 (B-IT) Referentin: Jana Berger M.Sc. Lehrstuhl Informatik 2 Thema: Applying Software Model Checking: Experiences and Advancements Abstract: We present a comprehensive investigation into the applicability of formal methods for software model checking, with a particular emphasis on the automotive sector. The work is grounded in a practical study carried out with Ford Motor Company, offering a basis for the exploration and enhancement of both commercial and academic model checking tools. In the first part of the study, we collaborated with Ford Motor Company to formalise their natural language requirements for two automotive case studies employing the commercial C code model checker, BTC EmbeddedPlatform. Our investigation focused on the verification of auto-generated C code from Simulink models, analysing the associated challenges and providing practical recommendations for both engineers and tool developers. Building on this foundation, we then applied academic C code model checkers to the same case studies, comparing their performance against BTC EmbeddedPlatform. This comparison yielded a fresh set of insights and recommendations aimed at enhancing academic model checking tools. The second part of this work addresses the development of new tools and methods aimed at advancing the state of the art in software model checking. First, we designed NITWIT, a novel violation witness validator based on C code interpretation. This approach yields fast execution with a minimal memory footprint and is agnostic of existing model checkers. Next, we developed a tool for writing specifications with formal semantics, tailored specifically for the automotive industry. This tool is capable of auto-generating (a) specifications in Simulink, BTC and pure C code for verification, streamlining the formal verification process, and (b) natural language representations with fixed semantics for stakeholders and non-experts to review and use. Finally, to better facilitate the testing and benchmarking of C code model checkers, we built a tool that generates C code. The style of the generated code borrows from the two case studies, thus providing a valuable resource for stress-testing model checkers while respecting non-disclosure agreements, offering a wide variety of controllable code metrics, features and transformers. In summary, this thesis provides new and relevant insights as well as novel tools aimed at improving the application and comparison of software model checking, particularly in the context of the automotive industry. It offers practical solutions to real-world challenges, and helps to bridge the gap between commercial and academic approaches to software model checking. Es laden ein: die Dozentinnen und Dozenten der Informatik

1 0

Einladung: Informatik-Oberseminar Herr Fabian Ohler
by Maaßen, Leany 25 Nov '23

25 Nov '23

+********************************************************************** * * * Einladung * * * * Informatik-Oberseminar * * * +********************************************************************** Zeit: Freitag, 08.12.2023, 13:00-14:00 Uhr Ort: Raum 5053.1 (kleiner B-IT-Hörsaal)/Informatikzentrum, Ahornstraße 55 Referent: Herr Fabian Ohler, M. Sc. Lehrstuhl Informatik 5 Thema: Die Rolle funktionaler Domänenmodelle in der Entwicklung allianzgetriebener Softwareplattformen Abstract: Organisationen können durch die Bildung von Allianzen gemeinsame Ziele verfolgen. Gestalten sie dabei zusammen eine Plattform, müssen bereits in einem frühen Stadium eine Vielzahl an Prozessen zusammengeführt und die Verteilung der Funktionen auf die Akteure geklärt und koordiniert werden. Im Vortrag wird die methodische und technische Unterstützung für die unternehmensübergreifenden, von Kollaboration geprägten Entwurfsaktivitäten der zugehörigen Systemarchitektur am Beispiel einer Plattformentwicklung zur Kooperation öffentlicher Verkehrsträger und Verkehrsverbünde vorgestellt. Dafür wird eine empirisch ausgearbeitete Methode zur Herleitung eines funktionalen Domänenmodells und für die Weiterentwicklung zu einer Systemarchitekturbeschreibung präsentiert. Die Rolle des funktionalen Domänenmodells in diesem Kontext wird beleuchtet. Ergänzend werden Software-Werkzeuge zur Unterstützung der Entwicklungsphasen der Feinspezifikation und Validierung vorgestellt. Diese basieren auf in Experteninterviews erhobenen Erkenntnissen über die Defizite bislang eingesetzter Werkzeuge sowie daraus abgeleiteten Anforderungen. Die erarbeitete Methode wurde bereits erfolgreich in zwei Domänen eingesetzt, so dass eine Übertragbarkeit auf weitere Domänen zu erwarten ist. 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

1 0

Einladung: Informatik-Oberseminar Istvan Sarandi
by Sárándi, István 24 Nov '23

24 Nov '23

+********************************************************************** * * * Einladung * * * * Informatik-Oberseminar * * * +********************************************************************** Zeit: Dienstag, 5. Dezember 2023, 13.00 Uhr Ort: Großer b-it Raum (5053.2), Geb. E2, Informatikzentrum, Ahornstr. 55 Der Vortrag findet hybrid statt. Zoom: https://rwth.zoom-x.de/j/63672895840?pwd=L2ZoLzRDMzV3dzIrZ2hsNTAwaFNldz09 Meeting ID: 636 7289 5840, Passcode: 964382 Referent: Istvan Sarandi, M.Sc. Lehrstuhl Informatik 13 Thema: Robust and Efficient Methods in Visual 3D Human Pose Estimation Abstract: Computer vision algorithms for perceiving humans in the real world are crucial for several impactful emerging technologies, including self-driving cars and mobile service robots. In this talk, I will present three contributions to improving the state of the art in deep learning-based 3D human pose estimation, that is, localizing major anatomical landmarks of the human body in 3D space from RGB images only. The central themes are robustness and efficiency, which constitute the main challenges in robotics applications. We start by addressing robustness to occlusions, i.e., when objects block the line of sight between the person and the camera. After presenting the first systematic study of how occlusions deteriorate 3D pose estimation accuracy, we propose to mitigate the problem using an effective synthetic occlusion data augmentation strategy. We then turn to the problem of truncation, i.e., when only a part of the body is within the camera's field of view. We develop a truncation-robust heatmap representation, which also allows learned recovery of the metric scale. Building upon this capability, I present an end-to-end learned absolute pose estimation method called MeTRAbs, for robustly reconstructing human poses in the camera's reference frame at state-of-the-art accuracy. In the third part, I present the largest-scale experiment reported in the 3D pose literature to date, by merging a total of 32 datasets, in the pursuit of improved in-the-wild generalization, outside the controlled environments of motion capture studios. We overcome the challenge of differently annotated datasets through a novel affine-combining autoencoder formulation for capturing the common information in different landmark annotation formats. Importantly, these methods can run on low-powered robot hardware in real time. I conclude with a discussion of possible extensions to the presented works, as well as exciting future challenges for the field as a whole. Es laden ein: die Dozentinnen und Dozenten der Informatik

1 0

Einladung: Informatik-Oberseminar Jannis Klinkenberg
by Klinkenberg, Jannis 20 Nov '23

20 Nov '23

+********************************************************************** * * * Einladung * * * * Informatik-Oberseminar * * * +********************************************************************** Zeit: Freitag, 01. Dezember 2023, 09.00 Uhr Ort: IT Center, Kopernikusstraße 6, Seminarraum 003 Referent: Jannis Klinkenberg M.Sc. Lehrstuhl Informatik 12 Thema: Reactive Runtimes for Parallel Programming Models on Shared, Distributed, and Heterogeneous Memory Systems Abstract: The persistent drive to enhance the scale and precision of scientific simulations over the past decades has created an ever-growing demand for computational resources. This, in turn, has catalyzed several innovations in the architectural design and manufacturing procedures of computer systems, such as the introduction of multi-core processors and multi-socket shared memory systems, which are typically operated in High Performance Computing (HPC) installations. Historically, scientific applications have primarily been developed with the presumption of a uniform execution environment, where every compute node and core within an installation operates at a consistent, unchanging speed and where the execution time can be accurately predicted. However, in the last decades, both hardware and software have become increasingly complex and often exhibit dynamic execution behavior, causing performance fluctuations and run time variability. Consequently, a priori load balancing within and between compute nodes becomes increasingly challenging in such environments. Applications as well as runtime systems therefore demand for new techniques to be able to dynamically react to changing execution conditions and efficiently balance the load. This thesis presents reactive concepts, runtime implementations, and runtime extensions designed to address the growing complexity of both hardware and software. The objective is to provide portable, vendor-independent solutions to improve application performance, minimize run time variability and mitigate impending load imbalances. Locality-aware task scheduling extensions in OpenMP improve the data locality on contemporary shared memory NUMA architectures by dynamically identifying physical data locations and reactively adjusting the task distribution and scheduling. Further, continuous performance introspection combined with reactively migrating or replicating tasks in distributed memory can efficiently detect and mitigate emerging imbalances at execution time. Lastly, abstracting regular memory allocation allows specifying additional requirements or hints regarding how the data is used throughout the execution, which can be exploited to dynamically guide the data placement on systems with heterogeneous memory. Systematic evaluations demonstrate the effectiveness of the presented concepts and show that placing the burden on runtime systems to proficiently handle these low-level aspects is crucial to achieve high performance on current and future architectures. Es laden ein: die Dozentinnen und Dozenten der Informatik

1 0

Re:
by Nagl, Manfred 26 Oct '23

26 Oct '23

Das war ein Fehlversuch. Sorry M. Nagl Von: Nagl, Manfred Gesendet: Mittwoch, 25. Oktober 2023 16:34 An: Vortraege(a)informatik.rwth-aachen.de Betreff: Prof. Dr.-Ing. Dr. h.c. Manfred Nagl, Emeritus Lehrstuhl Software Engineering RWTH Aachen University 52074 Aachen Tel. +49 241 8021350 Mobil +49 171 5463727

1 0