+**********************************************************************
*
*
* Einladung
*
*
*
* Informatik-Oberseminar
*
*
*
+**********************************************************************
Zeit: Freitag, 12. Juli 2019, 10.00 Uhr
Ort: Informatikzentrum, E3, Raum 9222
Referent: Dipl.-Inform. Malte Nuhn
Thema: Unsupervised Training with Applications in Natural Language
Processing//
Abstract:
The state-of-the-art algorithms for various natural language processing
tasks require large amounts of labeled training data. At the same time,
obtaining labeled data of high quality is often the most costly step in
setting up natural language processing systems.Opposed to this,
unlabeled data is much cheaper to obtain and available in larger
amounts.Currently, only few training algorithms make use of unlabeled
data. In practice, training with only unlabeled data is not performed at
all. In this thesis, we study how unlabeled data can be used to train a
variety of models used in natural language processing. In particular, we
study models applicable to solving substitution ciphers, spelling
correction, and machine translation. This thesis lays the groundwork for
unsupervised training by presenting and analyzing the corresponding
models and unsupervised training problems in a consistent manner.We show
that the unsupervised training problem that occurs when breaking
one-to-one substitution ciphers is equivalent to the quadratic
assignment problem (QAP) if a bigram language model is incorporated and
therefore NP-hard. Based on this analysis, we present an effective
algorithm for unsupervised training for deterministic substitutions. In
the case of English one-to-one substitution ciphers, we show that our
novel algorithm achieves results close to human performance, as
presented in [Shannon 49].
Also, with this algorithm, we present, to the best of our knowledge, the
first automatic decipherment of the second part of the Beale
ciphers.Further, for the task of spelling correction, we work out the
details of the EM algorithm [Dempster & Laird + 77] and experimentally
show that the error rates achieved using purely unsupervised training
reach those of supervised training.For handling large vocabularies, we
introduce a novel model initialization as well as multiple training
procedures that significantly speed up training without hurting the
performance of the resulting models significantly.By incorporating an
alignment model, we further extend this model such that it can be
applied to the task of machine translation. We show that the true
lexical and alignment model parameters can be learned without any
labeled data: We experimentally show that the corresponding likelihood
function attains its maximum for the true model parameters if a
sufficient amount of unlabeled data is available. Further, for the
problem of spelling correction with symbol substitutions and local
swaps, we also show experimentally that the performance achieved with
purely unsupervised EM training reaches that of supervised training.
Finally, using the methods developed in this thesis, we present results
on an unsupervised training task for machine translation with a ten
times larger vocabulary than that of tasks investigated in previous work.
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
Ahornstraße 55
D-52074 Aachen
Tel. Frau Jansen: +49 241 80-216 06
Tel. Frau Andersen: +49 241 80-216 01
Fax: +49 241 80-22219
sek(a)i6.informatik.rwth-aachen.de
www.hltpr.rwth-aachen.de
Tel: +49 241 80-216 01/06
Fax: +49 241 80-22219
sek(a)i6.informatik.rwth-aachen.de
www.hltpr.rwth-aachen.de
+**********************************************************************
*
*
* Einladung
*
*
*
* Informatik-Oberseminar
*
*
*
+**********************************************************************
Zeit: Montag, 4. Oktober 2021, 10.00 Uhr
Ort: Zoom-Videokonferenz (https://rwth.zoom.us/j/93693721093?pwd=OWo0eXJaajgram9lY1hxUVE1N0lXZz09)
Referent: Svenja Noichl M.Sc.
Lehr- und Forschungsgebiet Informatik 9
Thema: InfoBiTS: Informatische Bildung für Technikferne Seniorinnen und Senioren
Abstract:
Digitale Kompetenzen gewinnen im Zuge der fortschreitenden Digitalisierung in allen Bereichen des alltäglichen Lebens zunehmend an Relevanz. Dies gilt auch für ältere Personen, welche bisher wenig oder keine Berührungspunkte mit digitalen Technologien hatten. Unter Digitalkompetenz oder digitaler Kompetenz wird hier die Kombination aus Medienkompetenz und Informatikkompetenz verstanden, also die Kombination aus Aspekten der Medienkunde, Medienkritik, Mediennutzung und Mediengestaltung sowie Grundkenntnissen über unter anderem die Funktionsweisen von Informatiksystemen. Während in bestehenden Angeboten für ältere Menschen, wie z. B. Computer-, Smartphone- oder Tabletkursen, bei Peer-Learning Angeboten oder bei Technikbegleitung und Sprechstunden, zumeist die Medienkompetenz adressiert wird, soll das hier entwickelte Angebot einen größeren Fokus auf die Informatikkompetenz legen. Durch die Vermittlung von Ideen und Konzepten aus der Informatik, soll so über die reine Nutzungskompetenz digitaler Endgeräte hinaus, übertragbares Wissen dieser Domäne gefördert werden. Die Zielgruppe sind Personen ab 50 Jahren, welche keine bis wenig Vorerfahrung mit digitalen Technologien besitzen. Um den Lernerfolg in dem für die Zielgruppe neuen Gebiet bestmöglich zu unterstützen, ist die Berücksichtigung der Geragogik unerlässlich. Drei wichtige Aspekte stellen hierbei (1) das Lernen mit Gleichgesinnten, (2) das Lernen in einem geschützten Raum sowie (3) schnelle Hilfe bei Fragen und Problemen dar. Das hier entwickelte Angebot setzt daher nicht auf eine alleinige Nutzung der entwickelten Lernapp (InfoBiTS), sondern bettet diese in ein Kurskonzept ein. Bewährt hat sich dafür ein Workshopsetting. Ein Onlinesetting ist mit Einschränkungen ebenfalls möglich. Die InfoBiTS-App beinhaltet vier Lernmodule, welche sich mit den Themen Kommunikation, Funktionsweise des Internets, mobile Geräte und das Internet sowie Datenschutz und Datensicherheit befassen. Die Module adressieren hierbei jeweils Kompetenzen aus dem Curriculum für Seniorinnen und Senioren, welches im Rahmen dieser Arbeit entwickelt wurde und auf nationalen und internationalen Schulcurricula sowie Interessen der Zielgruppe, welche in einem Fragebogen mit 123 Teilnehmenden erhoben wurden, basiert. Für die konkrete Themenauswahl waren darüber hinaus Themen aktueller Relevanz sowie Anknüpfungspunkte an den Alltag der Zielgruppe, z. B. die Kommunikation mit (entfernt lebenden) Kindern und Enkeln, maßgeblich. Während die Pilotstudie im Workshopsetting durchgeführt wurde, erfolgte die Evaluation, aufgrund der Einschränkungen durch die vom Coronavirus SARS-CoV-2 ausgelöste Pandemie, im Onlinesetting. Insgesamt nahmen 19 Personen zwischen 50 und 84 Jahren teil. Die Evaluation zeigte insbesondere, dass sich das Gefühl der Kontrolle im Umgang mit und über die Technik im Vergleich zu vor dem ersten Modul und nach dem letzten Modul signifikant verbesserte. Weiterhin deuten die Ergebnisse einer modulbezogenen Selbsteinschätzung sowie die Bearbeitung der Aufgaben innerhalb der Module darauf hin, dass die angestrebten Lernziele in den Modulen weitestgehend erreicht werden konnten, was auf eine Förderung der adressierten Kompetenzen aus dem Curriculum hindeutet. Letztlich weißt die Auswertung des DigComp 2.1, dem europäischen Referenzrahmen für digitale Kompetenzen, nicht nur auf eine Verbesserung der erwarteten Kompetenzen hin, sondern auch auf eine Verbesserung weiterer Kompetenzen, wie beispielsweise im Bereich des Umgangs mit technischen Problemen.
Es laden ein: die Dozentinnen und Dozenten der Informatik
+**********************************************************************
*
*
* Einladung
*
*
*
* Informatik-Oberseminar
*
*
*
+**********************************************************************
Zeit: Donnerstag, 23. September 2021, 11.00 Uhr
Zoom: https://rwth.zoom.us/j/92021395779?pwd=ckRtKzJhcHFmeWxaU3YxSEZRWTJTUT09
Meeting-ID: 920 2139 5779
Kenncode: 534446
Referent: Dimitri Bohlender M.Sc.
Lehrstuhl Informatik 11
Thema: Symbolic Methods for Formal Verification of Industrial Control Software
Abstract:
Many of the systems that we rely on, and interact with on a daily basis, are driven by
software. Unfortunately, design and implementation of such systems is naturally prone to
error, as it is done by humans and involves reasoning about the vast number of states a
system may reach. While testing is the common approach to alleviating the risk of writing
faulty software, it can only help with finding errors, but not prove their absence.
By way of contrast, formal methods have mathematical foundations and enable rigorous
reasoning about the behaviour of formally modelled systems. In particular, they
give rise to formal verification procedures for proving a system's compliance with certain
formal specifications. Although many such procedures can simplistically be thought of
as an automatic exploration of a system's state space, the explicit enumeration of each
reachable state can often be avoided. To this end, symbolic methods reason about many
states at a time by representing sets of states and transition relations as logical formulas.
My thesis is concerned with advances in symbolic methods for formal verification of
the software-driven reactive systems that are used in the setting of industrial automation.
While these systems often operate in safety-critical environments, the specifications and
peculiarities of the domain impede the use of existing verification machinery for general-purpose
programming languages, leaving engineers in need of computer-aided reasoning
about the control software semantics. Our contributions address this issue in platform-agnostic
ways, but are presented using the example of programmable logic controllers
which are tailored to the industrial automation domain and therefore widely used.
In this talk, we give an overview over our contributions with a focus on the approaches
that leverage constrained Horn clause solving. After a short introduction, we sketch how
a logical characterisation of control software safety in terms of constrained Horn clauses
can be derived from reactive systems safety foundations. To exploit the modularity of
control software, the characterisation is also extended and combined with mode
abstraction – a domain-specific analysis for approximating the state space. Furthermore,
we present approaches for the design and verification of control software that is resilient
to potential restarts of the controller. We show how the choice of persistent variables can
be reduced to parameter synthesis, and solved by extending the previous verification procedures.
Es laden ein: die Dozent*innen der Informatik
+**********************************************************************
*
*
* Einladung
*
*
*
* Informatik-Oberseminar
*
*
*
+**********************************************************************
Zeit: Freitag, 24. September 2021, 09.00 Uhr
Ort: Zoom Videokonferenz
https://rwth.zoom.us/j/98629457476?pwd=WHB2U0FUMzVTaWU2S0paeUJESy8vdz09
Meeting ID: 986 2945 7476
Passcode: 073217
Referent: Paul Voigtlaender, M.Sc.
Lehrstuhl Informatik 13
Thema: Video Object Segmentation and Tracking
Abstract:
Video Object Segmentation (VOS) is the computer vision task of segmenting
generic objects in a video given their ground truth segmentation masks
in the first frame. Strongly related are the tasks of single-object
tracking (SOT) and multi-object tracking (MOT), where one or multiple
objects need to be tracked on a bounding box level. All these tasks
are highly related and have important applications like autonomous
driving and video editing. At the same time, all of these tasks remain
very challenging till today. In this talk, we present our work on
VOS, MOT, and SOT.
Firstly, we present a VOS method, FEELVOS, which follows the feature
embedding-learning paradigm. FEELVOS is one of the first VOS methods
which use a feature embedding as internal guidance of a convolutional
network and learn the embedding end-to-end with a segmentation loss.
Following this approach, FEELVOS achieves strong results while being
fast and not requiring test-time fine-tuning. This feature embedding-learning
paradigm together with end-to-end learning has by now become the
dominating approach for VOS.
We further extend the popular MOT task to Multi-Object Tracking and
Segmentation (MOTS) by requiring methods to also produce segmentation
masks. We propose a semi-automatic labeling method and use it to annotate
two existing MOT datasets with masks. We release the resulting KITTI MOTS
and the MOTSChallenge benchmarks together with new evaluation measures and
a baseline method. Additionally, we promote the new MOTS task by hosting a
workshop challenge. MOTS is a step towards bringing the communities of VOS
and MOT together to facilitate further exchange of ideas.
Finally, we present Siam R-CNN, a Siamese re-detection architecture
based on Faster R-CNN, to tackle the task of long-term single-object
tracking. In contrast to most previous long-term tracking approaches,
Siam R-CNN performs re-detection on the whole image instead of a local
window, allowing it to recover after losing the object of interest.
Additionally, we propose a tracklet dynamic programming (TDPA) algorithm
to incorporate spatio-temporal context into Siam R-CNN. Siam R-CNN
produces strong results for SOT and VOS, and performs especially well
for long-term tracking.
+**********************************************************************
*
*
* Einladung
*
*
*
* Informatik-Oberseminar
*
*
*
+**********************************************************************
Zeit: Donnerstag, 23. September 2021, 14.00 Uhr
Ort: Zoom-Videokonferenz und Raum 9222
Für die Teilnahme vor Ort wird um Anmeldung gebeten
Link:
https://rwth.zoom.us/j/95250295111?pwd=cjhFMGk0bStZcGNvcmYrYURPWnQ0dz09
Referent: Martin Ritzert M.Sc.
Lehrstuhl für Informatik 7
Thema: Learning on Graphs with Logic and Neural Networks
Abstract:
In the domain of graphs we show strong connections between logic and
machine learning in both theory and practice. In a purely theoretical
framework we develop sublinear machine learning algorithms for
supervised learning of logical formulas on various graph classes.
Further we show that learning first-order logic on arbitrary graphs is
intractable unless P=NP. At the intersection of theory and practice, we
prove an equivalence between graph neural networks and the 1-dimensional
Weisfeiler-Leman algorithm. As a practical application, we approximate
combinatorial problems with recurrent graph neural networks. The
proposed architecture is unsupervised and can be applied to all maximum
constraint satisfaction problems.
Es laden ein: die Dozentinnen und Dozenten der Informatik