Here you can find
query_builderLecture: Foundations of Artificial Intelligence (FAI)
In Foundation of Artificial Intelligence you will learn the basics of how computer try to mimic intelligence or human behavior.
After a brief, philosophical introduction to various kinds of intelligence we will focus on agents such as “robotic vacuum cleaner” which work autonomously to achieve goals.
Besides formal details such as agent architectures and environment specifications we will talk in greater detail how information is handled and decision are made from an algorithmic viewpoint. This entails for one finding information in graphs (uninformed search, informed search, local search), but also how an agent deals with uncertainty and not available information.
We will furthermore provide an introduction to the field of machine learning where we discuss various techniques to let a machine recognize objects in the real world (classification, clustering, regression), their prerequisites and how to evaluate them.
- History of AI
- Definition of Intelligence
- Agent Architecture
- Properties of Environments
- Uninformed Search (BFS, DFS)
- Informed Search (Greedy, A*)
- Local Search (Genetic Algorithms)
- Uncertainty / Probabilistic Models
- Machine Learning
- Classification (Naive Bayes, Decision Trees)
- Applications of AI
query_builderLecture: Knowledge Based Systems (KBS)
In this lecture we will discuss the various methods to transform information in a representation suited for computers and how to turn this information into knowledge.
We will start with the foundational questions of what knowledge even is and advance towards various way of encoding it. We discuss how the degree of similarity between pieces of information can be determined to answer questions such as “is the color black similar to white?” by using various human-curated resources.
In the subfield of machine learning we explore how the computer learns to recognize objects in the real world either from human-prepared data or all on its own. We will discuss several machine learning algorithm that are commonly used today and highlight some of their properties that one should consider when choosing one algorithm.
query_builderLecture: Natural Language Human-Computer Interaction (NL HCI)
In this lecture, we will work on a real world application of natural language based human-computer interaction. The best known application in this field are probably personal assistants on the smart phone like Siri, Cortana, or Ok Google.
We will have a look on the scientific foundations of such personal assistants including:
- language identification – how do I know that a request is made in German, English or French?
- question classification – How does the computer know what is being requested?
- entity disambiguation – What entities are mentioned in the question?
To enrich theory with practical experience we will build our own personal assistant system which we are going to steadily improve over the semester.
query_builderLecture: Language Technology
In this lecture we provide an overview over the large field of language technology. We start with examples in daily life that should be familiar to the most of you.
We will guide you through the various stages of how language, written or spoken, has to be processed in order to let computers work with language. This starts at trivial sounding tasks such as detection of word or sentence boundaries, recognizing a word’s part of speech (e.g. noun, verb, etc.) or identifying the language of a text.
More advanced topics are spelling and grammar correction, extraction of keywords and word sense disambiguation.
You will learn about the many challenges that lie hidden behind a trivial impression of the respective problems and learn concepts and their shortcomings to tackle those problems.
mode_editPractical Course: Text-Analysis Tools
In this practical class you will work on a task from the domain of natural language processing (NLP). In the beginning we will provide you with a tutorial of how to use a professional framework for working on problems of the NLP domain (DKPro). You will learn the basics of how to read, process and output data and how to use existing and add own user-created components in your project to fit them to your needs. Students that participate should have a profound knowledge of Java.
The center of this practical class is to learn how to use the framework in order to tackle a real-world NLP problem. This includes for one an analysis of the necessary steps to solve the problem, but also the implementation of your ideas in actual code without reinventing the wheel (using existing components in the framework for frequently reoccurring tasks). Thus, each participating student has to provide at the end of the semester an own solution / code for their respective chosen task. Students are encouraged to exchange information between each other and read through code-examples etc., but bulk copy-paste from websites (other people’s solution) is not accepted. We check for plagiarism!
At the end of the tutorial time you will choose a project. This project is either one of a list of proposed projects we provide or, if you have suited ideas, a self-selected topic. Popular topics are for instance determination of text similarity or sentiment analysis.
mode_editPractical Course: Master Research Projects
Offered on demand
Here are two examples that show how a project might look like:
1.) Exploring opinionated twitter conversations in the context of popular german TV-shows:
In this project 3 Masterstudents designed and implemented data collection, automatic topic- and sentiment- analysis and interactive visualisation.
2.) Geolocated Hatespeech detection:
In this project two Masterstudents designed and implemented a system to automatically detect and visualise geolocated tweets which contain
Get in contact!
It is advantageous if there are already interests or project ideas. But we also have our own ideas and we are happy to advise you.
This seminar introduces students to scientific writing of research papers that conforms to the standards of international conference papers.
The students will select a topic of a collection of prepared subject areas that are related to Natural Language Processing (NLP). During the seminar, students will go through the entire publication-creation life-cycle which composes of literature research, formulation of hypothesis, implementation / line of argument to (dis-)prove the hypothesis, writing of the actual paper and eventually presentation of the results.
We will provide guidance for these individual steps and how to structure and present information that it conforms to scientific standards.
The task of the student will be on one hand literature research and analysis to select relevant work and hypothesis generation grounded on this literature. On the other hand, the student will have to organize their texts in a way that people not directly involved in the creation process of the paper can understand and follow the line of thoughts. To get a better feeling for the challenging nature of writing scientific results down in an understandable fashion, students will peer-review their work (students exchange their current working version with their fellow students and mutually criticize their work for clarity, correctness/soundness and meaningful comparison(s)).
At the end of this seminar, the students present their result and hand in their paper that is graded for the same criterions that apply in peer-reviewing.
Each participant has to hand in an own paper, group work among two or more students is not permitted.
We offer thesis topics for Bachelor and Master students in the field of Natural Language Processing. Please contact us for details about currently available topics.
The following list of finished theses should give you an idea of the range of topics offered:
Identifying the native language of a text’s author,
The Influence of Smileys as a Feature in Opinion-based Text Classification,
Vergleichende Evaluation von UIMA-Textanalysekomponenten,
Vergleich von Methoden des Decompounding im Deutschen,
Identifying Semantically Equivalent Twitter Messages,
Entwicklung eines Frameworks zur Extraktion von Netzwerken aus Texten,
Automatische Korrektur von Präpositionen und Artikeln,
Vergleich von Textähnlichkeitsmaßen anhand von Geschwindigkeit und Leistungsverhalten,
Konfiguration und Verifikation von UIMA Pipelines,
Vergleich von Topic und N-Gramm Modellen in Wortvorschlagssystemen,
Predicting Cloze-Test Difficulty with Semantically Sensitive Language Models,
Automatische Bestimmung des Erstellungszeitpunktes von Textdokumenten,
Comparing spell-checking tools with respect to the quality of automatic corrections,
- Entwicklung eines deutschen Social Media Models für HeidelTime,