Projekt Kognitive Systeme/Project Cognitive Systems

General Information

  • For a general course description please read the corresponding pages (Bachelor, Master) from the WIAI module guide.
  • You find administrative information at UnivIS.
  • To get up-to-date information you should sign up for the course in the virtual campus.
  • At the first meeting (normally, first week of the start of lectures) we will give an introduction in suggested project topics.
  • A topic is usually worked on jointly by two to four students (except for 15 ECTS projects). Groups can be and mostly are formed in the first week of the course (for example related to topics of interests, etc.). Groups are established typically in the second week via posting in the vc (names and topic).
  • We strongly recommend students interested in writing a thesis in our group to participate in a project before!
  • Project topics presuppose knowledge and skills as introduced in the lecture/practive modules of the KogSys group.

Proposed Topics

  • We have a shared pool of topics for bachelor- and master-level projects as well as for bachelor- and master theses, which are described in the VC-Course "KogSys: General Information": VC-Course.
  • Selection of research domains addressed in the projects: Inductive programming, applications of machine learning to image classification (e.g., pain faces, plant growth), grammar learning, human-robot interaction with NAO or Pepper, cognitive models of analogy and inductive learning.
  • A list of past and current project topics is given below.

Current and Past Projects

SS 2024

6 ECTS Project:

Neuro-symbolic AI:

The emerging research field of Neuro-symbolic AI (NeSy), often also Hybrid AI, deals with combining the representational power and interpretability of symbolic AI and Machine Learning systems with the computational capabilities of (deep) neural networks. However, the implementation of said combination is often trivialized in conversation. The goal of this project is to come up with and implement a Neuro-symbolic system, with the main focus on the interface between the two (or more) components.

Supervisor: Johannes Langer

15 ECTS project: Several topics in the context of intelligent tutor systems (ITS)

Supervisors: Adrian Völker, Alisa Münsterberg & Ute Schmid

WS 2023/2024

6 ECTS Project:

Evaluating AI Systems and ML Models:

In order to design AI systems and ML models that make safe, reliable, beneficial and ethically responsible decisions, appropriate evaluation data sets, methods, and frameworks are needed. In this project, students work in groups to develop technical or empirical approaches for evaluating AI systems and ML models. In particular, the transformative nature of systems such as Chat-GPT, the need for Explainable Artificial Intelligence in high-stakes applications, and the question of the existence of a general Artificial Intelligence, give rise to exciting research topics for evaluation. Starting from a self-selected data set or empirical question related to Artificial Intelligence and Machine Learning, students implement an experiment and assess the AI system or ML model using appropriate evaluation criteria. The topics of the project cover ethics and veracity of AI evaluation, metrics and frameworks for evaluation, experimental designs for valid, reliable and objective AI evaluation, re-evaluation of high-impact AI research, AI evaluation by measuring intelligence and possibly novel evaluation techniques, like cognitive XAI.

Supervisors: Bettina Finzel & Ute Schmid

15 ECTS Project:

XAI Demonstrator; (KMU-KI-EZ), Supervisors: Judith Knoblach & Ute Schmid

SS 2023

15 ECTS Projekt: ML/Cobot Demonstrator; (KMU-KI-EZ) Supervisor: Judith Knoblach & Ute Schmid

WS 2022/2023

  • 15 ECTS Project: Active Learning for interactive labeling of RGB UAV forest data; (BaKIM) Supervisor: Jonas Troles
  • 15 ECTS Project: Combining CRP and ILP for explanations in relational visual domains; (PainFaceReader) Supervisor: Bettina Finzel
  • 15 ECTS Project: Post-processing of Language Models with Logical Rules for Increased Performance in Named Entity Recognition; (IIS/CAI) Supervisor Emanuel Slany & Ute Schmid

SS 2022

  • Project: Statistical Machine Learning -- joint with Survey Statistics; Supervisor: Sarem Seitz
  • 15 ECTS Project: The Object-Centric Concept Learner: An Interpretable Framework for Learning Relational Visual Concepts; Supervisor: Bettina Finzel

WS 2021/2022

  • 15 ECTS Project: Interactive Segmentation Learning; (BaKIM)  Supervisor: Jonas Troles & Ute Schmid

SS 2021

  • Project: Statistical Machine Learning -- joint with Survey Statistics; Supervisor: Sarem Seitz
  • 15 ECTS Project: Automatic Verification of Convolutional Neural Networks: A Landmark-based Approach for Facial Action Unit Classification; (PainFaceReader) Supervisor: Bettina Finzel
  • 15 ECTS Project: XAI for Intelligent Tutor Systems -- Leaf Classification by Contrastive Explanations, Supervisor: Ute Schmid

WS 2020/2021

In winter term three topics are supervised by Ute Schmid which are in collaboration with different research partners:

  • 15 ECTS Project: Intelligent Tutor Systems -- Algorithmic Debugging and Feedback by Analogy (Subkraki), Supervisor: Ute Schmid
  • 15 ECTS Project: Dare2Del Demo, Supervisor: Ute Schmid
  • 6 ECTS Project: Learning and Explaining Driving Recommendations (with fortiss/IBM Munic), Supervisor: Ute Schmid

SS 2020

Hands-on Machine Learning


Das Projekt findet im SoSe 2020 wohl ausschließlich digital statt.
Das Projekt richtet sich bevorzugt an Masterstudierende, die das Modul Machine Learning bereits erfolgreich absolviert haben. Ziel ist es, sich vertieft mit ausgewählten Ansätzen des maschinellen Lernens auseinanderzusetzen und praktische Erfahrung im Umgang mit Maschine Learning Tools (wie TensorFlow, Keras, R, RapidMiner) zu sammeln. Der Anwendungsbereich wird jeweils in Bezug zu einem aktuellen Drittmittelprojekt, einem Projekt mit einem Anwendungspartner oder einem Promotionsprojekt der Gruppe Kognitive Systeme gewählt. Studierende im Bachelor Angewandte Informatik können  das Projekt auch ohne Vorkenntnisse in ML belegen, wenn sie das Modul Einführung in die KI erfolgreich absolviert haben. Hier wird dann ein wissensbasierter Ansatz für den selben Anwendungsbereich umgesetzt.

Spezifisches Thema SoSe 2020: ML aus Multi-Sensor-Daten zur Phänotypisierung

Im Projekt soll das Wachstumsverhalten von zwei Arten von Pflanzen in Abhängigkeit von Umweltbedingungen erfasst und zur Analyse oder Klassifikation (z.B. Erkennung der Pflanzenart oder von Wachstums-Anomalien) genutzt werden. Daten werden dabei über einen "intelligenten Topf" (auf Arduino-Basis) erhoben, der mit verschiedenen Sensoren ausgestattet ist, deren Meßwerte via WLAN Anbindung übermittelt werden. Das Auslesen der Sensoren und die Anbindung an ein WLAN vermitteln praktische Einblicke in die IoT-Programmierung. Darauf aufbauend sollen entweder wissensbasierte KI-Ansätze oder Methoden des maschinellen Lernens für Anomalie-Erkennung oder Klassifikation umgesetzt werden. Mögliche Ansätze sind: Fuzzy-Regeln, probabilistisches logisches Programmieren, Zeitreihenanalyse, Klassifikation mit tiefen Netzen.

Supervisor: Oliver Scholz, Ute Schmid

SS 2019

Bots With Blocks: Wenn Computer Tetris spielen lernen


In diesem Projekt beschäftigen wir uns mit Tetris und der Implementierung eines Bots dafür. Während bei anderen Spielen mit großen Suchräumen wie Schach und Go nur am Ende einer Partie klar ist, ob man gewonnen hat oder nicht (binärer Outcome), werden bei Tetris für das Abräumen von Lines direkt auch innerhalb des Spiels Punkte vergeben. Dadurch lassen sich effizienter Heuristiken für Spielstrategien finden. Außerdem können Spiel-Zustände intuitiver bewertet werden, da klar ist, dass bestimmte Block-Konfigurationen wünschenswerter sind als andere (breite/niedrige Berge vs. schmale/hohe). Wir wollen uns anhand einer vorhandenen Tetris-Library anschauen, welche Herangehensweisen es an das Implementieren von Algorithmen gibt, die Tetris automatisch spielen.

Die Studierenden werden in Gruppen an den Implementierungen arbeiten. Das Endresultat soll ein wissenschaftliches Paper sein aufbauend auf Experimenten mit dem implementierten Bot.

Supervisor: Johannes Rabold

WS 2018/2019

15 ECTS project: Instance Segmentation of 3D Point Clouds of Wheat Field Data with Deep Learning Methods

Description: Implementation of Future Work for Instance Segmentation of 3D Point Clouds with SGPN.

Student: Melanie Vogel

Supervisor: Johannes Rabold

SS 2018

15 ECTS project: Learning to Discriminate Categorial Feature Values -- An Intelligent Biology App for Tree Classification based on Leaf Forms

Description: Biological classifications often rely on categorial -- often binary -- decisions about feature values. For example, trees can be classified by various characteristics of their leafs such as round vs. oval form or smooth v. ripped edge, Most classification apps rely on a correct recognition of feature values by the user. That is, they ask for the value of one feature after the other to traverse a path in a decision tree. However, in many domains, it is difficult for a novice to recognize feature values correctly. For example, the edge might be wavy but the novice classifies it as rippled. Our goal is to develop a training app which can give hints to the user if she or he does not give the correct feature value. Thus, the app will be a trainer for exact observation. As a first step, in the project, it is explored whether a deep learning classifier will be sufficiently accurate to recognize trees/leafs.

Supervisor: Ute Schmid, Student: Florian Ellinger 

15 ECTS project: Deep Learning for Instance Segmentation (3D Point Clouds of Wheat Fields)

Description: 3D Semantic/Instance Segmentation using PointNet++ and SGPN.

Supervisor: Johannes Rabold

Topic: Deep Learning for Semantic Segmentation

Supervisor: Johannes Rabold


In the project we will research applications of various deep learning architectures on the segmentation of instances in images. Identification of instances of a certain type (e.g. leaf, blossom) is prerequisite for phenotyping of plants. This research problem is in the context of a joint project with Fraunhofer.


  • Background in machine learning, e.g. from KogSys-ML-M

Admission policy:

  • Maximum of 20 participants with 5 groups
  • If more participants, priority is given to people that:
    • Passed the exam of KogSys-ML-M
    • are enrolled in a higher semester
    • intend to write their master's thesis at our chair (reason: we strongly recommend, taking a project in order prepare for the master's thesis)

We will meet in the first class, where you can state, that you want to attend the project. In the second week, we fix the participants.

Students: ... Report: [pdf]

WS 2017/2018

15 ECTS project: Pain Recognition with Cartesian Genetic Programming

Description: Pain recognition has recently received significant attention. Over the past years a lot of work is done in this field, trying to find the one equation which defines the pain expression of all humans. Hence it is not a trivial challenge and it depends on a lot of factors. Different methods has been tried in automating pain recognition and some of them achieved good results. We chose Cartesian Genetic Algorithm(CGP) to identify pain by generating a Regular Expression(RegEx) that combines Action Units(AUs). We systematically review the mutation, fitness function and the evolutionary algorithm to achieve good results. The implementation steps are summarized and the statistical result are shown. There are two underlying motivations for us to write this paper. First, wheather it is possible to achieve a simple RegEx which correctly identifies pain expressions of all humans and second to find out how CGP fits with our problem and what results it achieves. (supervision: Michael  Siebers/Ute Schmid)

Students: Krist Fama, Daniel Schäfer, Report: [pdf]



Topic: Relational Learning [Unvis: KogSys-Proj-B, vc-course]

Description: In the project, we want to compare classical feature-based learning (SVM) with relational learning (ILP). We will focus on domains which are more naturally modelled relationally (social networks, chemical structures, Michalski trains) and domains which can be modelled feature-based or relational (leaves of trees, lamps). For feature-based learning we will introduce methods for extracting features from relational representations. Approaches will be evaluated with respect to efficiency of learning, classification performance and comprehensibility of the learned hypotheses to humans. (Supervisor: Christina Zeller/Ute Schmid)

  • Gruppe 1: Vergleich der Güte der Klassifizierung potenziell krebserregender Moleküle durch induktiv logische und merkmalsbasierte Lernverfahren; Students: Bettina Finzel, Katrin Grabe, Martin Hillebrand, Hannes Hornig, Rosa Ricci; Report: [pdf]
  • Gruppe 2: Klassifikation von Michalski Trains mit ILP und SVM; Students: Matthias Puchta; Report: [pdf]
  • Gruppe 3: Michalski Trains -– Learning with Relational Data; Students: Matthias Delfs, Felix Hitzler; Report: [pdf]
  • Gruppe 4: How does the Representation of Machine Learned Relational Rules Affect Human Comprehensibility?, Student: Meike Schaller; Report: [pdf]
  • Gruppe 5: Neuronale Netze zum Zählen von Objekten in Bildern; Students: Andreas Wiegand, Dennis Gross, Ludwig Schallner; Report: [pdf]

Topic: Human Robot Interaction with Pepper [Unvis: KogSys-Proj-B, vc-course]

Description: Das Ziel dieses Projekts ist es ein System zu schaffen, welches anstrebt einen
Spielenden möglichst lange dazu zu bewegen mit diesem System zu interagieren
aus reiner Interesse und Spielspaß. Dazu wird der humanoide Roboter Pepper
als Hardware genutzt auf welchem ein Programm läuft, welches mit einer Per-
son das Gesellschaftsspiel Geistesblitz spielt und dabei auf diverse Ereignisse
während dem Spielen anhand eines internen Modells reagiert. (Supervisor: Christina Zeller/Ute Schmid)

Students: Michael Groß, Adrian Schwaiger, Tim Rütermann, Sebastian Rottmann

Report: [pdf]

Topic: AI Birds [Univis: KogSys-Proj-M, vc-course]

Joint Bachelor/Master-Project with SmartEnvironments.

Description: This project is a continuation of the successful AI Birds project from summer 2016. Goal is to program an intelligent agent who can play Angry Birds. Last year, the BamBirds team became world champion! The realised approach was mainly knowledge based with a reasoning system written in Prolog at the core. This year we want to extend the BamBird agent by an online learning component. (Supervisor: Ute Schmid)


SS 2016

Joint Project with SmartEnvironments: [VC-Course]

Topic: AI Birds – Qualitative Physics Knowledge and Strategy Learning to Master a Strategy Game

Description: Das Computerspiel Angry Birds wurde in der Künstlichen Intelligenz Forschung als ein Maßstab ausgewählt für intelligent mit Alltagswissen agierende Software. Um nämlich im Spiel erfolgreich sein zu können, muss ein grobes Verständnis physikalischer Zusammenhänge auf ein Spiel übertragen werden und strategische Entscheidungen unter Unsicherheit getroffen werden. Das macht Angry Birds schwieriger als frühe Arcade-Spiele oder rein strategische Spiele ohne Unsicherheit, also totales Wissen über Spielzustand und Spielmechanik wie Go. (Supervisor: Ute Schmid)

Report: [pdf]


WS 2015/2016

Topic: Lernen von Klassifikatoren zur Unterscheidung von Ekel und Schmerz aus der Mimik

Description: In diesem Projekt soll untersucht werden, wie sich Schmerz- und Ekelgesichtsausdrücke voneinander unterscheiden. Zur Verfügung stehen Daten eines Versuches der Physiologischen Psychologie, in dem Probanden Schmerz- und Ekelmimik zeigen. Die Daten liegen bereits in symbolischer Form vor (keine Bilder/Videos). Aus diesen Daten sollen abstrakte Beschreibungen (Klassifikatoren) für Schmerz- und Ekelgesichter gelernt werden. (Supervisor: Michael Siebers)

Students: Feras Barmo, Andreas Böhler, Jan Boockmann, Mark Gromowski, Benjamin Heibel

Reports: ohne temporale Information [pdf], mit temporaler Information [pdf]


SS 2015

Joint Project with Mobi: Mining Vistior-Traces of Bamberg Zaubert

Es wird untersucht, wie durch die Analyse anonymer und anonymisierter Sensordaten Besucherbewegungen erfasst und vorhergesagt werden können. Hierzu sollen Verfahren des Machine Learnings eingesetzt werden, um typische Wege, Verweildauern und Geschwindigkeiten zu ermitteln; mit Hilfe dieser Information könnte dann beispielsweise vorhergesagt werden, dass in einer halben Stunde an einer gegebenen Stelle (z.B. Verpflegungstand) ein deutlicher Anstieg der Besucherzahlen zu erwarten wird. Diese Verfahren sollen mit einem Datenstrommanagementsystem kombiniert werden, um online relevante Ereignisse zu erkennen und an mögliche Adressaten (z.B. Betreiber des Verpflegungsstands) zu übermitteln.Die zentrale Frage, die mit Hilfe des Projekts beantwortet werden soll, lautet somit:Zu welchem Zeitpunkt hätte man welche Situationen bereits vorhersagen können? (Supervisors: Ute Schmid und Michael Siebers)

Reports: Visualisierung von Besucherströmen [pdf]; Vorhersage von Besucherzahlen [pdf]


WS 2014/2015

Topic: Complexity of Number Series Problem -- A comparison of human performance with the inductive programming systems Igor and MagicHaskeller and the special purpose system SeqSolver

Description: Based on results of a previous bachelor project and a previous bachelor thesis, given empirical results for the complexity of 20 systematically constructed number series (humans, Igor) are compared to the inductive programming system MagicHaskeller and the special purpose system SeqSolver. (Supervisor: Ute Schmid, Michael Siebers)

Participants: Barbora Hrda (CiTH), Dea Svoboda (CitH), Henrik Marquard (BA AI)

Reports:[MagicHaskeller for Number Series] [Alternating Series with IGOR]


WS 2014/2015


Topic: Impact of Communication Strategies on the Problem Solving Performance in a Multi-Agent Setting

Description: Based on a multi-agent treasure-hunt system realized in a bachelor thesis, different communication strategies between agents will be explored. (Supervisor: Michael Siebers)

Students: Daniel Bernhard, Bettina Finzel, Michael Groß, Christian Teichmann (CitH)

Report: [pdf]


WS 2014/2015

Topic: Designing and Implementing a Rule-Based Classifyer System

Description: A system for rule-based classification of, typically feature vector represented data, is to be designed and implemented. The rule base should be easily modifiable and exchangable. When a new entity is classified, the involved rules should be presented as explanation for the classification decision. (supervised by Michael Siebers and Ute Schmid)

Students: Philip Grandl, Johannes Hofmann, Tim Rütermann

Report: [pdf]


SS 2014

Topic: Enduser Programming with Igor: Learning Flashfill-Style String Transformations

Description: Enduser programming offers inductive learning systems which generalize programs from observed regularities in user behavior. For example, Flashfill offers support for complex string transformations in Excell. Flashfill is based on a domain-specific language and has an induction algorithm specifically designed for typical user tasks in Excell. Igor is a general, domain-independent inductive programming algorithm which allows to generalize recursive functional programs from input/output examples. In the project, we want to explore how to realize string transformations with Igor and compare its performance with Flashfill. (Supervisor: Ute Schmid)

Students: Peter Hohmann (BA AI), Fabian Höpfel (BA AI), Alexandra Rohm (BA AI)

Report: [pdf] Problems [maude] Sources [tar.gz]


SS 2014


Topic: Playing "I spy, with my little eye", with NAO.

Description: An autonomous gaming environment with human-machine interaction should be created. (Supervisor: Christian Reißner)

Participants: Christina Zeller, Mike Imhof

Report: [pdf]



Past Projects (Master)

Past Projects (Bachelor)