Available Topics & Tasks

• [Intelligence] Advanced Political Opinion Slalom (Master Thesis, Bachelor Thesis, Practicum) Based on a finished project, develop an NLP application that works on the protocols of speeches in Parliament (publicly available). Extract three properties: (1) How congruent are the messages of politicians that belong to the same party? (2) How congruent are the opinions expressed by individual politicians over time? (3) How congruent are the statements of individual politicians and/or aggregated over parties in the N months before and M years after an election? Requires excellent natural-language processing and machine/deep learning skills.
• [Intelligence] Businenss as Usual: Categorization of Political Speeches by (Extra-)Ordinariness (Master Thesis, Bachelor Thesis, Practicum) Analyse videos of political speeches in order to build a model of the usual topics, figures of speech, behaviour, etc. of individual politicians and then, classify new speeches (or parts thereof) as usual/extrordinary/truthful/etc. Requires good knowledge in audio and video processing, deep learning methods, machine learning and statistical data analysis. Due to its size, the topic is available for groups of students as well.
• [Intelligence] HOT TOPIC: Chatbot for the Safer Internet Program (Master Thesis, Bachelor Thesis, Practicum) Develop an intelligent chatbot that canbe used as first-level response for children who have troubles with social media apps. The chatbot should be able to answer the most common questions that children between 8 and 12 years have when dealing with social media on WhatsApp, Tictoc and like applications. Non-standard cases should be referred to a human operator. Requires good skills in natural language processing an machine learning. If performed as a diploma thesis, must include user-based refinement of requirements + a user-based evaluation.


• [Multimedia] Dancey: Dance Karaoke with OpenPose (Master Thesis, Bachelor Thesis, or Practicum) Build a system for dance simulation based on Youtube. Essentially, take the idea of the "Just Dance" game series and wrap it onto aribtrary Youtube dance videos. Requires streaming such videos, synchronous frame extraction, keypoint extraction with OpenPose from both the dance video and the webcam, mapping of keypoints between the dancers, rendering of a merged view and visualization (simply, by OpenCV methods). On top of that implement a game that computes a score based on the level of alignment between the keypouints (e.g. arm and leg position/movement). Requires good computer vision skills & knowledge of the Python interfaces of OpenCV and OpenPose (mostly trivial). Visualization skills and game design experience are also desired, but can also be picked up during the project. If done as a dilpoma thesis, has to include a user-based evalution + one iteration of refinement based on user feedback.
• [Multimedia] Deep FFMPEG Video Classification (Master Thesis, Bachelor Thesis, or Practicum) Implement an FFMPEG extension that is able to classify videos (films) into different categories by deep learning depending on the recording system (home HD recording, professional), type of content (film, documentary, etc.) and genre (drama, action, etc.). Due to the usage of the FFMPEG library requires good programming skills and, particularly, C++ programming experience.
• [Software] Digital Credentials for Teaching (Master Thesis, Bachelor Thesis, or Practicum) Develop a system for the management of digital (micro-)credentials in the TISS teaching environment. The system should be based on a recent whitepaper and include a blockchain-based authentication mechanism as well as an interface to the present TISS teaching environment. If performed as a diploma thesis, must include an expert-based evaluation. Requires good knowledge in blockchain technology, authentication methods and Java enterprise programming.


• [Intelligence] Heimlich & Co Agents for our Strategy Game Engine (Master Thesis, Bachelor Thesis, or Practicum) Develop three agents (MCTS-based, depth search, baseline) for the board game Heimlich & Co (multiple players, turn-based, partial information, chance). The implementation has the be performed in the Java strategy game environment used in the course Strategy Game Programming. Requires good Java programming skills as well as knowledge and experience in strategic planning and reinforcement learning. If performed as a diploma thesis, the solution must include research-based state of the art solutions for the problems of dealing with partial information and elements of chance.
• [Intelligence] Intelligent Services for the TU Information Systems (Master Thesis, Bachelor Thesis, Practicum) Build components for the smart enhancement of the TU Information Systems (TISS). Topics include matching of data cagetories (e.g. project contents and research profiles, course contents and student profiles, etc.), recommender systems (e.g. "Students who did course A also subscribed course B"), smart search (e.g. by synonyms, proposed keywords, etc.) and others. Participating students have to have a good knowledge of machine learning methods in the area of text and data analysis, in particular, NLP, deep learning, recurrent networks (e.g. LSTMs) and, of course, word embedding methods such aus BERT, GPT-2, etc. Programming is done with Python + Jupyter only; the embedding of the smart components in the service infrastructure is not part of the topic. A powerful GPU infrastructure for training and evaluation is provided. Please note that the present entry comprises a number of topics & can also be subscribed by groups of students. If performed as a diploma thesis, must include a small user-based evaluation of the outcome.
• [Intelligence] Looking through the Fog of War (Master Thesis, Bachelor Thesis, or Practicum) Investigate how much planning makes sense in strategy games (reinforcement learning) under partial informations ("fog of war"). The idea is that for deeper exploration assumptions have to be made, which might be wrong & thus lead the exploration process into the wrong direction. Goal should be to develop a rule based on game complexity (what is that?) that limits the exploration process. For gaming experts: Please notice that this is not about the balance of exploration vs. exploitation but about the balance of vertical vs. horizontal exploration. The topic is ideally suited for students who passed VU Strategy Game Development and requires solid programming skills.


• [Multimedia] Mechanical Radio (Master Thesis, Bachelor Thesis, Practicum) Build a system that generates endless audio streams consisting of information on a specific topic (given in the form of keywords), general information (e.g. weather, news) and music that fits to the chosen topic. The entire system should run on an embedded computer and perform all intelligence-based tasks (selection of information and text) on the edge. The final stream should have production quality and be both informative and entertaining. If done as a master thesis, has to include an extensive user-based evaluation.
• [Multimedia] Presentation Hardware AV Quality Estimator (Master Thesis, Bachelor Thesis, or Practicum) For an existing solution for presentation, collaboration and knowledge sharing (so-called BYOD infrastructure) build a system which can assess the quality of replayed media with regard to different quality parameters (e.g., latency, image quality, synchronicity of Audio- and Video). In modern multimedia presentation system, media data streams need to be delivered in top quality to the output sources: HDMI, USB, Network, etc. In this context optimal quality means: low latency, best resolution (4k) and color trueness, no flickering, and perfect audio- and video synchronization. To assure the required quality, a considerable number of parameters need to be checked after each change of components influencing the playback of media streams (e.g. driver and system updates, streaming infrastructure updates, etc.). As this is a very time-consuming task if done manually, it should be automated by a software solution (e.g. implemented as a filter in ffmpeg or gstreamer). All hardware and software infrastructure will be provided by the institute. A successful implementation of the practicum will be rewarded with an award of 1000€. Requires good knowledge in embedded programming, networking and multimedia communications.

• [Multimedia] Presentation Hardware Network Limiter (Master Thesis, Bachelor Thesis, or Practicum) For an existing solution for presentation, collaboration and knowledge sharing (so-called BYOD infrastructure) build a system which can be placed between it and the network, which allows to control network quality parameters like packet loss, latency, bandwidth and so on. Background: BYOD presentation functions strongly depend on the quality of the network. Varying quality (e.g. latency, packet loss, bandwidth) will result in varying quality of provided functions and maybe even to failures. Reproducing such failures is often impossible as insufficient network quality is difficult to reproduce. All hardware and software infrastructure will be provided by the institute. A successful implementation will be rewarded with an award of 1000€. Requires good knowledge in embedded systems and an interest in multimedia communications/quality of service.
• [Multimedia] Presentation Hardware Network Scanner (Master Thesis, Bachelor Thesis, or Practicum) For an existing solution for presentation, collaboration and knowledge sharing (so-called BYOD infrastructure) build a system which can scan a customer’s network environment and checks e.g., if specific ports are opened or the bandwidth between the system and a BYOD presentation device is sufficiently good for e.g., mirroring a 4k video via Miracast. Background: Certain functions of BYOD infrastructures (e.g., device discovery) strongly depend on the network infrastructure. If a network is not configured 100% correct, errors may occur which are very hard to reproduce (e.g., green flickering in BYOD-sessions as a result packet loss due to a not opened port on a firewall). All hardware and software infrastructure will be provided by the institute. A successful implementation of the practicum will be rewarded with an award of 1000€. Requires good knowledge in networking and multimedia communications.
• [Intelligence] Royal Bot the Second (Bachelor Thesis, Master Thesis or Practicum) Build a mobile robot that recognises its environment and interacts with humans, requiring the understanding and interpretation of their motives, thus developing a basic form of consciousness. The robot exists in a first version and is able to recognise the following settings (+interaction): (1) cheering crowds (stand still, look tangentially away from the crowd and slowly wave back), (2) one person kneeling in front of the bot (draw a miniature sword and knight that person). It is, hence, a jovially royal bot. Now, please extend the framework by additional features and more consciousness. For example, when recognizing groups of people in uniforms, the robot should stand erect, move slowly forward, hoist a small flag and play some sombre yet enchanting piece of music. Further requirements will be defined by the supervisor in the pre-course meeting. Please recognise that apart from the humorous requirements this is a serious topic with hard to achieve goals.


• [Multimedia] Rugby Video Analyzer (Master Thesis, Bachelor Thesis, or Practicum) Develop a windows-based video player that is able to analyze recordings of rugby union matches. Functions should include the recognition of scrums, a jump over option (also for timeouts), an outcome analyser for scrums, lineouts and other set pieces, the eliminiation of in-screen ads and banner ads, etc. Requires solid computer vision skills and fluency in a major programming language. If done as a diploma thesis, it must include a data-based evaluation.
• [Virtuality] Smartphone-based Interactive Virtual Climbing (Master Thesis) For our existing physical-virtual climbing environment Vreeclimber, implement an augmentation of the virtual environment that includes (1) the usage of smells over an existing interface (Vragrancer, as used in the Virtual Jumpcube application), (2) controlling the environment of the current climber externally (by, for example, adding birds/spiders/etc., thunderstorms, other environmental effects, etc. - whatever can be thought of) through a mobile app, and (3) the evaluation of the effect of interaction on immersion in a user study. Requires good knowledge in Unity programming, VR design and 3D modelling. Please note that a fully functional model of the climbing wall including motor management and tracking does exist and can be used out of the box.
• [Intelligence] A Toolbox for Artificial Consciousness: Second Iteration (Bachelor Thesis, Master Thesis, or Practicum) Extend the existing Python framework for artificial consciousnes by a generalizing mechanism for the understanding of happiness and fear. Based on visual stimuli the agent should be able to generalize the notion of fear it has of dogs as well as the notion of happiness it perceives from seeing cats using reinforcement learning. Alongside, improve the simulation capabilities by a visual rendering, dynamic components and improving re-usability of components. Students of Media and Brain 1/2 are preferred.


• [Intelligence] Transfer Learning for Image Hunters (Bachelor Thesis or Practicum) Develop a deep classifier that is able to identify strongly camouflaged animals in images. The classifier should run on an embedded system (practically, a Raspberry Pi 4; edge computing) and employ an existing object classifier that is re-trained to the specific task (transfer learning). Once recognized, the classifier should be able to determine if, in case a bullet would go through the center of the image, the animal would be killed, wounded or remain unharmed. Based on the classification result, a hash value should be computed that can be read from the edge computer via an Android mobile app and forwarded to a given REST interface. Requires good deep learning skills, experience with edge computing and a minimal knowledge of mobile app programming.
• [Virtuality] The Virtual Artisan 2 (Bachelor Thesis, Master Thesis or Practicum) Based on Unity or Unreal develop virtual reality applications for the training of artisans in practical skills such as welding, turning (metal and wood), etc. Applications should be as realistic as possible which requires creativity in the haptic area, in particular. Further requirements are a minimum understanding of game engines as well as readiness to improve existing physics engine procedures (e.g. for a correct implementation of the welding process). If performed as a diploma thesis, the task has to include a small sclae evaluation with test users and quantitative measurement.
• [Virtuality] Virtual Reality Evaluation of Robot Sonification (Master Thesis, Bachelor Thesis, or Practicum) Perform a VR-based evaluation of different forms of sonification on an existing robotics environment (robot arm, stereo cam, joystick, headphones). Requires the implementation of VR app that uses a stereo data input stream for the visualization of a robot and a ball that needs to be grabbed by the user. The app should, furthermore, provide different forms of sonification that should help the users in grabbing the ball with the robot via HMD and joystick. The hypotheses are that using sonification makes (1) the learning process faster, (2) the grabbing process faster, and (3) the grabbing process more reliable. Evaluate theses hypotheses with a small group of users.


• [Multimedia] Voice Identification with an Artificial Corpus (Master Thesis, Bachelor Thesis, or Practicum) Develop an approach for the sampling of a corpus of artificial words (based on literature in form of texts/ebooks) that represents the typical distributions and combinations of phonemes in existing languages statistically correct. Should focus on the Serbian, Hungarian and Turkish languages.
• [Intelligence] Words, Sentences, Meaning: What Matters in Politics? (Master Thesis, Bachelor Thesis, or Practicum) Based on an existing system for political opinion mining, investigate the question what matters more in political speeches: meaningful expression of complex concepts or just iterating of buzzwords. Use natural language processing techniques to model the different concepts and data provided by the Austrian parliament to evaluate the research questions. Requires good knowledge in natural language processing and machine learning.

Finished Tasks, since 2019

• [Intelligence] AI Recommendation of Thesis Reviewers (Master Thesis) Develop a deep learning model that is able to recommend reviewers for student master theses at TU Wien from (1) the abstract of the thesis and (2) the research profiles of the researchers. Requires good knowledge of a deep learning framework, natural language processing, LSTMs and other recurrent networks.
• [Multimedia] Audio Analysis Software for Raspberry Pi (Bachelor Thesis) Develop content-based audio analysis components that are able to run on a Raspberry Pi processor. Potential topics include sound effects and the recognition of voice/music transitions. Eventually, the software should run on our Raspberry Pi radio system Tuvoli. Requires basic knowledge in audio analysis and signal processing.
• [Multimedia] Babelfish App (Bachelor Thesis, Practicum) Develop a mobile app that translates spoken content from one language to another using already existing services for speech recognition, translation and speech synthesis. The service should be limited to German and English.


• [Multimedia] Base Dosimeter (Bachelor Thesis) Develop a dosimeter (long-time noise recording) exclusively for the base frequencies. Should include proper recording hardware, if necessary, band pass filtering, RMS computation and optionally, recording of the source signal on an embedded device (e.g. a Raspberry Pi) + a data analysis and reporting component on the PC. Requires maker skills and good programming knowledge in one high-level programming language.
• [Virtuality] Content Management for Virtual Climbing 1 (Bachelor Thesis) For our existing physical-virtual climbing environment Vreeclimber, migrate two of four existing contents to the latest Unity version and integrate them with the interaction template of the climbing wall. The contents are: climbing a space ship or a crater on the Moon. Requires 3D modelling skills and a good understanding of the requirements of VR applications as a major issue will be finding the optimal trade-off between level of detail and framerate requirements of the game engine. Please note that a fully functional model of the climbing wall including motor management and tracking does exist and can be used out of the box.
• [Multimedia] Implementation of a Mobile Interactive Climbing App (Practicum) Climbr is a mobile app that allows for intuitive visual editing of climbing routes. In this task, it shall be re-implemented using Flutter (similar to JS/React; see flutter.dev). In this task, you first learn Flutter as well as implementing the required UI patterns. Then, you re-implement the existing app on a quality level that allows for publication in the Android and Apple app stores.


• [Virtuality] Construction of a Virtual Tasting Device (Bachelor Thesis) Build a virtual taste device that imitates the approach taken in https://dl.acm.org/citation.cfm?id=3123440 - only for taste stimulation, not for virtual cocktails -, i.e. creation of gustatory sensations by electrical stimulation of the tongue. The final prototype should include a hardware prototype, a software interface for Unity & project documentation (BoM, building instructions, etc.). Requires good electronics skills. A well-equipped lab + funds for hardware are available.
• [Virtuality] Distributed Multi-User Virtual Miniature Golf Application (Bachelor Thesis) Develop an application for miniature golf playing with virtual reality. Includes a bit of hardware design (attaching a Vive handheld to a stick) and the development of a gaming environment in Unity or Unreal. Once the basic application has been developed, the framework should be extended to distributed multi-user usage - potentially by using collaborative VR technology developed at the IMS. If performed as a diploma thesis, the project must include a user-based evaluation of the prototype.
• [Intelligence] Embedded Parasite Recognition (Bachelor Thesis) Implement an embedded-system (based on a Raspberry Pi) for the recognition and classification of parasites on small fast-moving insects. Requires good knowledge in computer vision, deep learning and Python programming.


• [Virtuality] Finger Tracking based on Deep Learning (Bachelor Thesis) Based on an existing prototype for hand image generation (based on 3D models of hands) develop a deep learning model that is able to recognize hand poses (fist, open hand, etc.) as fast as possible. Input images can be generated for all degrees of freedom (hand size, skin color, anomalities, etc.) and all perspectives (front, side, etc.) - thus millions of training images can be generated quickly. Training should be startet with CNN architectures, but might also include recurrent networks. For application, the model is fed with images from a consumer webcam. Goals are the maximization of the frame rate (for VR >=90fps would be desirable; 10-20fps are realistic for state-of-the-art Nvidia graphics cards) and the maximization of the accuracy (e.g. measuresd as mean average precision for five categories on the scale between open hand and fully closed fist). Requires good engineering skills and an interest in 3D modelling, VR and machine learning.
• [Multimedia] Game Design for Augmented Reality Bouldering (Practicum) For our existing physical-virtual climbing environment Vreeclimber, implement two more games for bouldering with augmented reality. Pracitcally, this requires the design of targets (sprites), a recognition algorithm in OpenCV and of game designlike in the old days of level/arcade games. Required skills are Python programming, graphics design and computer vision experience. Please note that a fully functional model of the climbing wall including motor management and tracking does exist and can be used out of the box.
• [Multimedia] Implementation a Mobile App for Sign Language Learning (Practicum) Lectary is a platform-independent mobile video-based app for the learning of sign language. The app is, for example, part of the curriculum at the University of Klagenfurt. In this project, Lectary shall be re-implemented in Flutter (similar to JS/React; see flutter.dev). In this task, you first learn Flutter as well as implementing the required UI patterns. Then, you re-implement the existing app on a quality level that allows for publication in the Android and Apple app stores. Please note that this is a real-world project with social impact. Hence, we expect a professional, user-centred and dependable approach by the student.


• [Software] A Like Button for the Legislative (Bachelor Thesis) Develop a server-based system that satisfies the following requirements: (1) Periodic (e.g. annually) grabbing of the contents of the Austrian Database of Laws (RIS), (2) flexible UI for usage on desktop & mobile systems, (3) a server job that computes the changes in content per paragraph from one year to another, (4) content reading functions for browsing, search & random access for the changes/paragraph, (5) presentation of the individual changes with a feedback widget for (-2="hate it"..+2="love it" & "don't know/care"), 86) at times, ask the user for her/his political preference:"left - center - right", (7) reporting functions for aggregated results that filter out too many too quick answers (spam), out of Austria geoblocking, indiscriminate haters/lovers, etc. Open issues are: (1) the implementation of a reward function, a seeding algorithm, offering of "thematic paths" in the browsing function. Requires solid programming skills, can be done by a group of two students, ideally suited for students who passed VU Ruby Business Programming.
• [Intelligence] Political Opinion Slalom (Master Thesis) Develop an NLP application that works on the protocols of speeches in Parliament (publicly available). Extract three properties: (1) How congruent are the messages of politicians that belong to the same party? (2) How congruent are the opinions expressed by individual politicians over time? (3) How congruent are the statements of individual politicians and/or aggregated over parties in the N months before and M years after an election? Requires excellent natural-language processing and machine/deep learning skills.
• [Virtuality] Robot Arm Sonification (Bachelor Thesis) Explore the use of sonification in the control of robot arms: Build a setup that consists of a robot arm + an underwater work task (e.g. manipulating a cable) and/or provide a software environment for avatar-based control of such a robot arm. Provide sonification of movement and touch sensors. If done as a diploma thesis should involve an evaluation of easier learning and/or more precise handling of the robot arm in the defined task. Requires good programming skills and interest in robot technology.

Please see the static snapshot of the IMS website for tasks started and/or finished between 2000 and 2018 as well as tasks started before 2019 and finished since.