György Kovács

For Students

Courses

• Introduction to Artificial Intelligence (D0032E)
• Machine Learning and Pattern Recognition (D0033E)

Master Supervision

My main research interest is in Speech and Language Technology, thus this is also my primary interest in supervising student theses. Topics of interest include, but is not limited to:

• Automatic Speech Recognition: a key task in speech technology and speech processing is transcribing audio into text, that is taking audio input, and providing as output a sequence of words/letters. Here, one key challenge is doing so successfully in various conditions (e.g. with background noise, with speakers that have different accents, speech impediments, or belong to various age groups). In the task (depending on the available data) one can target different languages. An example of a thesis in this area is available below.
• Paralinguistic Speech Processing: another interesting part of speech processing when what we are curious about in the speech signal is beyond what is being said. This could mean many different things, such as the emotion expressed by the speaker, the state of the speaker otherwise (sleepy/intoxicated, or under physical load), or their accent.
• Audio event classification: is again an area with many possible tasks to address. One example is the classification of cough sounds (see below) into categories of "Covid" and "non-Covid" coughs. Another example is the classification of bird species based on audio recordings of their calls.
  
• Sentiment analysis in written text: here, potential thesis topics can include various classification tasks based on short texts. This could, for example mean the categorization of messages into positive/negative class, the classification of text based on the emotions expressed in it, or detecting signs of depression in text. One related topic that I find especially interesting is the detection of hateful language (particularly in social media), and how this detection relates to the sentiments expressed in the text. The exact topic to be addressed depends on the data available.
• Bot detection: another interesting task in social media analysis is categorizing users as humans or bots based on the posts/messages of the user.

Another important research interest I have is Earth Observation, or particularly, image processing as applied to Earth Observation. Here, we could work on various image classification/detection tasks based on satellite images, or drone images.

I am also open to supervising other topics. To get an idea of the range of possible topics, please see the list of theses I have supervised, and the papers on which I worked with students.

Theses Supervised

2024

• Mohamed Dinawi, Accent Classification using Machine Learning in English Language [Link External link, opens in new window.]
• Theo Hembäck (co-supervisor), UCL: Unsupervised Curriculum Learning for Medical Diagnosis, Scene Recognition and Historical Document Classification [Link External link.]
• William Schill, Sentiment Analysis in Swedish Texts: Understanding Emotional Context [Link External link, opens in new window.]
• Ludvig Hedlund, Towards On-Premise Hosted Language Models for Generating Documentation in Programming Projects [Link External link.]
• Rafael Silva, Assessing Code Quality and Performance in AI-Generated Code for Test Automation [Link External link.]

2023

• Ishika Gupta, Punctuation Restoration as Post-processing Step for Swedish Languge Automatic Speech Recognition [Link External link, opens in new window.]

2022

• Hector Nyblom, Swedish Language End-to-End Automatic Speech Recognition for Media Monitoring using Deep Learning [Link External link, opens in new window.]
• Jacob Wedin (co-supervisor), Classifying Electricity Tower Image Data with Unsupervised Curriculum Learning [Link External link, opens in new window.]
• Nils Widmark, Short-term electricity consumption forecasting using deep learning and external variables [Link External link, opens in new window.]

2021

• Anil Kumar Kothapalli, Short-Term electricity consumption prediction: Elområde 4, Sweden [Link External link, opens in new window.]
• John Jernberg, Identification of alkaline fens using convolutional neural networks and multispectral satellite imagery [Link External link, opens in new window.]
• Adam Zoltan Kenyeres, Social Media Bot Detection [Link External link, opens in new window.]
• Sumit Rakesh (co-supervisor), Exploiting Leap Motion and Microsoft Kinect Sensors for Static and Dynamic Sign Gesture Recognition [Link External link.]

Beyond the thesis

In some fortunate cases (and this is something I really like to encourage, so that students can gain further experience), the cooperation does not end with the thesis, but also leads to a scientific publication

• Abid, N., Kovács, G., Wedin, J., Paszkowsky, N.A., Shafait, F. and Liwicki, M., 2022, November. UCL: Unsupervised Curriculum Learning for Utility Pole Detection from Aerial Imagery. In 2022 International Conference on Digital Image Computing: Techniques and Applications (DICTA) (pp. 1-9). IEEE.
• Kenyeres, A.Z. and Kovács, G., 2022. Twitter bot detection using deep learning. In XVIII. Conference on Hungarian Computational Linguistic (MSZNY 2022), Szeged, january 27–28, 2022 (pp. 257-269). University of Szeged. [Nominated for Best Paper award]
• Rakesh, S., Kovács, G., Mokayed, H., Saini, R. and Pal, U., 2021, June. Static palm sign gesture recognition with leap motion and genetic algorithm. In 2021 swedish artificial intelligence society workshop (SAIS) (pp. 1-5). IEEE.

In some other cases, our work with students may not lead to a thesis (many students work with different researchers, and choose one of them later, to supervise their thesis work), but can still lead to interesting publications

• Nilsson, F., Al-Azzawi, S.S.S. and Kovács, G., 2022. Leveraging sentiment data for the detection of homophobic/transphobic content in a multi-task, multi-lingual setting using transformers. In 14th Forum for Information Retrieval Evaluation, FIRE 2022, December 9-13, 2022, Kolkata, India (Vol. 3395, pp. 196-207). CEUR-WS.
• Nilsson, F. and Kovács, G., 2022, May. FilipN@ LT-EDI-ACL2022-Detecting signs of Depression from Social Media: Examining the use of summarization methods as data augmentation for text classification. In Proceedings of the Second Workshop on Language Technology for Equality, Diversity and Inclusion (pp. 283-286).
• Södergren, I., Nodeh, M.P., Chhipa, P.C., Nikolaidou, K. and Kovács, G., 2021. Detecting COVID-19 from Audio Recording of Coughs Using Random Forests and Support Vector Machines. In Interspeech (pp. 916-920).
• Lavergne, E., Saini, R., Kovács, G. and Murphy, K., 2020, December. Thenorth@ haspeede 2: Bert-based language model fine-tuning for italian hate speech detection. In Proceedings of the Seventh Evaluation Campaign of Natural Language Processing and Speech Tools for Italian Final Workshop, EVALITA (Vol. 2765, pp. 142-147).

PhD Supervision

• Present, Sana Al-Azzawi (co-supervisor)
• 2024, Nosheen Abid (co-supervisor), Unsupervised Curriculum Learning Case Study: Earth Observation UCL4EO [Link External link, opens in new window.]