About

I am a PhD student at FIIT STU in Bratislava, under supervision of a.o. Prof. Wanda Benešová.

My research efforts are focused mostly around tackling of music information retrieval tasks with deep learning methods. Specifically, approaching the problem of polyphonic music transcription as a multi-class classification problem, I train neural network models to recognize notes in musical audio signal.

My broader interests include:

  • deep learning
    • neural network architectures
    • optimization and learning algorithms
    • unsupervised learning and generative modelling
    • approximate inference and graphical models
    • reinforcement learning
  • signal processing methods for music informatics

News

  • I attended Transylvanian Machine Learning Summer School TMLSS2018 in Cluj-Napoca and won Best Poster award for presenting my work during the poster sessions. My notes might come as a separate post soon. In the meantime, you can read more about the event in this outstanding report written directly by event organizers.
  • Inference on BSDTI just released my own Implementation of Distilling Binary Soft Decision Tree according to the paper by Frosst & Hinton. Check it out!
  • I attended Deep Learning Summer School DeepLearn2017 in Bilbao, Spain. Here are some of my notes from this exciting event: https://goo.gl/btM4Fv

Projects


Automatic Music Transcription using WaveNet

Authors: Lukáš Marták, Márius Šajgalík, Wanda Benešová

We exploit modelling capacity of deep end-to-end neural network architecture called WaveNet, to perform a multi-class multi-label classification of frames of polyphonic music into sets of notes present in the polyphony. The approach is evaluated on the MusicNet benchmark dataset for deep learning approaches to music transcription.


Modelling Music Structure using Artificial Neural Networks

Authors: Lukáš Marták, Márius Šajgalík

My Master Thesis takes multiple approaches to the task of polyphonic note transcription. The project compares various neural network architectures used for modelling of polyphonic textures in musical audio data. This project was also a basis for the subsequent work focused on raw audio modelling with WaveNet which was also later published as a separate project.


Publications

Papers
  • Martak, L. S., Sajgalik, M., & Benesova, W. (2018). Polyphonic Note Transcription of Time-Domain Audio Signal with Deep WaveNet Architecture. 2018 25th International Conference on Systems, Signals and Image Processing (IWSSIP), 1–5. https://doi.org/10.1109/IWSSIP.2018.8439708
  • Beňuš, Š., Trnka, M., Kuric, E., Marták, L., Gravano, A., Computación, D. De, & Aires, U. D. B. (2018). Prosodic entrainment and trust in human-computer interaction. Proceedings of International Conference Speech Prosody 2018, (June), 220–224. https://doi.org/10.21437/SpeechProsody.2018-45
Theses
  • Modelling Music Structure using Artificial Neural Networks.
    Marták, Lukáš; Šajgalík, Márius. Diploma thesis. Bratislava: FIIT STU, 2016.

Teaching

  • Neural Networks (Autumn 2018)
  • Computer Vision (Spring 2018)
  • Digital Signal Processing, Graphics and Multimedia (Autumn 2017)
  • Neural Networks (Autumn 2017)