Short Biography

Pavol Harar got his master's degree in system engineering and informatics from Brno University of Technology in 2015 as a transition from economics to IT. During which he became very excited about data mining and machine learning. Fascinated by the possibilities of neural networks he applied for PhD programme at Faculty of Electrical Engineering and Communications to bootstrap his research in the field of deep neural networks and artificial intelligence. During his second year of teaching, he was awarded the best teacher at Department of Telecommunications and 8th best teacher at the whole faculty. Currently he continues his research in cooperation with logopedists and mathematicians as a part of NuHAG group at Faculty of Mathematics, University of Vienna.

Full CV

Research

Currently, I am researching the deep learning approaches mainly for audio analysis and medical diagnostics under the supervision of Monika Dörfler at Numerical Harmonic Analysis Group (NuHAG) in Vienna. I am also part of Brain Diseases Analysis Laboratory (BDALAB) under the supervision of Jiri Mekyska. My PhD supervisor is Radim Burget from Signal Processing Laboratory (SPLAB) Bellow, you can find a few of my research interests that mostly span my todo list.

  • Analysis of pathological voices with deep learning

  • Exploration of novel time-frequency representations

  • Investigation of new preprocessing steps for learning algorithms

Get in touch

Publications

2018

Towards Robust Voice Pathology Detection @ Neural Computing And Applications

In search towards a robust voice pathology detection system, we investigated three distinct classifiers within supervised learning and anomaly detection paradigms. We conducted a set of experiments using a variety of input data such as raw waveforms, spectrograms, mel-frequency cepstral coefficients (MFCC), and conventional acoustic (dysphonic) features (AF). In comparison with previously published works, this article is the first to utilize combination of four different databases comprising normophonic and pathological recordings of sustained phonation of the vowel /a/ unrestricted to a subset of vocal pathologies.
article
poster
2018

BUT-CZAS – Database of Czech anechoic speech @ Elektrorevue

The article introduces a novel database of human voice recordings named BUT-CZAS (Brno University of Technology, Czech Anechoic Speech), acquired in the anechoic chamber. The database consists of 405 high-quality mono recordings of the reading task in the Czech language acquired using bit depth of 24 bit and sampling rate of 48 kHz.
article (CZ)
web

Training

Lots of new people, new concepts, emerging technologies and no administrative work. Just hands on the problem and at the end, the great feeling of satisfaction. That's why I love workshops. Here you can find a list of those I attended and found very worth it. Please recommend me the upcoming good ones you know about.

3 days

Mathematical Challenges of Structured Function Systems @ ESI, UNIVIE (19 Mar 2018 - 21 Mar 2018)

This ESI workshop built up on recently discovered links between Gabor and wavelet analysis on the one hand and real algebraic geometry, complex analysis, system theory, special functions, and approximation theory on the other hand.
3 days

Machine Learning weekend @ Kiwi.com, Brno (27 Oct 2017 - 29 Oct 2017)

A thorough course led by top machine learning engineers from Google, DoiT, Seznam.cz and Kiwi.com for the authors of the best entry task solutions. 25 participants out of ~400 applicants.
2 days

Novathon by PBZ @ Zagreb (23 Sep 2017 - 24 Sep 2017)

A 24-hour fintech hackathon organized for distilling the best ideas and creating cutting-edge applications for responsive, mobile and branch banking.
5 days

Systematic approaches to deep learning methods for audio @ ESI, UNIVIE (11 Sep 2017 - 15 Sep 2017)

Assembly of scientists from both applied mathematics and computer science focused on gaining a systematic formal understanding of the underlying structures and principles of deep learning.
3 days

Python weekend @ Kiwi.com, Brno (23 Sep 2016 - 25 Sep 2016)

Intensive programming workshop for 16 selected programmers led by senior developers from Kiwi.com. The whole weekend was mainly focused on HTTP Requests, Multithreading / Multiprocessing, Redis and Flask.
1 day

Code retreat @ BUT by Seznam.cz, Brno (17 Jan 2016)

Test driven development workshop for the best students aimed on looking at problems from different perspective. This whole day event was led by M. Jančařík from Seznam.cz.

Teaching

A list of invited talks and other presentations I gave at various public or proprietary events.

05 Jun 2018 @ Strobl18

Poster session: Towards Robust Voice Pathology Detection

Presentation of the poster in the so-called "engineering corner" of the Harmonic Analysis and Applications conference near Salzburg.
poster
web
02 Feb 2018 @ BUT

Battle-proven machine learning workflow from venv to dockerization

Hands-on Python, Pandas, Scikit-learn, XGBoost, Keras, Docker. Everything from virtual environment, through pipelines to dockerization.
slides
repo
12 Dec 2017 @ Mergado DevTalks

How to train fox's brain to predict the future

Presentation of our successful prediction system for e-shops based on Deep Learning. Discussing the data processing, feature engineering, training and deployment into production.
13 Sep 2017 @ Systematic approaches to deep learning methods for audio

Voice Pathology Detection Using Deep Learning: a Preliminary Study

In contrary to standard machine learning approaches, in this presentation, we discussed a preliminary investigation of voice pathology detection from raw waveform using combination of convolutional and LSTM layers.
slides

Projects & Grants

opvvv

International Mobility of Researchers @ Czech Ministry of Education, Youth and Sports (Apr 2018 - Mar 2019)

This grant (CZ.02.2.69/0.0/0.0/16_027/0008371) is focused on enhancing international cooperation and developing human resources in research. The main objective is to support professional growth of researchers and to develop research organizations by strengthening their human resources.

vutbr

Multimodal processing of unstructured data using machine learning and sophisticated methods of signal and image analysis @ BUT (2017 - 2019)

This research project (FEKT-S-17-4476) supports specific research in multiple areas, such as paraclinical methods of Parkinson's disease progression and treatment efficacy monitoring with respect to hypokinetic dysarthria, advanced techniques of data analysis using deep learning and more.

bdalab

Effects of non-invasive brain stimulation on hypokinetic dysarthria, micrographia, and brain plasticity in patients with Parkinson's disease @ Czech Ministry of Health (2016 - 2019)

This research project (16-30805A) aims for the identification of potential therapeutic effects of rTMS as a tool that could contribute to the speech therapy of hypokinetic dysarthria in Parkinson's disease patients. The results will also enhance our understanding of brain mechanisms underlying specific long-term effects of rTMS.

profitak

Bootstrapping R&D department @ PROFITAK (Jun 2017 - Feb 2018)

This commercial project's main aim was to explore the company's resources, create commercial machine learning MVPs interconnected with the existing infrastructure and evaluate the integration process with respect to sustainability and profitability as steps of establishing a new Research & Development department. Michal Janik, the CEO of Profitak solutions, s.r.o. can provide testimonials.