About me

I am a third year PhD student at the Physics of Complex Systems Department, Eötvös Loránd University, Budapest, Hungary. My thesis is about applications of machine learning in science; therefore, I mainly work with scientific data sources.


Projects

Below a few recent research topics are listed that I am involved in.


Automated joint scoring of rheumatoid arthritis progression

deep learning | diagnostic | medical | segmentation | regression

Rheumatoid arthritis is an autoimmune disease that often results in joint degradation and eventually painful joints. To track the detailed progression and response of drugs for a patient, a medical expert needs to score many relevant joints. It is a time-consuming process. We trained a convolutional neural network to perform the joint scoring accurately. Our method resulted in the top 3 for the RA2 DREAM Challenge.

alt text


Whistler segmentation for monitoring ‘space weather’

deep learning | physics | segmentation

When lightning strikes somewhere, its electromagnetic wave travels along the Earth’s magnetic field lines. This signal can be measured in the other hemisphere (for example in Hungary the whistler signals from the region of South-African can be measured). As the wave propagation speed is frequency-dependent in mediums, the whistler has a unique shape on the spectrogram. The exact shape and position of the signal carry information about the medium which the signal traveled through, this way an automated real-time segmentation of the whistlers can serve as a ‘space weather’ monitoring system.

alt text


Identification of tiger mosquitoes based on MosquitoAlert smartphone app photos

deep learning | mosquito | monitoring

Certain mosquito species are responsible for spreading potentially deadly viruses. Due to numerous factors (global warming, global transport&travel), aedes albopictus (tiger mosquito) were found in European countries in the recent years. The MosquitoAlert is a citizen science project, where the users can submit photos of mosquitoes via a smartphone app, and the photoes are evaluated by entomology expoerts. We published a paper about training a deep learning model that can help to partially automate this process, therefore allowing large-scale monitoring of tiger mosquito occuerence worldwide, which can help to prepare for an outbeark.

alt text


AI in pathology

deep learning | diagnostic | cancer research | medical | big data

Often histology is the final diagnostic procedure to decide if someone has cancerous tumor or not. During this project we are training CNNs to assist pathologists to make this decision. More to come soon…

alt text


Data driven antibiotic resistance prediction

machine learning | antibiotics | bioinformatics | genomics

Antibiotic resistance is a rising global problem. Having a fast and precise qualitative antibiotic resistance estimation could help a targeted antibiotic selection. This work proposes a machine learning based data-driven approach, where the antibiotic resistance is predicted from the DNA of the bacteria. For details, please read our publication.

alt text


Cosmological parameter constraints

deep learning | cosmology | weak lensing | physics

During this project we trained CNNs on simulated weak lensing maps to estimate cosmological parameters. For further details please read our related publications 1 and 2 in the topic.

alt text