UKENG

Марія Добко

Research Engineer at The Machine Learning Lab mainly working in computer vision and deep learning. Has strong experience in deep learning, and applying machine learning techniques to medical domain and healthcare. Current projects:

CardioVision: a deep learning-based solution for prediction of the stenosis level on MPR images of coronary artery. The project won first place at Microsoft AI for Good Idea Challenge .

WSMIS: a method of weakly-supervised semantic segmentation that demonstrates its efficiency for detecting anomalous regions on chest X-rays.

 

GitHub · LinkedIn  · GoogleScholar

E-mail: 

Czech Technical University (Jan 2019 – Jun 2019) 
Semester of research, Center for Machine Perception, Czech Republic
Domain: Computer vision in medicine; Research topic: Lung Nodule Classification in Computed Tomography Scans Using Deep Learning

 

Ukrainian Catholic University (Sep 2015 – Jun 2019)
BSc in Computer Sciences, magna cum laude, GPA: 92.6/100, Ukraine
Specialization in Artificial Intelligence.

ML lab UCU, Faculty of Applied sciences  (Jun 2019 – Present)
Computer Vision & Machine Learning researcher

 

SoftServe, Data Science Group  (Jul 2019 – Jun 2021)
Machine Learning Engineer

 

Eleks, Data Science Office  (Jul 2018 – Jan 2019)
Junior Data Scientist
Implemented computer vision models for defects detection on image data (unsupervised anomaly detection on textural images), text based recommendation systems (resume to vacancy recommendation, personalized tourist attractions recommendation).

 

Captain Growth  (Jan 2018 – Jun 2018)
Junior Data Scientist
Developed algorithms for data analysis, anomaly detection and forecasting in time series data.

 

Ukrainian Catholic University, Faculty of Applied Sciences  (Jan 2017 – Jun 2017)
Teaching Assistant 
Communications and Academic Writing course for B.S. students in Computer Science and M.S. students in Data Science: managed communication channels between the professor and the students; participated in the evaluation of home assignments; and assisted in the preparation of course materials.

Combining CNNs With Transformer for Multimodal 3D MRI Brain Tumor Segmentation With Self-Supervised Pretraining
M. Dobko, D-I. Kolinko, O. Viniavskyi, Y. Yelisieiev
BRATS 2021 challenge, preprint – arXiv

Transformer-based Self-supervised Learning for Medical Images 
M. Kokshaikyna, M. Dobko
WiML at NeurIPS 2021 poster session, medium blog

Weakly-Supervised Segmentation for Disease Localization in Chest X-Ray Images
O. Viniavskyi, M. Dobko, O. Dobosevych
International Conference on Artificial Intelligence in Medicine, AIME 2020. paper, code

NoPeopleAllowed: The Three-Step Approach to Weakly Supervised Semantic Segmentation
M. Dobko, O. Viniavskyi, O. Dobosevych
The 2020 Learning from Imperfect Data (LID) Challenge – CVPR Workshops. arXiv, code

CNN-CASS: CNN for Classification of Coronary Artery Stenosis Score in MPR Images
M. Dobko, B. Petryshak, O. Dobosevych
CVWWW 2020, Rogaska Slatina, Slovenia. arXiv, code, proceedings

Classification of the Coronary Artery Stenosis Score in Multiplanar Reconstruction Images
M. Dobko, B. Petryshak, O. Dobosevych
Accepted to WiML workshop at NeurIPS 2019, Vancouver, Canada. Presented during poster session.

Lung Nodule Detection in Computed Tomography Scans Using Deep Learning
M. Dobko, Supervisor: Dr. Jan Kybic
Bachelor’s thesis, June 2019. paper, code

 

  • Polyp detection and segmentation from endoscopy images
  • Image matching: SuperGlue implementation and ablation study