Jacek GRELA

---

ML | Physics | Data

---

---

Profile

---

I am an academically-proven researcher with experience in data science and working knowledge of machine learning algorithms. I enjoy conducting data-driven research and developing actionable solutions. I firmly believe that my physics background coupled with technical expertise equips me with a distinctive advantage in problem-solving. As an independent researcher I thrive in novel environments and take a proactive attitude to problems.

---

Work experience

---

• Jagiellonian University

  March 2020 -

Researcher


1. Development of the Single Label One-Versus-All (SLOVA), a post-processing method which enhances usual neural-network-based classifiers with uncertainty-aware outputs to counteract the issue of over-confidence often exhibited by such models.
   [paper]
2. Utilizing statistical physics to investigate the dynamics of biological brains, we bolster the brain criticality hypothesis by demonstrating the brain's resilience in the face of stroke-like lesions.
   [paper] [github]
3. Integrating multifractal-based feature selection techniques with gradient boosting decision trees to construct an early Alzheimer's disease classifier using structural MRI scans.
   [github]

• Queen Mary University of London

  August 2018 - March 2020

Postdoctoral research assistant


1. Applying random matrix theory to the stability analysis of complex systems, investigating changes in stability hierarchy.
   [paper] [github]
2. Collaboratively organizing a Reinforcement Learning study group.
   [github 1] [github 2]

• LPTMS CNRS Universite Paris-Sud

  October 2016 - October 2018

Postdoctoral researcher


1. Exploiting the relationship between random matrices and non-interacting fermionic gas, we derive the complete distribution of its kinetic energy.
   [paper]
2. Exploring transient dynamics within non-reciprocal systems, we establish a crucial link between eigenvectors and the magnitude of transient phenomena.
   [paper]
3. Obtaining an exact solution for a multivariate Brownian bridge induced by random matrix dynamics through the application of Burgers' equation.
   [paper]

---

Selected projects

---

• [DATA] [ML] Diagnosis of Alzheimer’s disease

Development and evaluation of ML algorithms for identification of early-stage Alzheimer’s disease based on fractal properties of MRI scans.

1. Used nested cross validation
2. Addressing imbalanced dataset via resampling methods
3. Results for KNN, Support Vector Classifier and Gradient Boosted Decision Tree
4. Tech stack: scikit-learn, lightgbm

• [DATA] Exploratory Data Analysis of social media data

Formulation and development of an extensive EDA study of social media dataset.

1. Aim of the analysis is to search for patterns of organization in the data. We focus on both single and composite actors of influence from clandestine communities to bot-like users
2. Study found cooperative voter communities gathered around certain tags, singled out bots and identified a curious pattern of link promotion in terms of upvote-to-downvote ratio
3. Preparation of three-part exploratory data analysis report in a series of posts [1], [2] and [3]
4. Tech stack: plotly, pandas, statsmodels

• [DATA] [ML] Identification of pulsars

Preparation of feasibility report on using various ML models to identify pulsar stars.

1. Use of SMOTE algorithm to mitigate unbalanced classes
2. Results on Support Vector Classifiers, K-Nearest Neighors and a Neural Network using Average precision metric
3. Tech stack: scikit-learn, imblearn

• [DATA] Large-scale social media data scrape

Development and execution of a large dataset scraping pipeline.

1. Set of scripts for scraping data from a social media website wykop.pl
2. Multi-level dataset on links, voters and tags
3. Full dataset is available on Kaggle wykop-data-2022
4. Tech stack: beautifulsoup, pandas

• [ML] Meme trends analyzer

Idea and development of MMX - an ML-based analyzer of meme trends.

1. Flask-based backend API
2. Identifying popularity of memes by feature extraction using a pretrained neural network combined with vector search in mongoDB
3. Test frontend is provided
4. Tech stack: docker, flask, tensorflow_lite

• [ML] Video captioning pipeline

Development of a high-throughput, asynchronous client-server pipeline for captioning video content.

1. Video captioning server based on flask (single GPU setup) or torchserve (multiple GPU setup)
2. Fully asynchronous sample clients provided
3. Captions handled by huggingface's BLIP-2 + OPT img2txt model
4. Preparation of a blog post explaining the torchserve setup is here
5. Tech stack: flask/torchserve, huggingface, asyncio

• [ML] How generative diffusion models work?

A series of blog posts paired with accompanying notebooks that elucidate the essential mechanisms enabling generative diffusion models.

1. Materials discussing essential mechanisms behind generative diffusion models
2. Accompanying notebooks containing elementary working examples
3. Preparation of pedagogical blog posts [1] [2] and [3]
4. Tech stack: pytorch, seaborn

---

Education

---

• Jagiellonian University

  2012-2016

PhD in Physics
Thesis "Dynamics of eigenvalues and eigenvectors in non-Hermitian matrix models" on random matrix theory.

• Jagiellonian University

  2010-2012

MPhys in Theoretical Physics

• AGH University of Science and Technology

  2006-2010

BSc in Applied Physics

---

Courses & Certificates

---

1. Data Scientist Associate Certificate
2. Building Transformer-Based Natural Language Processing Applications
3. Statistics for Data Science and Business Analysis
4. Data Science in Python: Data Prep & EDA
5. Creating web applications in Python & Azure deployment
6. Python programming - basic level

---

Skills

---

• IT Skills

1. Programming: advanced
2. Machine Learning: advanced
3. Data Analysis: advanced
4. Cloud computing: basic

• IT Tools

1. Operating system: Linux
2. Programming: Python, C/C++
3. Version Control: GIT
4. Virtualization: Docker
5. Project Management: Trello
6. Cloud: Azure, AWS

• Machine Learning / Data Science

1. Deep Learning: pytorch, keras, huggingface
2. Machine Learning: scikit-learn, pandas, scipy
3. Visualization: matplotlib, seaborn
4. Databases: SQL, MongoDB, pyspark

• Soft skills

1. Analytical thinking
2. Problem solving
3. Scientific communication

• Other skills

1. Languages: english (C1, CAE exam), french (B1)
2. Scientific software: Mathematica, LaTeX
3. Other: flask, asyncio

---

Selected papers

---

• Investigating structural and functional aspects of the brain's criticality in stroke

  Sci. Rep. 13 (1), 12341 (2023)

J. Janarek, Z. Drogosz, J. Grela, J. K. Ochab, P. Oświęcimka

• SLOVA: Uncertainty estimation using single label one-vs-all classifier

  Appl. Soft Comput. 126, 109219 (2022)

B. Wójcik, J. Grela, M. Śmieja, K. Misztal, J. Tabor

• Glass-like transition described by toppling of stability hierarchy

  J. Phys. A: Math. Theor. 55 154001 (2022)

J. Grela, B. A. Khoruzhenko

• Non-intersecting Brownian bridges in the flat-to-flat geometry

  J. Stat. Phys. 183, 49 (2021)

J. Grela, S. N. Majumdar, G. Schehr

• Full Dysonian dynamics of the complex Ginibre ensemble

  J. Phys. A: Math. Theor. 51, 425203 (2018)

J. Grela, P. Warchoł

• What drives transient behaviour in complex systems?

  Phys. Rev. E 96, 022316 (2017)

J. Grela

• Kinetic energy of a trapped Fermi gas at finite temperature

  Phys. Rev. Lett. 119, 130601 (2017)

J. Grela, S. N. Majumdar, G. Schehr

• Exact spectral densities of complex noise-plus-structure random matrices

  Phys. Rev. E 94, 042130 (2016)

J. Grela, T. Guhr

• Hydrodynamical spectral evolution for random matrices

  J. Phys. A: Math. Theor. 49 085203 (2016)

P.J. Forrester, J. Grela

• Diffusion method in Random Matrix Theory

  J. Phys. A: Math. Theor. 49 015201 (2015)

J. Grela

• Dysonian dynamics of the Ginibre ensemble

  Phys. Rev. Lett. 113, 104102 (2014)

Z. Burda, J. Grela, M. A. Nowak, W. Tarnowski, P. Warchoł

[complete publcation list]