Artem Pulkin

[ɐrˈtsʲɵm]

🏠 Amsterdam NL 🇳🇱

Expertise

Machine learning, computational science, research code development

Education 🎓

2012-2017 Docteur ès Sciences EPFL in physics Lausanne CH 🇨🇭 Specialized on: numerical electronic structure, quantum simulations. Thesis: Electronic Transport in 2D Materials with Strong Spin-orbit Coupling (03/2017); supervisor: Oleg Yazyev

2010-2012 Master of Science Chalmers in applied physics Göteborg SE 🇸🇪 Thesis: Spintromechanical Aspects of Charge Transport in Nanostructures (06/2012); supervisor: Robert Shekhter

2006-2010 B.Sc. in Physics cum laude V.N. Karazin’s State University Kharkiv UA 🇺🇦

Training

Coursera: Machine Learning from Stanford University

Experience 🔬

Apr'19-Apr'22 Postdoc @ QuTech Delft university of technology NL 🇳🇱

I researched a stack of machine learning tools: deep neural networks DNN, generative models (reverse Monte-Carlo, RMC), adversarial attack approaches in the context of electronic structure/nanoscale atomic dynamics. I developed a DNN/atomic descriptor code for nanoscale dynamics miniff. I discovered novel electronic materials as a part of a multi-disciplinary team of quantum researchers.

Jul'17-Mar'19 Postdoc @ Caltech US 🇺🇸

I developed and implemented a computational many-body quantum chemistry framework to model two-dimensional crystalline materials. I investigated low-energy spectral properties of two-dimensional molybdenum disulphide with numerical modeling.

Oct'12-Apr'17 Doctoral assistant @ EPFL CH 🇨🇭

I carried out a scientific project in the quantum materials modelling domain. I discovered a new class of electronic band structure effects in two-dimensional semiconductors. I collaborated with world-leading experimental groups to prove my findings experimentally.

Jun'12-Aug'12 Research assistant @ Seoul National University, KR 🇰🇷

Aug'10-Jun'12 Research assistant @ Chalmers, SE 🇸🇪

In numbers

15 publications >500 citations 14 talks

>10 countries

>30 collaborators

Projects

More on github/pulkin

miniff https://gitlab.kwant-project.org/qt/miniff

A machine learning project in python to simulate molecular dynamics with classical force fields. Uses deep learning to train multiple neural networks at once from a hybrid dataset including both dependent variable values and their gradients. Combines the power of cython, numpy and torch to deliver maximal performance in a high-quality python code. Demonstrates my experience of full-stack machine learning research including dataset generation and performance-aware inference.

Adversarial ML playground colab

An adversarial machine learning project where I investigate the robustness of deep learning computer vision setups to various flavors of gradient-based adversarial attacks. I decided to publish (parts of) the project to make an easy hands-on introduction for those interested in the topic.

Awards 🏆

postgraduate 💰 Personal Swiss NSF grant to study abroad 80k CHF, 18 months, postdoctoral level (Early Postdoc.Mobility) grant P2ELP2_175281

💰 Personal computing time at national supercomputing facilities (SURF NL) Approximate equivalent of 26k EUR, 24 months project 45873

graduate 🥇 Olympiad in Physics for University Students (national in Ukraine) – first prize

🏅 Youth Physicists Tournament (national in Ukraine, team) – multiple prizes

🥇 Open Olympiad in Applied Physics (MIPT Moscow) – first prize

💰 Kharkiv City Mayor and Kharkiv State Governor scholarships for gifted youth

high school 🥇 Dozens of prizes in physics and informatics (olympiads, student projects; top-10 and top-1 in national competitions)

💰 Multiple scholarships

Skills 🔨

Software development in 🐍 Python (8 years): scientific stack: numpy, torch, scipy, pandas; HPC and parallel/distributed/concurrent computing (MPI, OpenMP, multiprocessing, async); performance-driven development with C and cython; styling, testing, documenting, packaging; other: FastAPI, django, OpenCV, OpenCL, cPython bytecode.

C/C++: HPC and parallel environments (MPI, OpenMP); Lapack; embedded platforms; interfacing other languages; decompiling and reverse-engineering.

Other: ☕ Java, Fortran, Julia, Javascript, Matlab.

Infrastructure: git, CI/CD (Travis, Gitlab-CI, Azure pipelines), docker, HPC, AWS (EC2, S3).

IDEs: Pycharm, vim, VSCode.

Machine learning: supervised learning (DNN, linear fits, logistic fits, SVM); unsupervised learning (PCA/SVD, K-means, anomaly detection); dataset generation, feature extraction, adversarial models.

Soft: critical analysis, problem solving, communicating (organizing discussions, presenting, paper/grant/documentation writing), full-cycle project management (idea - funding - implementation - reporting), supervision.

Languages

English (prof), Ukrainian (mother), Russian, French (basic), Dutch (basic).

Hobbies

Sports, ✈ travels, cross-stitching, soldering, 🔒 lock picking, 🕹️ board and video games, open-source projects.