About
Education
Research
Projects
Skills
Coursework
Contact
Download CV (PDF)Bekhzod_CV_Machine_Learning_Engineer_EN.pdf
Copy email addressvirtsudo@proton.me
Send an emailvirtsudo@proton.me
Open GitHubgithub.com/virtsudo
Open LinkedInlinkedin.com/in/virtsudo
Cybersecurity Analyst
Software Engineer
ItalianoSwitch to Italian
Light theme
Dark theme
System theme

Open to junior roles and internships in machine learning, on-site or remote.

BekhzodAnvarov.

Machine Learning EngineerSelf-Supervised Learning (JEPA) · Deep Learning · Anomaly Detection

  • PyTorch (CUDA)
  • scikit-learn
  • XGBoost
  • Hugging Face
  • Polars
  • NumPy
  • Matplotlib
  • Jupyter

whoami

Bekhzod Anvarov

cat role.txt

Machine Learning Engineer

cat focus.txt

Self-Supervised Learning (JEPA)

Deep Learning

Anomaly Detection

cat location.txt

Turin, Italy

// about me

About

Computer Engineering graduate (MSc, Cybersecurity track) from Politecnico di Torino, applying deep learning to security data.

For my thesis I built a self-supervised Joint-Embedding Predictive Architecture (JEPA) in PyTorch and tested whether its representation keeps a classifier more accurate than raw features under concept drift. I have also trained FFNN, RNN, GNN and BERT models for security tasks and have used Python for 5+ years. Open to junior roles and internships in machine learning, on-site or remote.

// background

Education

  1. Master of Science in Computer Engineering (Cybersecurity track)

    Politecnico di Torino

    2026

  2. Bachelor’s Degree in Computer Engineering

    Politecnico di Torino

    2022

// research

Research experience

Master’s thesis · Politecnico di Torino · 2026

JEPA Representation Learning for Intrusion Detection on Real-World Drifting Network Traffic

Supervisors: Prof. Guido Marchetto, Prof. Alessio Sacco, Cristiano Serra, MSc

Does a representation learned from unlabeled network traffic keep an intrusion classifier more accurate than raw flow features on traffic recorded years later?
  1. Built a self-supervised JEPA model in PyTorch with a context encoder, an EMA target encoder, a predictor and grouped masking of correlated features. Pre-trained one encoder per year on unlabeled flows of real Internet backbone traffic from 2011, 2016 and 2021 (MAWIFlow benchmark).

  2. Evaluated it against raw flow features by training four classifiers (logistic regression, decision tree, random forest, XGBoost) on each year and testing them on every year (ROC-AUC, PR-AUC). Replaced the benchmark’s scaling, which lost most of the resolution in the features.

  3. Found that the representation makes the classes more linearly separable. In domain, logistic regression on it closes most of its ROC-AUC gap to XGBoost on raw features, but the representation does not reduce the drop across years, and the 2011 labels mean something different from the later ones. Also implemented VICReg and SIGReg regularizers, which did not improve the classifiers.

  • Python
  • PyTorch (CUDA)
  • scikit-learn
  • XGBoost
  • Polars
  • NumPy
  • Matplotlib

// featured work

Projects

Built an FFNN for network intrusion detection and FFNN, RNN and GNN models for malware classification. Fine-tuned BERT to analyze SSH logs and tag them with MITRE ATT&CK tactics.

  • Python
  • PyTorch
  • Jupyter
  • Hugging Face

// capabilities

Skills

Machine learning

  • PyTorch (CUDA)
  • scikit-learn
  • XGBoost
  • Hugging Face
  • Polars
  • NumPy
  • Matplotlib
  • Jupyter

Programming

  • Python
  • C/C++
  • SQL (MySQL, SQLite)

Tools & environments

  • Linux shell
  • Git/GitLab
  • LaTeX

Focus areas

  • Self-supervised learning (JEPA)
  • Representation learning
  • Anomaly detection
  • Concept drift

Languages

  • UzbekNative
  • RussianC1
  • EnglishB2
  • ItalianB1

// at university

Relevant coursework

  • AI and Cybersecurity

    deep learning (CNN, RNN, LSTM), NLP, anomaly detection, traffic classification.

  • Data Science and Database Technology

    classification, clustering, association rules, data warehouses, OLAP.

// get in touch

Contact

Turin, Italy