Hakim Ziani

International PhD researcher (Cotutelle, CentraleSupélec Paris & University of Turin) specializing in generative models for heavy-tailed data. Work published at international venues(ICML, EPEW...); reviewer at ICML and NeurIPS. Previously a Data Scientist at KilowattSol, building predictive maintenance models for photovoltaic infrastructure and deploying AI-driven solutions across R&D and engineering teams.

Email  /  CV  /  Projects  /  LinkedIn  /  Github

profile photo

Focus

  • Generative Models
  • Probabilistic Modeling
  • Heavy-Tailed Data
I am currently exploiting Markov Chain based Phase-Type distributions to develop novel generative models and probabilistic tools for modeling heavy-tailed data and extreme events, with applications in machine learning, time series anomaly detection and forecasting.

I am also maintaining phdist an open-source Python library for Phase-Type distributions in deep learning and probabilistic modeling.

Experience

Before my PhD, I worked as a Data Scientist at KilowattSol, where I designed and deployed predictive models for photovoltaic infrastructure. My work covered soiling ratio forecasting, brownfield and greenfield yield prediction , with a focus on delivering actionable insights for real-world energy systems. I also developed and deployed dashboards to support decision-making for engineering and operations teams.

Alongside my research, I teach Python programming to engineering students at CentraleSupélec as part of the Coding Weeks program. I mentor students through end-to-end projects, from problem formulation to implementation, debugging, and code review.

Selected Publications

  1. Ziani, A., Horváth, A., & Ballarini, P. (2025). Approximating heavy-tailed distributions with a mixture of Bernstein phase-type and hyperexponential models. arXiv:2510.26524. arXiv


  2. Ziani, A., Horváth, A., & Ballarini, P. (2026). Phase-Type Variational Autoencoders for Heavy-Tailed Data. arXiv:2603.01800. arXiv

  3. Ziani, A., Horváth, A., & Ballarini, P. (2026). Markov Chain Decoders Overcome the Heavy-Tail Limitations of Lipschitz Generative Models. arXiv:2605.18931. arXiv

  4. Ziani, A., Yakoubi, O. (2022). Machine Learning with CBC Test for Early Response Prediction to Neoadjuvant Chemotherapy. MOAD’22.

Projects


PH-VAE: Phase-Type Variational Autoencoder
International Conference on Machine Learning
ICML Generative Models Heavy-Tailed Data
Project Paper
phdist: Phase-Type Distributions Library
Open-Source Python Library
Python Deep Learning PyTorch
Project GitHub

More Links

You can find more about my work on GitHub, LinkedIn, and Google Scholar .