I am a machine learning engineer with a background in theoretical physics, focused on models that integrate structure, constraints, and domain knowledge into learning systems.
My work sits at the intersection of machine learning, dynamical systems, and scientific modeling, with an emphasis on robustness, interpretability, and generalisation.
- Physics-informed and structure-aware machine learning
- Time-series modeling and dynamical systems
- Generative models with inductive bias
- Bridging scientific modeling and production ML
Physics-aware anomaly detection for networked systems, where available data is sparse and reliable extrapolation is required.
The model prioritises physically meaningful anomalies over purely data-driven sensitivity, favouring stability and interpretability under limited observability.
Focus: interpretable modeling, robustness, uncertainty estimation
Implementation: linear regression, probabilistic anomaly detection
Comparative study of physics-informed LSTM and purely data-driven LSTM models for oscillatory time-series analysis and anomaly detection.
The work favours physics-informed constraints to stabilise detection performance under stricter decision thresholds, trading raw flexibility for more consistent behaviour.
Focus: modeling assumptions, robustness, diagnostics
Implementation: LSTM architectures, physics-informed loss terms, threshold-based evaluation
Exploration of VAE-based generative models in settings where unconstrained samples violate known physical structure.
The approach constrains generation to preserve physical validity, accepting reduced diversity in exchange for consistency and downstream usability.
Focus: representation learning, constraints, generalisation
Implementation: variational autoencoders, structured latent spaces, constraint-based regularisation
- Repository: (In progress)
Remaining useful life estimation on the NASA turbofan dataset under heterogeneous operating conditions, with explicit uncertainty quantification.
The model emphasises calibrated uncertainty over sharp point estimates, accepting wider intervals to reduce overconfident end-of-life failures.
Focus: uncertainty quantification, robustness, representation learning
Implementation: random forests, uncertainty calibration, hyperparameter optimisation
- PhD and habilitation in theoretical physics
- Experience leading research-oriented ML projects
- Strong focus on principled modeling and clean software design
Languages
- Python (primary)
- Bash / shell scripting
ML & scientific computing
- PyTorch, NumPy, SciPy
- Custom Physics-informed ML frameworks
Data & systems
- Linux-based workflows
- Experiment tracking and reproducibility
- Version control with Git