ML Researcher @ Computational Geoscience Inc · Machine Learning · Inverse Problems · Scientific Computing · GenAI
I build physics-informed deep learning models that bridge the gap between mathematical theory and real-world applications — from weather prediction to medical image analysis. My work focuses on designing architectures grounded in differential equations and optimization theory to make neural networks more efficient, interpretable, and scientifically meaningful.
- Physics-Informed Deep Learning — Integrating PDEs (advection, diffusion, reaction) into neural architectures for improved modeling of spatiotemporal phenomena
- Optimization for Deep Learning — Multiscale and multigrid strategies for efficient training of CNNs on high-resolution data
- Inverse Problems & Scientific Computing — Parameter estimation and data-driven solutions for forward/inverse problems
- Medical Image Computing — ML-driven precision medicine
N. Zakariaei*, S. Rout*, E. Haber, M. Eliasof NeurIPS 2024 A physically inspired CNN architecture using a novel semi-Lagrangian push operator for advection, enabling non-local information propagation in spatiotemporal prediction tasks such as weather forecasting and video prediction. [Paper] [Code]
S. Ahamed, N. Zakariaei, E. Haber, M. Eliasof TMLR 2025 A Multilevel Monte Carlo–inspired approach that uses coarse-to-fine gradient estimation and full-multiscale training to reduce CNN training costs by 4–16× on high-resolution data while preserving accuracy. [Paper]
📊 47+ citations across publications · Full list on Google Scholar
| Domain | Tools & Frameworks |
|---|---|
| Programming Languages | Python · C/C++ · MATLAB · Julia |
| ML Frameworks & Libraries | PyTorch · JAX · Scikit-learn · NumPy · SciPy · OpenCV · Pandas |
| LLM & Generative AI Tools | Hugging Face Transformers · OpenAI API · LangChain · LlamaIndex · vLLM · PEFT · LoRA · SentenceTransformers |
| DevOps & Version Control | Docker · Git |
🎓 Just defended my PhD! Super lucky to have been supervised by Prof. Eldad Haber at UBC.
🔬 Currently working on Overcoming Spectral Bias in Neural Operators.
📫 Reach me: LinkedIn