Shashank Kumar Roy, PhD
Incoming Postdoctoral Researcher — MIDS, Katholische Universität Eichstätt-Ingolstadt, Germany
Artificial Intelligence for Data Assimilation | Uncertainty Quantification
PhD Physics @ ICTS-TIFR · Former Postdoctoral Researcher @ IMT Atlantique, France
📄 Download PDF versionCareer at a Glance
Postdoctoral Researcher
MIDS — Mathematical Institute for Machine Learning and Data Science, Katholische Universität Eichstätt-Ingolstadt · Ingolstadt, Germany
Postdoctoral Researcher (Ingénieur de recherche)
Lab-STICC, IMT Atlantique · Brest, France
Uncertainty quantification and ensemble methods for 4DVarNet, a deep-learning-based data assimilation framework, with Prof. Ronan Fablet.
PhD in Physics
Thesis: A study of dynamical instability and filter stability using ensemble Kalman filter (degree awarded 2025).
Education
Doctor of Philosophy in Physics — July 2020 – Jan 2024 (awarded 2025)
International Centre for Theoretical Sciences,
Tata Institute of Fundamental Research, Bangalore, India.
Advisors: Prof. Amit Apte & Prof. Samriddhi Sankar Ray.
Thesis: A study of dynamical instability and filter stability using ensemble Kalman filter
Masters in Physics (coursework, Integrated-PhD) — 2017 – July 2020 · CPI 70.8 International Centre for Theoretical Sciences, Tata Institute of Fundamental Research, Bangalore, India.
Bachelors in Physics (Honors) — 2014 – 2017 · 84% Ramjas College, University of Delhi, New Delhi, India.
Publications
Accepted Shashank Kumar Roy, Ronan Fablet. Spatiotemporal Coherent Displacements for Ensemble-Based Neural Data Assimilation. Environmental Data Science (EDS), special issue. hal-05585517
Preprint Shashank Kumar Roy, Amit Apte. A note on sensitivity of Lyapunov vectors to trajectory perturbations. EGUsphere. doi:10.5194/egusphere-2023-2168
Under review Pinak Mandal, Shashank Kumar Roy, Amit Apte. Probing robustness of nonlinear filter stability numerically using Sinkhorn divergence. Submitted to Physica D. arXiv:2208.10810
Published Pinak Mandal, Shashank Kumar Roy, Amit Apte. Stability of nonlinear filters — numerical explorations of particle and ensemble Kalman filters. 2021 Seventh Indian Control Conference (ICC), IEEE, pp. 307–312. doi:10.1109/ICC54714.2021.9703185
Talks & Conferences
Bridging Variational Data Assimilation and Flow Matching for Posterior Sampling
EnKF Workshop 2026 · Rosendal, Norway
Uncertainty Quantification in 4DVarNet via Spatiotemporal Coherent Displacements
Workshop on ML for Bi-Level Optimization, Data Assimilation and Uncertainty Quantification · Brest, France
Spatiotemporal Coherent Displacements for Ensemble-Based Neural Data Assimilation
Climate Informatics 2026 · Lausanne, Switzerland
Performance Gains and Advantages of 4DVarNet in End-to-End Learning for Data Assimilation
European Geosciences Union (EGU) General Assembly 2025 · Vienna, Austria
Automatic Differentiation for 4DVar: Some Results in PyTorch with a QG Model
Space Applications Centre, ISRO · Ahmedabad, India
4DVarNet: A Neural Network Model for Data Assimilation
Department of Data Science, IISER Pune · Pune, India
Workshop: AI for Data Assimilation — 4DVar and 4DVarNet for Ocean Datasets
National Atmospheric Research Laboratory, Department of Space · Gadanki, India
Research Projects & Hackathons
Modeled the distribution of sea surface temperature at six locations and interpolated to a new location using a GAN to perform regression.
Predicted annual mean global distributions of temperature and precipitation from emissions and concentrations of key anthropogenic climate forcings: SO₂, BC, CH₄ and CO.
Sequential estimation of the conditional distribution of the state of a 40-dimensional nonlinear dynamical system with partial observations via Kalman filtering.
Analytical calculation of Stokes flow inside a sphere generated by a given surface flow, using vector spherical harmonics; 3D flow visualizations built with the Python package mayavi.
Project titled An Interdisciplinary Study of Light Pollution in the Indian Context (Extension): collection, analysis and quantification of light pollution from the Google light-pollution map.
Teaching Experience
Supervised and assisted fellow students; conducted sessions on quantitative analysis of historical climate data using Python & Xarray, and led discussions of the underlying concepts.
Curated Jupyter notebooks for demonstrations and conducted tutorial sessions.
Solved college- and university-level physics problems on an online platform.
Skills
Scientific
Technical
Licenses & Certifications
- Deep Learning — Neuromatch Academy · view
- Data Assimilation (online course) — Imperial College London · view
- Applications of AI for Anomaly Detection — NVIDIA DLI · view
- Fundamentals of Deep Learning — NVIDIA DLI · view
- Accelerating Data Engineering Pipelines — NVIDIA DLI · view
- IBM Certified Associate Developer — Quantum Computation · view
- IBM Quantum Challenge 2021 — Advanced · view
Awards & Achievements
- Secured the 10th best score in the IBM Quantum Challenge 2021.
- Department of Atomic Energy Fellowship for pursuing a PhD in Physics.
- Joint Entrance Screening Test 2017: AIR 95 (98.8 percentile); IIT-JAM 2017: AIR 259.
- School Topper 2014 with 95.6% in 12th (CISCE Board) & 3rd topper at the district level.
A CV is the evidence that ..
Life must be understood backwards but lived forwards.
— Søren Kierkegaard (a 19th-century Danish philosopher)