University of Maryland, College Park
Graduate Research Assistant, PARETO Group
College Park, MD, USA
- Developed & trained a Reinforcement Learning agent on the XRO climate model to induce sustained ENSO events by controlling 9 coupled ocean–atmosphere modes with physics-based dynamics; discovered critical preconditions for multi-year El Niño/La Niña events through learned policy analysis.
- Extended DeepMind's GenCast diffusion model with gradient-based guidance for controllable tropical cyclone trajectory forecasting over 4-day horizons; designed a parallel ReadOut GNN decoder for binary storm classification from diffusion latents with dynamic class weighting for <1% positive class ratio.
- Implemented inner latent optimization within DPM-Solver++ sampling using gradient clipping, selective spatial masking, and local affine warping for stable multi-step guided generation; built an ERA5-to-storm-mask data pipeline processing 0.25° global reanalysis data with variable-radius cyclone annotations.