ml researcher @ algoverse ai · 2025

this is my algoverse ai research work, benchmarking physics-informed neural networks (pinns) across different data types and optimizers on reaction and wave pdes. switching from fp32 to fp64 precision cut both error and memory usage in ways that were consistent across multiple pde setups, and i quantified those tradeoffs across the full grid of dtype/optimizer combinations. the work is on track for a neurips workshop submission.