Research

• Large-scale data reduction

  Error-bounded lossy compression for scientific data, and the question of what a reconstruction has to preserve to stay useful.

• GPU-accelerated compression

  Compressor design for modern accelerators: prediction, quantization, and entropy coding that actually parallelize.

• In situ neural-hybrid reconstruction

  Learning-based methods that improve reconstruction quality alongside a classical compressor rather than replacing it.

• HPC/AI cyberinfrastructure

  Moving scientific data across HPC, cloud, and edge, and the software that has to hold up while it does.

Publications are listed on the PI's page; the code is under Software.