PhD student at INRIA MIND & INRIA THOTH working on physics-informed distributed data compression using deep learning.
Recent Work
Under review at ICLR 2027
LATEC: Learned Axial Tensor Compression for High-Dimensional Scientific Data
Hippolyte Verninas, Hadrien Hendrikx, Thomas Moreau
We introduce LATEC, an error-bounded compressor with a dimension-agnostic learned GNN predictor. Pretrained on synthetic 2D and 4D fields, a single model compresses scientific data from 3D to 5D while keeping pointwise error guarantees, achieving the best average compression ratio and the best ratio on 9 of 13 scientific fields.
