Lukáš Hübner

This paper addresses non-reproducibility in parallel phylogenetic inference by introducing a bit-reproducible tool and a general algorithm for associative parallel reduction operations.

The paper introduces a new data structure, genealogical forests, which compresses genealogical trees by storing shared subtrees only once, improving computational efficiency for population genetics analyses.

This study shows that adjusting numerical thresholds in phylogenetic inference software can significantly speed up tree calculations without affecting tree quality.

The paper evaluates fault tolerance mechanisms in RAxML-NG for phylogenetic inference, showing a 2% average slowdown from recovery mechanisms and 8% with fault-tolerant MPI.

The paper introduces HyperPhylo, a tool for efficient data distribution in phylogenetic inference with site repeats, showing improved performance over naive methods.