Dylan D. Ray

This study investigates how convolutional neural networks in population genetics implicitly compute summary statistics and reveals their ability to efficiently approximate long-range linkage disequilibrium.

GHIST 2024 evaluates population genetic inference methods through four demographic history challenges, revealing the dominance of site frequency spectrum-based methods and the potential of flexible models for complex histories.

The paper proposes a graph convolutional network to directly learn from tree sequences for population genetic inference, showing comparable or better performance than CNNs on benchmark tasks.

The paper shows that mechanical and biochemical feedbacks combine to create complex contractile oscillations during cytokinesis.