Beyond synthetic benchmarks, we show applications to mechanical parameter inference from microscopy images and to inverse design problems in epithelial tissues. 4/5
We benchmark three strategies for bilevel optimization in vertex models: automatic differentiation, implicit differentiation, and equilibrium propagation.
This gives a practical comparison of their accuracy, speed, and memory trade-offs. 3/5
Very glad to see this great transatlantic collaborative work published 🎉!
Congratulations to all authors: @henrydebelly.bsky.social, @andreu Fernandez Gallén, @ericneiva.bsky.social, and @oweinerlab.bsky.social
Funding from MSCA @ec.europa.eu & @cnrsbiologie.bsky.social
We hope VertAX can help make inverse problems & differential programming more accessible in tissue mechanics.
Thanks to @apasqui.bsky.social, Maxence Ernoult, Italian colleagues, and other members of the team for this collaboration.
Comments very welcome.
Code: github.com/VirtualEmbry... 5/5
A real computational tour de force by @ericneiva.bsky.social, carried out as Marie Skłodowska-Curie postdoctoral fellow in the team, now published in the Journal of Computational Physics: urlr.me/AuzK86
Congratulations, Eric!
@cnrsbiologie.bsky.social
Grateful to @ec.europa.eu for funding support
The abstract deadline for the CSHL Cell Modeling in Space and Time is this Friday 10 April. Don’t miss out!
A differentiable JAX-based framework for vertex modeling and inverse design of epithelial tissues - VirtualEmbryo/VertAX
VertAX unifies forward modeling and inverse modeling in the same framework.
It lets users define custom energy functions and cost functions in pure Python, while leveraging JAX and bilevel optimization for differentiation, JIT compilation, and GPU acceleration. 2/5
