With Jax to program the forward problem (thermal mechanical homogenization), leverage the auto-diff features of Jax to optimize microstructure with double negative thermal expansion coefficient and Poisson's ratio
Augment this small data set with developed shape blending to train a generative model for inverse design