KATOsuper: Surrogate-accelerated neural topology optimization with sensitivity-consistent Fourier neural operators

Topology optimization (TO) remains computationally intensive due to repeated finite element analysis (FEA) evaluations at each iteration. KATOsuper addresses this by coupling neural-reparameterized topology optimization with sensitivity-consistent Fourier neural operators, which resolves gradient inconsistency between predicted objectives and sensitivities that plagues existing neural surrogate approaches. The framework is objective-agnostic, making it applicable to various structural optimization problems.