Tag
This paper introduces Response Renormalization, a backward-pass framework to improve training stability in Deep Equilibrium Models by addressing near-singular Jacobian issues, demonstrated across various multiphysics applications.