R1 Regularization

2000년 도입 · 논문 518편에서 사용

R_INLINE_MATH_1 Regularization is a regularization technique and gradient penalty for training generative adversarial networks. It penalizes the discriminator from deviating from the Nash Equilibrium via penalizing the gradient on real data alone: when the generator distribution produces the true data distribution and the discriminator is equal to 0 on the data manifold, the gradient penalty ensures that the discriminator cannot create a non-zero gradient orthogonal to the data manifold without suffering a loss in the GAN game. This leads to the following regularization term: $$ R\_{1}\left(\psi\right) = \frac{\gamma}{2}E\_{p\_{D}\left(x\right)}\left[||\nabla{D\_{\psi}\left(x\right)}||^{2}\right] $$

출처: Which Training Methods for GANs do actually Converge?

소개 논문: Which Training Methods for GANs do actually Converge?