The paper proposes G-GLN, an extension to the recently proposed GLN family of deep neural networks. The proposed approach is competitive on several benchmarks, including contextual bandits and density estimation.