The paper presents a novel approach to multiple kernel learning (MKL) by formulating it as a second-order cone programming (SOCP) problem. The authors address the challenge of non-smoothness in the MKL problem, which makes it difficult to apply sequential minimal optimization (SMO) techniques. They propose a dual formulation of the problem using Moreau-Yosida regularization, which allows for the application of SMO. The paper includes theoretical results on optimality conditions and provides an algorithm for solving the regularized MKL problem efficiently. Experimental results show that the proposed algorithm outperforms general-purpose interior point methods in terms of computational efficiency.The paper presents a novel approach to multiple kernel learning (MKL) by formulating it as a second-order cone programming (SOCP) problem. The authors address the challenge of non-smoothness in the MKL problem, which makes it difficult to apply sequential minimal optimization (SMO) techniques. They propose a dual formulation of the problem using Moreau-Yosida regularization, which allows for the application of SMO. The paper includes theoretical results on optimality conditions and provides an algorithm for solving the regularized MKL problem efficiently. Experimental results show that the proposed algorithm outperforms general-purpose interior point methods in terms of computational efficiency.