The RMTF suite includes seven different multitask optimization problems (i.e., RMTF1 to RMTF7), in which each RMTF problem has 2 or 3 practical tasks and each task is a 2-objective real-world LMOP. Concretely, RMTF1-RMTF6 are formulated to simulate the training of DNNs on multiple different binary classification tasks, where each task endeavors to optimize the weights (i.e., the variables) of the involved DNN for concurrently minimizing its complexity and classification error. Here, the considered DNN has only two hidden layers and two parameters (h1 and h2) are used to control the neurons in these two hidden layers, respectively. Since each neuron needs to specify an activation function, five different activation functions (A1-A5) are used in RMTF1-6 problems. Besides, three cost functions, i.e., the MSE, the mean absolute error (MAE), and the root mean squared error (RMSE), are included to evaluate the classification error. Moreover, the widely used two types of regularization (i.e., L1 and L2 regularizations) are adopted to reflect the complexity of the DNN. Finally, fourteen different datasets (D1 to D14) from various fields are included in RMTF1-6 to train the DNN for bi-classification. In summary, RMTF1-RMTF6 are formulated by considering from different deployments involved in these DNN-based classification tasks, including the training dataset, the error function, the activation function, the regularization, and the structure of the DNN. Regarding the RMTF7, it is defined as a 2-task portfolio optimization problem on different datasets, where each task is a 2-objective LMOP that aims to find the portfolio of instruments having the largest expected return and the lowest risk.
Evolutionary transfer optimization (ETO) has been becoming a hot research topic in the field of evolutionary computation, which is based on the fact that knowledge learning and transfer across the related optimization exercises can improve the effi-ciency of others. However, rare studies employ ETO to solve large-scale multiobjective optimization problems (LMOPs). To fill this research gap, this paper proposes a new multitasking ETO algorithm via a powerful transfer learning model to simultaneously solve multiple LMOPs. In particular, inspired by adversarial domain adaptation in transfer learning, a discriminative reconstruction network (DRN) model (containing an encoder, a decoder, and a classifier) is created for each LMOP. At each generation, the DRN is trained by the currently obtained nondominated solutions for all LMOPs via backpropagation with gradient descent. With this well-trained DRN model, the proposed algorithm can: (1) transfer the solutions of source LMOPs directly to the target LMOP for assisting its optimization, (2) evaluate the correlation between the source and target LMOPs to control the transfer of solutions, (3) learn a dimensional-reduced Pareto-optimal subspace of the target LMOP to improve the efficiency of transfer optimization in the large-scale search space.