design of effective TL in EAs requires new representations of knowledge as well as new learning methods for knowledge transfer across tasks, since usually only objective function and limited problem-specific data are available in a-prior to an evolutionary optimization process. III. Evolutionary Transfer Optimization The study of ETO has attracted increasing interest in the field of computational intelligence. As shown in Fig. 3, existing ETO approaches can be categorized into homogenous ETO and heterogeneous ETO from the perspective of algorithmic design. The former focuses on knowledge transfer in evolutionary search across problems sharing a common search space, while the latter considers learning and transfer of knowledge across problems possessing diverse search spaces, such as having different numbers of dimensions, different decision variables, and different objectives. As shown in Fig. 3, based on the type of problems being solved, existing ETO studies can be categorized as follows: 1) ETO for Optimization in Uncertain Environment; 2) ETO for Neural Knowledge Reinforcement Learning Classification Regression Matrix Knowledge Deep Learning Optimization Classification Regression Optimization Deep Learning Reinforcement Learning Tree Knowledge Task B (Target Domain) Task A (Source Domain) Different Forms of Knowledge FIGURE 2 Illustration of transfer learning. Evolutionary Transfer Optimization (ETO) Optimization in Uncertain Environment * Optimization With Dynamic Changes in Decision Space * Optimization With Moving of Optimum *... Multitask Optimization * Implicit Multitask Optimization * Explicit Multitask Optimization *... Complex Optimization Multi/Many-Objective Optimization Machine Learning Applications * NP-Hard Vehicle Routing * Constrained Optimization * Computational Expensive Optimization *... * Multi-Objective Optimization * Many Objective Optimization *... * Image Classification * Feature Selection * Symbolic Regression *... Homogenous ETO: Knowledge Transfer Across Problems Sharing the Same Search Space Heterogeneous ETO: Knowledge Transfer Across Problems Possessing Different Search Spaces FIGURE 3 Category of existing ETO approaches. FEBRUARY 2021 | IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE 25