Designing physical structures and devices to achieve desirable performance is an important study in many disciplines, including physics and engineering. However, the design process is non-trivial because of the large design spaces as well as the non-unique optimal design. Usually, the design process requires an iterative trial-and-error process conducted by human experts through extensive simulations or experiments, which wastes much time and effort. The recent development of computer science has reshaped this research domain. In this talk, I will give a brief overview of these design methods, with a special focus on designing optical and photonic structures and devices. In this talk, I will briefly discuss three parts: 1) Traditional optimization methods; 2) Deep learning methods that use the neural networks as function approximators to speed up the evaluations and guide for optimization; 3) Reinforcement learning methods that can efficiently extrapolate in the design space and provide new design thoughts. One specific example will be discussed in each part.