E - TAIDE Knowledge Graph

We would like to invite Yusuke Nojima to introduce the evolutionary computation Thank you very much for your introduction. I'm Yusuke Nojima from Osaka Metropolitan University, Japan. Today, I'm here. I will talk about the evolutionary computation and take home message is the evolutionary computation is a powerful and flexible. And, you also can find the novel solutions and PSO is a kind of the evolutionary computation technique. It is simple and powerful. Let's start with a Natural evolution See this cycle of images. It is a Natural evolution but actually there are the several variants in a population. This circle represents a population in one generation and each generation in a each population, there are the strong individual and some weak individuals. And, then those who are better adapted to their environment have a greater chance of surviving, breeding, and passing their genetic information to the next population. So, individual in the next population will have more well adapted to their environment, then they continue to this evolutionary process, the science individually existing in this process. I show two simply examples. The one is a polar bear as you can know the polar bear is living in a very cold place but he has a large body and a fluffy hair. This is because he keeps the body temperature inside. That shape was obtained by the natural evolution. All right. The other example is a Macropinna. This fish is living in the very deep sea and the most interesting point is that these are eyes and the eyes are looking upward all the time. Do you know why? Can you know why they have a such an interesting eyes? Anyway, this fish obtained is the interesting structure through the evolution. So, the basic about this later, the evolutionary computation is derived from the natural evolution. And, if we have a problem to be optimized the problem is like a envrionment in nature. We have a candidate solution and candidate solution like an individual in nature. And, each candidate solution is defined by a vector of design values and the evolutionary computation process will change these design variables to adapt to find the optimal solution for the problem at hand. So, the process is almost same as a natural evolutionary. At first, we randomly generate candidate solutions and create the fitness of each solution. We take the solution with good fitness and make a new solution through the evolutionary genetic operations like a crossover and mutation. So, offspring has a partial information from the parents and sometimes the offspring has a better genetic information between the parents so that means the offspring is expected to have a better information inside. Then, through this process, we will find and obtain the solution with the very fine fitness that can be the optimal solution for the problem or near optimal solutions in the problem. So, this is a simple explanation of the evolutionary computation and you can see you can watch the same explanation in this video. I recommend to watch this video later. Chang-Shing will provide my slide so you can access this video, the other videos, and webpage. Okay. I recommend the several applications that they are interesting enough to understand the mechanism of evolutionary computation. One is the 2D Buggy Car Design that we randomly generate the variety of shape of the buggy and that puts in the base strange load and finds the better shape through the evolution. This is the another example of Flappy Bird. Maybe, some of you will play this game. It's very very simple. You just pass through a pair of pipes and to avoid these objects but if you experience this application, you know this is a simple but very difficult. Then, the evolutionary computation will find the very nice controller through the evolutionary So, you can access these webpages from this QR code or the link and as I said you will find my slide in the webpage so please visit that these webpages if you are interested. I also recommend this one the 8-Queens problem and this is an interactive platform and you will know how to define solution and how to choose the collected genetic operations yeah I highly recommend this page. There are a number of advantages. The one is that the various solution representations. We can use the binary string 0 and 1, the integer string 1, 2, 3, 5 and so on, and also the real number 0.504, and the permutation string is a popular representation and is the order of the the string and also the tree structure can be used to represent a solution. And, also we don't need for the problem's mathematical model and we can optimize the problem at the black box optimization and thanks to that property, we can apply to various real-world optimization problems. So, this is one of the examples. The evolutionary computation finds the shape of front nose of Shinkansen, a good train, N700. The shape is very strange but probably the human designer cannot find this shape but the evolutionary computation finds this shape. And, the evolutionary computation also finds a very efficient main wing and this is an antenna in the spacecraft. The city is very strange but very effective so the evolutionary computation is very nice for the optimization problem and it can find the normal solutions. And, there are another advantage, the global search ability. The traditional optimization method is a single-point search for the starting point if the starting point is here, this point will be moved to the the local minimum but it's stuck. But, the evolutionary computation is a multiple-point search and they use the solution information and find the solution while avoiding local minimum,  local optimum. All right. The PSO is some optimization method inspired by collective behavior of swarm or blocks in nature. It can also have the the same advantage as evolutionary computation. The structure or algorithm structure is very similar and have the same almost same properties. They can solve the blackbox problem and they can perform the blackbox operation and find a normal solution and also it has a global search ability. The good thing is that it's a simple structure. The population is composed of the number of particles. The particle movement is defined by a best position in its experience and best position in the swam and finds the optimal solution. Ver easily. Here is a one of examples and this is a function of optimization problem and the peak is the optimum solution and using the population, the PSO finds the solution, the optimum solution. Later, after my talk, you will use a particle swarm optimization (PSO) to optimize the fuzzy membership functions. Now you know the fuzzy membership functions. The shape is like this and then you can change the shape by PSO. It is very powerful. All right. That's all my talk. Thank you very much and then enjoy next activities.