We should be grateful for the AI-related technologies that exist in the technical field. Humans have not only spent decades researching and perfecting mathematical algorithms to make these wonderful and complex algorithms work, but in the process, we are constantly seeking breakthrough developments as an inspiration and in-depth study. So that the next generation of intelligence can exist on our planet. Nature, and everything it contains, is deeply rooted in the operation of AI and will grow here. The wildlife documentary film by David Attenborough is impressive. They have an incredibly detailed interpretation of the behaviors and attributes of many living things on Earth, which allows us to understand how these organisms fit into the natural ecosystem and work together to make the planet we live in flourish and ultimately Let the earth be what it is today. Although I am not David Attenborough, I still want to bring you into my wildlife documentary. The star creatures we are going to explore are nothing else, they are artificial intelligence algorithms that are directly inspired by nature. First, I need to introduce you to two algorithm concepts, Search/Pathfinding and Predictive Modelling. Search (path finding) algorithm The search algorithm is essentially a program designed to find the best/shortest path to the target. For example, the travel salesman problem is a typical search optimization problem in which you will get a list of cities and their distances. Travel salespeople need to visit each city once, and you must find the shortest route for them to minimize travel time and expenses (ensure you can return to the original city). The practical application of this problem is the delivery truck. Suppose there are 100 people in London who place an order online, and all the boxes are packed in the same truck. Couriers (such as DPD) must calculate the optimal route (balance distance/time) to deliver these packages from the warehouse (and ultimately to the warehouse) and ensure that the company wastes as little time and money as possible during the delivery process. Predictive modeling algorithm Today, predictive modeling is the focus of speculation. Data scientists around the world are shouting "neural networks" on the roofs of their comfortable office buildings, and companies such as Google are rushing around to try to solve the world with these small and complex "artificial brains" and their different variants. Various questions on it. In fact, predictive modeling uses statistics to predict results. You often hear data scientists trying to solve two kinds of predictive modeling problems, regression and classification. Regression is a black magic that discovers the correlation between two sets of variables. Classification is the process of calculating the probability that a data set belongs to a different group. 5 biological heuristic learning algorithms 1. Artificial Neural Network (ARTIFICIAL NEURAL NETWORKS) Feedforward Neural Network - the most basic type of neural network Algorithm Type: Predictive Modeling Biological Inspiration: Cognitive Brain Function (Neurons) Application examples: sentiment analysis, image recognition/detection, language correction, robot We start with the most common artificial intelligence (AI) algorithms. Neural networks are part of a subcategory called machine learning in the field of artificial intelligence. They are designed and constructed to mimic brain function at the neuron level (ie, axon and dendritic interactions, passing information through the system), generating predictive output through a series of "layers." Each layer provides an additional layer of data representation that allows you to model the most complex problems. Neural network is probably the most widely used machine learning algorithm, and it is also a research hotspot in the field of data science and machine learning. This concept was originally proposed in 1958 and is called the "perceived machine." Later, Geoffrey Hinton refined it and promoted it by companies such as Google and Facebook. Neural networks can be used to solve various types of problems, including natural language processing, visual recognition, and the like. This supervised learning algorithm supports both regression and classification problems, and examples of its application can be found in everyday consumer products, including smartphones and connected home devices. 2. Genetic algorithm Individual reproduction in genetic algorithm Algorithm type: search / pathfinding Biological Implications: Survival/Evolution of the Fittest (Cell Propagation) Application examples: data mining / analysis, robotics, manufacturing / design, process optimization In order to solve the search problem, the genetic algorithm adopts an evolutionary method similar to "survival of the fittest" in a group of successive generations. Each generation contains strings that mark the chromosomes we see in the DNA. Each individual in the group represents a point in the search space, so each individual is a possible candidate solution. In order to increase the number of solutions, we put individuals in the process of evolution. Every individual in the whole will compete for resources and spouses Individuals who win in each competition (usually) produce more sub-individuals than individuals who fail in competition. More “ideal†candidate individual genes are transmitted in the population, leading to superior parents who will produce more promising offspring. 3. Group/Collective Intelligence (SWARM/COLLECTIVE INTELLIGENCE) Ant colony optimization example, a collective intelligence algorithm Algorithm type: search / pathfinding Biological Revelation: Ant Colony / Fish Group / Bird Group Application examples: robotics, video game AI, manufacturing, route planning Ant colony optimization and particle swarm optimization are the two most common algorithms that conform to the concept of "collective intelligence." At a basic level, the algorithms we discuss all need to utilize multiple working agents. Each work agent exhibits a very basic behavioral ability that works collectively (as a whole) to trigger more complex, more urgent behaviors to solve problems. Ant colony optimization (ACO) is very different from particle swarm optimization (PSO). Both are designed to achieve urgent behavior, but in a different way. Like the real ant colony, ACO uses pheromone odors to direct individual agents to choose the shortest path. Initially, a random pheromone is initialized in the problem space. Subsequently, the individual agent begins to traverse the search space, releasing the pheromone odor during the search. In each time period, the pheromone will decay at a certain rate. A single agent makes decisions based on the intensity of the pheromone odor in front of it to traverse the search space. The stronger the odor in a particular direction, the more likely it is to go in that direction. The scheme with the strongest pheromone smell is the optimal solution. PSO is more concerned with the overall direction. Some single agents are initialized and then they start in a random direction. Every time period, each agent needs to make a decision whether to change direction. This decision will be based on the direction of the optimal solution (pbest/global optimal), the nearest neighbor (local optimum), and the current heading. The new way forward is usually a "compromise" of all these values. 4. Reinforcement learning Strengthening the behavior of agents in the learning environment Algorithm Type: Predictive Modeling Biological Revelation: Classical Conditional Reflection Application examples: video games, autonomous vehicle control, production line software, financial systems With the development of psychology and a process similar to classical conditioning, reinforcement learning can make a positive digital response to the beneficial actions taken by agents. Learning the concept of reinforcement learning is often easier than learning classic conditioning. Pavlov's Dogs, a study conducted in the 1890s, when Russian psychologist Ivan Pavlov was studying the reaction of dog saliva to feeding. Here you can find an article that explains this very well. Essentially, if a reinforcement learning takes a good action and takes a step toward completing the task, it will receive a numerical reward. The agent will learn to use a strategy so that each step will receive the maximum reward. Applying raw input to the algorithm allows the agent to develop its own perception of the problem and improve how to solve the problem in the most efficient way. Applications that combine reinforcement learning algorithms with other machine learning techniques are very common, such as neural networks. This is often referred to as deep reinforcement learning. Neural networks are often used to predict the rewards that reinforcement learning should receive in making a particular decision. Today, Deep Mind is a Google-owned company that has made great strides in this area and can use the Deep Q Learning method to solve more common problems (such as an algorithm that can play the entire Atari game library). And in the game "GO", defeated the world champion without any help). They are currently using this approach to handle more complex games such as "StarCraft II." For reference, Q Learning is a model-free version of the reinforcement learning algorithm. It can find the optimal action selection strategy for any finite Markov Decision Process. At program initialization, the Q value corresponding to each action value is defined by the developer and updated by the reinforcement learning algorithm at each time period. The following figure shows an example of an equation for updating a Q action-value pair. Q Learning Value update equation 5. Artificial immune system Algorithm Type: Predictive Modeling Biological Inspiration: The immune system Use cases: security software, autonomous navigation system, scheduling system, fault detection software The immune system is a system that protects the body from substances and pathogens by generating an immune response. The Artificial Immune System (AIS) is an adaptive system inspired by theoretical immunology and observational immune function applied to solve problems. AIS is a sub-area of ​​bio-inspired computing and natural computing, related to machine learning and artificial intelligence. There are usually multiple algorithms associated with AIS: Clonal Selection Dendritic Cell Negative Selection Artificial Immune Recognition Like the biological immune system, AIS is able to classify all cells in the system into "self" and "non-self" categories. A distributed intelligence team was used to take action on all cells. The two most important types of cells involved in immune activity are B cells and T cells (white blood cells for you and me). T cells are divided into three categories, one is used to activate B cells, one is used to bind and destroy foreign invaders, and the other is to suppress autoimmune problems. B cells are responsible for the production of antibodies, specific proteins that bind to antigens (toxic/foreign substances). Artificial immune systems are often used to monitor cyber attacks by monitoring intrusion detection and are often integrated into enterprise software. Unlike the other algorithms mentioned in this article, free online learning materials on this topic are very limited, and this technology may be the least developed of all the technologies mentioned in this article. Stainless Steel Wire Rope,Ss 304 Welding Filler Wire,Stainless Tig Wire Rope,304 Stainless Steel Wire Rope ShenZhen Haofa Metal Precision Parts Technology Co., Ltd. , https://www.haofametals.com
