Machine learning is an important discipline in the AI field. As I have described earlier, I plan to focus on classification and identification of machine learning related to vision. Therefore, I may only focus on the role of video and vision in the analysis of each algorithm.
I am Mao Hua Wang 849886241. Technical blog http://blog.csdn.net/my_share
SA algorithm, fea-G algorithm, danova algorithm, SOA algorithm, shooting algorithm. This is from the web page, Baidu said they have their own algorithms.
Fea-G algorithm, danova algorithm, SOA algorithm, no information.
Shooting algorithms are called shot algorithms (results of Baidu and Bing ). However, the shooting algorithm is not a machine learning algorithm. It may be because the algorithm created by Baidu has been renamed or improved, so it is not made public, so there is no material. But I still want to explain what the shooting algorithm means. The shooting algorithm, both Baidu and the paper, is designed for Robot football matches. This is really a shot algorithm.
The robot points at R and the ball points at B. The best shot algorithm is to calculate 0 points and run the robot's walking route to arc c. Then shot the AG route. This is the shooting algorithm. It has nothing to do with machine learning. So this is not the case.
The SA algorithm is also called the annealing algorithm. It is of the same significance as the gradient descent method. Is an approximation model. It is actually a function approximation algorithm. A function may be a discrete point or a semi-cut function, which is a target function. Now I will use a function to approach that function. The error between the two functions is almost completed to a certain extent. To ensure a better solution, algorithms usually adopt slow cooling, multi-sampling, and low-end temperature, resulting in a long algorithm running time, however, from the searched data, it can improve the neural network, which is an important image recognition method. Therefore, it is necessary to make a detailed analysis of this direction.
Image metaphor of local search, simulated annealing, genetic algorithms, and Taboo Search: to find the highest mountain on the earth, a group of aspiring rabbits began to find a solution. 1. The rabbit jumped above the present. They found the highest mountain not far away. But this mountain is not necessarily Mount Everest. This is local search. It cannot guarantee that the local optimal value is the global optimal value. 2. The rabbit is drunk. He jumped randomly for a long time. During this period, it may go up or enter the ground. However, he gradually woke up and jumped in the highest direction. This is simulated annealing. 3. The rabbits took the amnesia pills and were shot into space, then randomly located somewhere on the earth. They do not know what their mission is. However, if you have killed some low-altitude rabbits in a few years, the productive rabbits will find Mount Everest themselves. This is the genetic algorithm. 4. rabbits know that the power of a rabbit is small. They told each other where the mountain was found and every mountain they looked for left a rabbit as a mark. They developed strategies for finding the next step. This is forbidden search. Gradient descent method, which works the same as the SA algorithm. There are only some improvements. Gradient Descent is part of a neural network. In fact, according to other data, it is better to use the annealing algorithm to improve the neural network. In my opinion, gradient descent is the fastest descent angle. For example, if you want to find a shortest path to the bottom of a basin, the gradient descent method takes one step at a time only from where you are currently, and the Newton method selects the direction, it not only determines whether the slope is large enough, but also determines whether the slope will become larger after you take a step. Therefore, it can be said that the Newton method looks a little farther than the gradient descent method and can go to the bottom faster.
Genetic Algorithms seem to be more complex. In terms of performance, it seems that they can easily converge to the global minimum value without being affected by the local minimum value. The disadvantage is that genetic algorithms are generally less efficient than other traditional optimization methods. Genetic algorithms tend to converge prematurely. There is no effective quantitative analysis method for the accuracy, feasibility, and complexity of the genetic algorithm. Taboo algorithms and semi-heuristic algorithms. The previous ones are all heuristic algorithms. I can understand a part of the rabbit's metaphor. I am talking about it. In fact, these are all of the same functions, and they are all close models. The taboo algorithm contains a table that records the rabbit's mountain. Of course, this table is limited, that is, to find the highest point by selecting the highest one. It is also called "tabu search algorithm ". From the current situation, it seems that the advantages and disadvantages are not found. It seems that you need to know more. There should be some new reasons. I have no idea about the astar algorithm. However, from the information I found, it seems that it has nothing to do with artificial intelligence. It seems like weak artificial intelligence. But it is also a kind of heuristic search algorithm. When searching for information, it seems that it is rarely related to artificial intelligence. So I don't have to go into depth. It's good to have such a concept. Ant Colony Algorithms and ant colony algorithms have the following advantages: ① they have strong versatility and can solve many path optimization problems that can be converted into connected graph structures; ② they also have positive and negative feedback characteristics, the positive feedback feature uses local solutions to construct global solutions, and the negative feedback feature, I .e., the volatilization of information elements, avoids the algorithm from falling into local optimum. ③ it has the characteristics of indirect communication and self-organization, there is no direct connection between ants, but indirect information transmission is carried out through the information elements in the path. Self-organization enables the power of the Group to solve the problem. However, the basic ant colony algorithm also has some disadvantages: ① from the complexity of the ant colony algorithm, this algorithm requires a long search time compared with other algorithms; ② after a certain degree of search, this algorithm is prone to the "stagnation" of the completely consistent solutions found by all ant financial, which limits the search space. From the text, it seems that the ant colony algorithm is quite complicated. Relatively new, performance seems acceptable, it is worth further understanding. So I am going to spend time learning about ant colony algorithms. Http://blog.163.com/ykn_2010/blog/static/1420333362012111411258466 /. It turns out that the ant colony algorithm can constitute a book. It seems that there are many contents. Particle swarm algorithm (PSO) is a type of Evolutionary Algorithm Similar to simulated annealing algorithm. It also starts from a random solution and finds the optimal solution through iteration, it also evaluates the solution quality through fitness, but it is simpler than genetic algorithm rules ,. This algorithm has attracted the attention of academia for its advantages such as easy implementation, high accuracy, and fast convergence, and demonstrates its superiority in solving practical problems. Particle swarm is a parallel algorithm. This is the description in Baidu encyclopedia. Particle swarm algorithms also have specialized tutorials. Published in 10 years. From the overall description, it seems to be updated and an upgraded version of the preceding annealing algorithm. According to the master thesis of Huazhong University of Science and Technology. It seems to be part of a neural network. You may need to know more about it. The greedy algorithm (beginning) must be noted that the greedy algorithm can be used together with the randomization algorithm, so there will be no more examples. In fact, many intelligent algorithms (also known as heuristic algorithms) are essentially greedy algorithms combined with randomization algorithms. Although such algorithms are also the result of local optimal solutions, they are closer to the optimal solution than pure greedy algorithms. For example, genetic algorithms and simulated annealing Algorithms (included in Baidu encyclopedia ). This statement illustrates the genetic algorithm. The annealing algorithm is a combination of greedy algorithms and randomization algorithms. It is worth noting that the greedy algorithm is not completely unusable. Once proven, the greedy policy is an efficient algorithm. Greedy algorithms are still one of the most common algorithms. This is because they are simple and easy to construct. It is not very difficult to construct greedy policies (in Baidu encyclopedia ). From the overall understanding, greedy algorithms have many disadvantages, and seem to have little to do with artificial intelligence machine learning. In the greedy algorithm, the best choice is made only in the current state, that is, the local optimal choice. Then, how does one feel like classification decision tree in the local optimum ?? But it is not a classification process. Pass. Cultural algorithm. In 2002, David proposed a GP-based dual-cultural algorithm framework. They also use cultural algorithms for image segmentation, dynamic optimization, and data mining. In September 1995, I tried my best to use the cultural arithmetic method to solve the problem and solve the problem. The dual-cultural algorithm framework is used for image segmentation and data mining. Bytes. However, the overall structure seems special. You may be concerned about it. The KMP algorithm is a pattern matching algorithm that does not understand its connection with machine learning. However, the matching function is needed. Considering that the KMP algorithm is not used in combination, we do not consider further understanding for the moment. There are still a lot of algorithms, which should be written here first. We also need to understand the integration of algorithms, online learning, offline learning, AI, Support Vector Machines, neural networks, and deep learning. Coming soon.
Algorithm summary in n of FPGA Machine Learning 1
