Summary of the n algorithm in FPGA machine learning 2

I have already mentioned many algorithms. For example, annealing algorithms, Ant algorithms, cultural algorithms, and taboo algorithms. Next we will write some other machine learning algorithms. Migration learning, integrated algorithms, vector machines, online learning, and other algorithms.

I'm Mao Hua Wang qq849886241. Technical blog http://blog.csdn.net/my_share

Migration learning algorithms: After machine learning, it is easier to learn other knowledge. For example, it is easier for people who learn tables to learn about chairs. People who learn to play chess are more likely to learn go. Boosting-based migration learning algorithm. Http://wenku.baidu.com/link? Url = oinVyarxctHwn8fbpLXueATxEkabROgIguH_QKHI9LfSEU35UZfzl7IuoISzkE8-Q6gCxRL9aLufvYWajoleX_PhEAh3mkPOGPxvzo3HE0W. In fact, this algorithm is available in the Dragon Star program. But I don't remember where it is. The process is complicated. In my memory, it is a relatively new machine learning algorithm. You need to know more about it in time.

PCA algorithm: Principal Component Analysis. The PCA method is very effective in terms of dimensionality reduction and feature extraction, whether it is image processing or machine learning. PCA algorithms are often used in a wide range of face recognition applications. The principal component face in face recognition is completed by the PCA algorithm. Neural networks are used in combination with face recognition. At present, face recognition is popular, so this algorithm requires in-depth understanding.

ICA algorithm: Independent Component Analysis Algorithm. The first example I see is related to MRI. It seems that it has a good performance in terms of images. The purpose of this method is to perform a linear decomposition of the observed data into statistical independent components. In fact, ICA and PCA belong to the same multi-variable data analysis method. However, each component obtained by ICA is not only independent of correlation, but also independent from each other and non-Gaussian distribution. Therefore, ICA can more fully reveal the essential structure (http://www.docin.com/p-564088508.html) between data ). Independent Component Analysis is a book that also describes PCA.

ICA is a powerful method in the field of blind signal analysis. It is also a method for finding hidden factors in Non-Gaussian distribution data. From the perspective of samples-features we are familiar with, the prerequisite for using ICA is that the sample data is generated by the implicit factor of independent non-Gaussian distribution, and the number of hidden factors is equal to the number of features, what we need is an implicit factor. PCA considers that features are generated by K orthogonal features (which can also be considered as implicit factors). What we need is the projection of data on new features. It is also a factor analysis. One is more suitable for signal reduction (because the signal is relatively regular and often not Gaussian distribution), and the other is more suitable for dimensionality reduction (What do we do with so many features, k orthogonal values ). Sometimes you need to combine the two.

ECO feature algorithm: Researchers use the ECO feature algorithm to implement ultra-high-precision object recognition. Http://www.cnbeta.com/articles/268834.htm. This is the only material. No information found. From the news description, this algorithm is good. To be followed up.

Artificial Immune Algorithm: a self-learning system developed based on the human immune system. It plays a role in pattern recognition. However, the results of this algorithm cannot be found. Fault Tolerance, stability, and robustness. The algorithm structure seems quite complex. You need to have a deep understanding of it before you can understand it.

Lasso algorithm: Algorithm. The documents are full of English, not very easy to understand (my English skills are limited ). Http://www.docin.com/p-149561253.html

The shooting algorithm is written here. In the previous n algorithms Summary 1, this algorithm is involved. However, there is no information.

Lasso-Based Face Recognition Algorithms. Http://oaps.lib.tsinghua.edu.cn/handle/123456789/137tsinghua's article. It seems that this lasso algorithm is still useful in face recognition algorithms. You need to learn more.

Decision tree: based on the known probability of occurrence in various situations, a decision tree is formed to calculate the probability that the expected value of the net present value is greater than or equal to zero and evaluate project risks, the decision analysis method that determines its feasibility is a graphic method that uses probability analysis intuitively (Baidu encyclopedia ). Applicable to probability models, that is, the probability of all situations must exist. In machine learning, a decision tree is a prediction model (Baidu encyclopedia ). Decision tree is a frequently used technology in Data Mining. It can be used for data analysis and prediction. Decision tree learning is also a common method in data exploration. The biggest advantage of decision trees is that they can understand the analysis process. Introduces logical expressions. Understanding about the surface of decision trees. However, you need to have a deep understanding of its specific application.

Neural Network: neural network. I wrote in my previous article that there is nothing to say. It is a very important machine learning algorithm. It is also established to simulate a human neural network. It is easy to fall into the local minimum value. But after the establishment, it will also have good results. There are also many improvements to it. Learn more. Google brain is a neural network with hundreds of millions of nodes. Therefore, its status cannot be ignored. In fact, I am also curious about how Google's car intelligence is implemented.

Deep Learning: its importance is nothing to say. Its reputation in machine learning is remarkable. It must be studied in depth. Let's talk about it. It is based on artificial neural networks. A multi-layer sensor with multiple hidden layers is a deep learning structure. In addition, it has good effects in speech recognition and image recognition.

Online Learning: Online learning is included in the Dragon Star program. It is a method of learning with new input. Baidu Library said that the method is good, but I did not say anything about it. According to my understanding, online learning algorithms are not a single algorithm. They work with neural networks and are a branch of neural networks. They cannot be used independently. Http://www.doc88.com/p-084413954013.htmlthis article can be the best I have ever found. There are very few materials. It may take some time to understand.

Offline algorithms: there is no data at all, and the concept of connection is vague. I still don't know if it is a machine learning algorithm.

Integrated Learning Algorithm: http://wenku.baidu.com/link? Url = Response

This article tells us that it is a method that combines many machine learning methods to achieve better results. The most important technology in generating and integrating individual networks is boosting [sch90] and bagging [bre96]. Integrated learning is a very important and popular branch in machine learning. It uses multiple weak classifiers to form a strong classifier. Its philosophical philosophy is that "Three skypixers have competed with Zhuge Liang ". Generally, a weak classifier can be composed of decision trees, neural networks, Bayesian Classifiers, and k-nn. Some scholars have theoretically proved that the idea of integrated learning can improve the performance of classifier, such as statistical reasons, computational reasons and representation reasons (http://www.cnblogs.com/tornadomeet/archive/2012/03/21/2409421.html ).

The three major algorithms in integration learning are boosting, bagging, and stacking. From these knowledge, we can see that it is a multi-algorithm integration process. Since there is no conflict and the effect is good, let's take a deeper look.

Support Vector Machine: I know from machine learning at Stanford University that support vector machine has excellent performance. Generally, SVM can only be used for the second-class problem, which is not effective for the second-class problem (Baidu encyclopedia ). It seems that there are some limitations. Let me take a look at the features of the image. The SVM method maps the sample space to a high-dimensional or even infinite-dimensional feature space (Hilbert Space) through a nonlinear ing p ), in this way, the problem of Nonlinear differentiation in the original sample space is transformed into the problem of linear differentiation in the feature space. simply put, it is the upgrade and linearity (Baidu encyclopedia ). It seems that binary classification can have good results. Http://blog.sina.com.cn/s/blog_89ba75c80101gxgn.htmlthis article is quite common.

EM algorithm: Maximum Likelihood Estimation. It's not a specific algorithm, but it often appears in machine learning algorithms.

Linear regression: Also available in Stanford machine learning. It is to give some real data about the house price and area, then use a function to approach, and then predict other house price and area. Http://wenku.baidu.com/link? Url = 9jwyim0cluj_rbpfplza_ce31rikmkg _-Just check it out.

What is the main idea of the Apriori algorithm ?? Is the correlation between two items, which is explained in http://blog.csdn.net/lizhengnan#/article/details/9061755. Two concepts involved: 3% of support: 3% of customers buy both milk and bread, and 40% of confidence: 40% of customers who buy milk also buy bread. Measure the two quantities to find the relationship between the data.

This figure shows many things.

Naive Bayes: Bayesian classification. I have heard a lot about it. It is a process of finding the posterior probability through the prior probability and Bayesian formula. Select a class with the highest posterior probability as the class to which the object belongs. However, as I know, the classification is good in some specific situations. It may also be used with other machine learning algorithms. Bayes is just such a formula.

K-means algorithm: K-means clustering algorithm. It is a method to find the appropriate clustering Center for these points. This clustering algorithm has some advantages in image segmentation. It is also very easy to find the center of each vertex. In fact, this algorithm also has many problems, such as the number of cluster centers. However, since it has some advantages in image segmentation, you can check it out.

FP-growth algorithm: an upgraded version of the Apriori algorithm. This is proposed to solve the problem that the performance of the Apriori algorithm tends to be low when the pattern is frequently mined.

Structure. I learned that the content of the Apriori algorithm is the same. But the performance is better.

 

The above is the machine learning algorithm that I can find. Some of the most common upgrade algorithms are not written in. Just familiar with it, not accurate. It's just a concept.

Summary of the n algorithm in FPGA machine learning 2