What is machine learning?

Machine Learning extracts rules or patterns from data to convert data into information. The main methods are inductive learning and analytical learning.

Data is first preprocessed to form features, and then a model is created based on the features. The machine learning algorithm analyzes the collected data and distributes weights, thresholds, and other parameters for learning.

If you only want to divide data into different classes, then the "clustering" algorithm is enough. If you want to predict, you need a "classification" algorithm.

The opencv library contains machine learning methods based on probability statistics. Newer algorithms such as Bayesian Networks, Markov Random Fields, and graph models are still growing, so opencv has not yet been included.

There are many machine learning algorithms:

1. Mahalanobis

2. K-means unsupervised clustering method

3. Naive Bayes classifier features Gaussian distribution & relatively strict statistical independence Conditions

4. Determine the classifier based on the Decision number and classify data based on the threshold, which is fast. ID3, C4.5

5. A combination of boosting multiple discriminative sub-Classifiers

6. A Random forest consists of multiple decision trees.

7. Use the boosting Algorithm for face detection/Haar Classifier

8. unsupervised Generation Algorithm for clustering with the expectation of maximizing em

9. The simplest classifier of K-Nearest Neighbor

10. Neural Networks (multi-layer sensor) Train classifier very slowly, but quickly recognize

11. SVM supports classification or regression. Achieve optimal classification in high-dimensional space through the classification hyperplane

12. The genetic algorithm draws on the biological genetic mechanism and randomizes non-linear computing algorithms.

In short, I personally think that machine learning, data mining, pattern recognition, expert systems, and other fields are still quite chaotic. There may be differences between academia and the business world. The theoretical research on algorithms and the use of these methods to generate commodities are of particular concern. Many branches can be divided according to different fields and methods. But one thing is certain. The formulas and proofs proposed in the 1980s s are now becoming a line of code with the support of some servers such as Tomcat and IIS, I climbed onto the network, searched for information useful to the master, and then moved it to the network to generate products or half products. Look at the network cable on your computer. It's so small, but it's hard to imagine what it took from your computer and what it sent to you. A little far away, continue to talk about data.

Currently, I have used many algorithms, including neural networks (sensor, BP, RBF, and many other algorithms), genetic algorithms, and SVM, analytic Hierarchy Process (AHP), various regression, gray systems (domestic methods for predicting uncertainty), rough sets, Bayesian Networks, and time series analysis (there are also many ).

Learning and studying paper-based algorithm formulas is only the first step. We cannot ignore the basics. How can we use these methods to find the data and information we need on the vast Internet to meet customers' needs, this makes it easier for the people who need it. It seems that many enterprises have already entered the data warehouse, and have tasted great sweetness. Some enterprises have also maintained a reserve army, focused on R & D, ready to go to the front line at any time, and occupied the market. The competition in the wireless network market has reached a fierce situation, and the era of universal computing is approaching. It depends on the wearable hardware products and the embedded and fast response of software products. All in all, more and more humane, no one is willing to hold a laptop in the toilet, right?