Machine Learning (Machine Learning) has many extensive applications, such as auto production, financial market forecasting stock market, medical automatic diagnosis and so on. This experience introduces its basic concepts and principles.
Tools/Materials
Method/Step
- Concept and definition
Machine learning is a field in the field of computer science, where existing data is used to predict or respond to future data. It is closely related to pattern recognition, computational statistics, artificial intelligence and other fields. Machine learning can use machines (computers and software) to mine meaning from known data, thereby giving the machine the ability to learn the environment. - Machine learning model
Machine learning is the process of converting measurements into parameters for future operations. After the machine obtains information from the environment, the data input to the system is divided into two parts, one is used to respond to the machine immediately, and the other part is input for learning. For example, in the automatic piloting of aircraft, the measurement of aircraft altitude is usually not directly used for control, but used to help select other flight parameters, such as speed, pitch angle, etc. - Elements of machine learning
- Data: This is the basis of machine learning. People can collect data from the network or sensors, and then the system learns it, and uses these parameters to identify parameters or train the system. The data can be collected by sensors, such as radar or cameras, or it can be polling data, stock market prices, accounting ledger data, etc. Or data collected by any other means. Machine learning is the process of converting measurements into parameters for future operations
- Model: It provides a mathematical framework for learning, which can be derived from human observation and experience.
- Machine learning methods
1) Supervised Learning: Supervised learning means that a specific training data set is applied to the system, and training is based on the collected training set. For example, karaoke scoring, if the collected data are all from popular singing, singers with famous sings may not get high scores.
2) Unsupervised Learning: Unsupervised learning does not need to use the existing prior knowledge, but clusters based on the collected data, so as to find hidden unknown features in the measurement data. For example, if unsupervised learning is used to train a facial recognition system, the system may classify the data set.
3) Semisupervised Learning: The semi-supervised method is between supervised learning and unsupervised learning. Usually only a small amount of input data is marked, and most are not.
4) Online Learning: The system continuously updates data, which can also be called "recursive learning". For example, the spam filtering system uses online learning. - Classification of machine learning:
(1) Control: A more elementary machine learning method, feedback control is used to compensate for the uncertainty in the system. The learning in control occurs in the adaptive control system and also in the general field of system identification.
(2) Artificial intelligence: Machine learning comes from artificial intelligence. The purpose of artificial intelligence is to make machines "think like humans", but some reasoning methods used in practice are different from human reasoning.
(3) Machine learning: It is considered by many to be real "machine learning", which arises from the process of self-statistics and optimization. The learning process can be carried out continuously in the recursive process. For example, based on years of stock purchase data, the trend can be predicted, but if a financial crisis occurs in a certain year, this model will have problems. Therefore, the recursive program will continue to merge new data. Pattern recognition and data mining belong to this category. - Common methods of machine learning:
(1) Return: Regression is a method of fitting data to a model. The model can be a multi-dimensional curve. The regression process fits data to a curve, producing a model that can be used to predict future data. Methods such as linear regression or least squares can be used. The example given in the figure below is based on the y=x function, adding noise to the y line, and using MATLAB's pinv pseudo-inverse function for least square fitting.
(2) Neural network: A neural network is a network used to simulate neurons in the human brain. Each "neuron" has a mathematical model used to determine its output from its input; for example, if the output is a step function with a value of 0 or 1, then if the input excitation will result in an output of 1, then the nerve Yuan can be said to be "inspiring."
The network forms multiple layers of interconnected neurons. Neural network is a form of pattern recognition. The network must use sample data for training, but does not require a priori model. The network can be trained to estimate the output of the nonlinear process, and then the network becomes a model.
Shows a simple neural network that flows from left to right, with two input nodes and one output node. There is a "hidden" neuron layer in the middle. Each node has a set of numerical weights that are adjusted during training. A complex neural network is a neural network between input and output with multiple intermediate layers (hidden layers)
(3) Support vector machine: Support vector machine (SVM) is a supervised learning model that uses related learning algorithms to analyze data for classification and regression analysis. The SVM training algorithm builds a model and assigns instances to categories. The goal of SVM is to generate a model for predicting target values ​​based on training data.
In SVMs, a kernel function is used to non-linearly map the input data in the high-dimensional feature space. In this feature space, a separation hyperplane is generated, which is a solution to the classification problem. The kernel function can be polynomial, S function and radial basis function. Only a subset of the training data is required; these are called support vectors. Training is done by solving a quadratic program, which can be done with many numerical software programs.
(4) Decision tree: The decision tree is a tree diagram used for decision making. It includes three types of nodes: decision nodes, opportunity nodes, and end nodes. The decision tree is easy to understand and explain, and its decision-making process is completely transparent. The difficulty in its application is to find the optimal decision tree for a set of training data.
There are two types of decision trees: classification trees that produce classification output and regression trees that produce numerical output. The following figure shows an example of a classification tree, which can help employees decide where to go for lunch.
(5) Expert system: Application of machine learning
(1) Face recognition: posture, lighting, obstructions (glasses, beard), makeup, hairstyle
(2) Character recognition: different handwriting styles.
(3) Voice recognition: voices of different people.
1) Use dictionary or language grammar to make judgments.
2) Sensor fusion: combine multiple modes, such as vision (lip shape) and voice to make judgments.
(3) Medical diagnosis: from symptoms to diseases
(4) Online advertising: predict whether a user clicks on an advertisement on the Internet.