Table of contents
5. Reinforcement Learning (RL)
Today, let's talk about the knowledge related to machine learning algorithms, let's take a look!
Machine learning algorithms mainly include supervised learning, unsupervised learning, semi-supervised learning, transfer learning, and reinforcement learning.
-
Supervised learning
Supervised learning is a very common type of machine learning in machine learning, which is to train a model with known input and output, and map the input to the output.
Features: Learning objectives (such as actual values, annotations, etc.) are given.
Supervised learning can be divided into classification and regression according to whether the target result is discrete or continuous.
1.1 Classification
The prediction target value of this type is discrete, such as predicting whether it will rain, there are only two final results, rain or no rain.
Purpose: To predict the classification result of a new sample based on the historical results, the task for two classification results is called a binary task; two or more are called a multi-classification task.
Common algorithms: decision trees, random forests, K-nearest neighbors, logistic regression, support vector machines, artificial neural networks.
1.2 Regression
The forecast target value of this type is continuous, and a typical example predicts the price trend of a real estate.
Algorithms: Linear regression, AdaBoosting, etc.
-
unsupervised learning
The characteristic of unsupervised learning is that there is no need to think of data labeling, but continuous self-learning and consolidation according to the model, and finally learning through self-summarization. The learning model mainly includes clustering and dimensionality reduction.
2.1 Clustering
It mainly refers to the process of forming a collection of physical or abstract objects into multiple categories by similar objects, which can be understood as grouping according to similar principles.
Algorithms: The common ones are K-means algorithm, BIRCH algorithm, DBSCAN algorithm.
2.2 Dimensionality reduction
For high-dimensional data, which is very dependent on system resources and algorithm performance, dimensionality reduction is to process the unimportant information in high-dimensional data while retaining most of the important information. To put it simply, it is a process from complex to simple. If you make complex problems as simple as possible, it will be much less difficult to deal with.
Advantages: save space, save time consumed by algorithms, and reduce consumption of system resources.
Algorithm: Principal Component Analysis Algorithm (PCA)
3. Semi-supervised learning
A learning method that combines supervised learning and unsupervised learning. Semi-supervised learning uses large amounts of unlabeled data, as well as labeled data, for pattern recognition work.
Semi-supervised learning is suitable for a small number of labeled samples and a large number of unlabeled samples, which can achieve high accuracy prediction.
4. Transfer learning
Transfer learning refers to a learning method in which a pretrained model is reused for another learning task.
Source domain: existing knowledge; target domain: new knowledge to be learned.
5. Reinforcement Learning (RL)
RL, also known as reward learning, evaluation learning, and reinforcement learning, belongs to one of the paradigms and methodologies of machine learning.
The task of reinforcement learning is to enable smart devices to learn and try continuously like humans, and then to make the most ideal processing plan in different environments, to strengthen the process of continuous decision-making, and to discover which one is the best through continuous attempts. the best way.
Case: AlphaGo used reinforcement learning algorithm to defeat world champion Lee Sedol, Google YouTube video recommendation algorithm, etc.
