News classification
Contact us
- Add: No. 9, North Fourth Ring Road, Haidian District, Beijing. It mainly includes face recognition, living detection, ID card recognition, bank card recognition, business card recognition, license plate recognition, OCR recognition, and intelligent recognition technology.
- Tel: 13146317170 廖经理
- Fax:
- Email: 398017534@qq.com
Machine learning overview
Machine learning overview
Machine learning overview
definition
Machine learning is the ability to teach computers to learn without detailed programming. For example, machine learning is often used to exercise computers to perform tasks that are difficult to accomplish with a programming wrist.
learning process
Generally speaking, both human and machine learning processes are roughly divided into the following four steps:
a) data storage: use observation, memory and association methods to provide factual basis for further reasoning;
b) Abstraction: transforming stored data into more general representations and concepts;
c) Generalization: Use generalized data to create learning and inferences that take further action in a new environment;
d) Evaluation process: providing a response mechanism for the learning process to weigh the applicability of the learned knowledge and bring about potential improvements;
Variety
Machine learning varieties include: supervised learning, reinforcement learning, and unsupervised learning, as shown in Figure 3-1.
a) Surveillance learning is the most common type of machine learning. Its exercise data is tagged. The purpose of exercise is to give new data (test data) the correct label. For example, if you stop the mail, you can classify the mail. At the beginning, we stop some mails and their labels (scum mail or non-scum mail) together. The learning model captures the contact between these mails and labels from time to time to stop self-adjustment and perfection. Then we give some new unlabeled messages, let the model stop the new mails as a sort of dregs.
b) Non-monitoring learning is often used for data mining. It is used to find out what is in the large amount of unlabeled data. Her exercise data is unlabeled. The purpose of exercise is to stop sorting or distinguishing the observed values. For example, non-monitoring learning should be able to distinguish "flowers" from a large variety of pictures based on the characteristics of all "flowers" pictures without giving any additional hints.
c) Reinforcement learning is often used in robotics (such as mechanical dogs), which accepts the current state of the robot. The purpose of the algorithm is to exercise the machine to make specific behaviors. The workflow is mostly: the machine is placed in a specific environment where the machine can continuously stop self-exercise and the environment will give a positive or negative reaction. The machine will be enhanced from past operational experiences and eventually find the best academic content to help it make the most effective behavioral decisions.
Machine learning classification
main application
a) Handwriting recognition: It recognizes handwritten text so that it can be converted directly into digital text.
b) Language translation: stop translating spoken or written text.
c) Speech recognition: recognizes speech and converts the speech segment into text.
d) image classification: stop the image from appropriate classification.
e) autonomous driving: enable the car to drive.
feature
A feature is a look-up value used to construct a prediction or model. For example, in picture classification, a pixel is a feature; in speech recognition, the pitch and volume of a sound are features; in an autonomous car, data from cameras, interval sensors, and GPUs are features.
It is important to extract meaningful features when setting up a model. For example, when categorizing pictures, the time of day is a meaningless feature, and when sorting the scum mail, the time of day is very useful information, because the scum mail is usually sent at night.
Common types of features in robots: pixel (RGB data), depth data (sonar or laser rangefinder), motion (encoder value), orientation selection or acceleration (gylon, accelerator or compass).
Note: In most cases, the more features available, the better, although this may be more time consuming; laser spacers are expensive, but the features they produce (360 degree depth maps) are useful for navigation; Cheap, deep data processing from cameras is very computationally intensive.
definition
Machine learning is the ability to teach computers to learn without detailed programming. For example, machine learning is often used to exercise computers to perform tasks that are difficult to accomplish with a programming wrist.
learning process
Generally speaking, both human and machine learning processes are roughly divided into the following four steps:
a) data storage: use observation, memory and association methods to provide factual basis for further reasoning;
b) Abstraction: transforming stored data into more general representations and concepts;
c) Generalization: Use generalized data to create learning and inferences that take further action in a new environment;
d) Evaluation process: providing a response mechanism for the learning process to weigh the applicability of the learned knowledge and bring about potential improvements;
Variety
Machine learning varieties include: supervised learning, reinforcement learning, and unsupervised learning, as shown in Figure 3-1.
a) Surveillance learning is the most common type of machine learning. Its exercise data is tagged. The purpose of exercise is to give new data (test data) the correct label. For example, if you stop the mail, you can classify the mail. At the beginning, we stop some mails and their labels (scum mail or non-scum mail) together. The learning model captures the contact between these mails and labels from time to time to stop self-adjustment and perfection. Then we give some new unlabeled messages, let the model stop the new mails as a sort of dregs.
b) Non-monitoring learning is often used for data mining. It is used to find out what is in the large amount of unlabeled data. Her exercise data is unlabeled. The purpose of exercise is to stop sorting or distinguishing the observed values. For example, non-monitoring learning should be able to distinguish "flowers" from a large variety of pictures based on the characteristics of all "flowers" pictures without giving any additional hints.
c) Reinforcement learning is often used in robotics (such as mechanical dogs), which accepts the current state of the robot. The purpose of the algorithm is to exercise the machine to make specific behaviors. The workflow is mostly: the machine is placed in a specific environment where the machine can continuously stop self-exercise and the environment will give a positive or negative reaction. The machine will be enhanced from past operational experiences and eventually find the best academic content to help it make the most effective behavioral decisions.
Machine learning classification
main application
a) Handwriting recognition: It recognizes handwritten text so that it can be converted directly into digital text.
b) Language translation: stop translating spoken or written text.
c) Speech recognition: recognizes speech and converts the speech segment into text.
d) image classification: stop the image from appropriate classification.
e) autonomous driving: enable the car to drive.
feature
A feature is a look-up value used to construct a prediction or model. For example, in picture classification, a pixel is a feature; in speech recognition, the pitch and volume of a sound are features; in an autonomous car, data from cameras, interval sensors, and GPUs are features.
It is important to extract meaningful features when setting up a model. For example, when categorizing pictures, the time of day is a meaningless feature, and when sorting the scum mail, the time of day is very useful information, because the scum mail is usually sent at night.
Common types of features in robots: pixel (RGB data), depth data (sonar or laser rangefinder), motion (encoder value), orientation selection or acceleration (gylon, accelerator or compass).
Note: In most cases, the more features available, the better, although this may be more time consuming; laser spacers are expensive, but the features they produce (360 degree depth maps) are useful for navigation; Cheap, deep data processing from cameras is very computationally intensive.