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
What is deep learning?
What is deep learning?
What is deep learning?
In order to make the learning performance of the neural network perform better, people can only try to stop the adjustment of a large number of repeated network parameters from time to time.
Deep learning is a multi-layer perceptron that contains multiple hidden layers. By combining the underlying features, it forms a more general high-level representation to describe the advanced attribute categories or features of the identified object. The intermediate representation of the self-generated data (although this representation is not known to humans).
Method of deep learning
Deep learning is "end-to-end" (end-to-end), the input is the original data (starting end), and then the output is directly the final destination (end), the intermediate process is unknown
Deep learning is a black box system, lacking explanatory
Cognitively, there is a direct change from one state or system to another state or system. This is the approach behind deep learning.
The big data era provides a valuable resource for us to understand the complex world – diverse and comprehensive data.
"All data" and "integration" in complexity science have a logical correspondence at a certain level.
Big data is a problem, and deep learning is one of them.
The artificial "carbon" is still in the air, and the intelligent "silicon" is unknown.
Artificial Intelligence's "Land and Lake Positioning"
Artificial intelligence, in simple terms, roughly imitates or reproduces the "carbon-based brain" with a "silicon-based brain."
Attribution of deep learning
Deep learning is a highly data-dependent algorithm, and performance is often enhanced from time to time as the amount of data increases.
There are two levels of analysis of machine learning:
(1) Facing the past (stopping the training of historical data collected), and discovering the form hidden under the data, called the descriptive analysis, mainly using the "reduction" approach.
(2) Facing the future, based on the built model, stop the prediction of the newly input data object, called predictive profiling, focusing on “deductive”
Modal definition of machine learning
The center of learning is to improve performance
If you want to do well in machine learning, you need to take three big steps:
(1) Modeling problem: how to find a series of functions to achieve the expected function
(2) Evaluation question: How to find a series of evaluation specifications to evaluate the quality of the function
(3) Optimization problem: how to quickly find the best performance function
Why use neural networks?
Cohesion: Trying to write a generic model, and then through data training, from time to time to improve the parameters in the model, know that the output results fit the expectations.
Cohesion theory believes that human thought is a combination of certain neural units. Therefore, it is possible to imitate the cognitive function of human beings at the network level, and to use the parallel processing form of the human brain to characterize the cognitive process.
Characteristics of artificial neural networks
Artificial neural network is a nonlinear, self-compliant information processing system. The system consists of a large number of disposal units that are connected to each other but are simple to use.
Artificial neural networks have four "non-" characteristics:
(1) Nonlinear: the activation function is nonlinear
(2) Non-limitation: the “scope” of any one neuron is not partial, but may involve the entire network through network connection.
(3) Very qualitative: the artificial neural network is constantly in the “update” state, with strong self-adapting, self-organizing and self-learning skills.
(4) Non-convexity: The current neural network usually adopts nonlinear activation functions such as Sigmoid, Tanh, ReLU, etc., which leads to the non-convexity of the objective function of the neural network.
In order to make the learning performance of the neural network perform better, people can only try to stop the adjustment of a large number of repeated network parameters from time to time.
Deep learning is a multi-layer perceptron that contains multiple hidden layers. By combining the underlying features, it forms a more general high-level representation to describe the advanced attribute categories or features of the identified object. The intermediate representation of the self-generated data (although this representation is not known to humans).
Method of deep learning
Deep learning is "end-to-end" (end-to-end), the input is the original data (starting end), and then the output is directly the final destination (end), the intermediate process is unknown
Deep learning is a black box system, lacking explanatory
Cognitively, there is a direct change from one state or system to another state or system. This is the approach behind deep learning.
The big data era provides a valuable resource for us to understand the complex world – diverse and comprehensive data.
"All data" and "integration" in complexity science have a logical correspondence at a certain level.
Big data is a problem, and deep learning is one of them.
The artificial "carbon" is still in the air, and the intelligent "silicon" is unknown.
Artificial Intelligence's "Land and Lake Positioning"
Artificial intelligence, in simple terms, roughly imitates or reproduces the "carbon-based brain" with a "silicon-based brain."
Attribution of deep learning
Deep learning is a highly data-dependent algorithm, and performance is often enhanced from time to time as the amount of data increases.
There are two levels of analysis of machine learning:
(1) Facing the past (stopping the training of historical data collected), and discovering the form hidden under the data, called the descriptive analysis, mainly using the "reduction" approach.
(2) Facing the future, based on the built model, stop the prediction of the newly input data object, called predictive profiling, focusing on “deductive”
Modal definition of machine learning
The center of learning is to improve performance
If you want to do well in machine learning, you need to take three big steps:
(1) Modeling problem: how to find a series of functions to achieve the expected function
(2) Evaluation question: How to find a series of evaluation specifications to evaluate the quality of the function
(3) Optimization problem: how to quickly find the best performance function
Why use neural networks?
Cohesion: Trying to write a generic model, and then through data training, from time to time to improve the parameters in the model, know that the output results fit the expectations.
Cohesion theory believes that human thought is a combination of certain neural units. Therefore, it is possible to imitate the cognitive function of human beings at the network level, and to use the parallel processing form of the human brain to characterize the cognitive process.
Characteristics of artificial neural networks
Artificial neural network is a nonlinear, self-compliant information processing system. The system consists of a large number of disposal units that are connected to each other but are simple to use.
Artificial neural networks have four "non-" characteristics:
(1) Nonlinear: the activation function is nonlinear
(2) Non-limitation: the “scope” of any one neuron is not partial, but may involve the entire network through network connection.
(3) Very qualitative: the artificial neural network is constantly in the “update” state, with strong self-adapting, self-organizing and self-learning skills.
(4) Non-convexity: The current neural network usually adopts nonlinear activation functions such as Sigmoid, Tanh, ReLU, etc., which leads to the non-convexity of the objective function of the neural network.