artificial intelligence
Studies have let the computer simulation, extension and expansion of intelligent people a technical science. Mainly derived from large amounts of data to the machine learning can be calculated faster results than people, this may be the big data in my eyes.
Perhaps the big data analysis results, and then further on a level modeling of artificial intelligence called it! I still shallow understanding, but also a lot of advice, self-study notes is not easy to summarize.
Machine Learning
Is the use of the algorithm or logic, calculates (learning from data that is how to accomplish the task of training the learning process), generating a model, decision-making and forecasting model on real events through large amounts of data.
1. come from a machine learning method points, machine learning algorithms can be divided into supervised learning (such as classification), unsupervised learning (such as clustering), semi-supervised learning.
2. Traditional algorithms include k- adjacent algorithms, decision trees, Bayesian classification, clustering, support vector machines.
Depth study
Artificial neural network is a neural network structure comprising a plurality of hidden layers (depth neural network), is connected by optimization method and the like neuron activation function, to improve training efficiency, after generating the model, made by the model to actual events decision-making and forecasting.
The relationship between machine learning and deep learning of
Machine learning is a method for implementing artificial intelligence. Deep learning is a technique (new algorithm) machine learning implementation.
Comparison of machine learning and deep learning
1, application scenarios: scenarios
Application of Machine Learning in fingerprinting, features of the object, the detection fields is substantially achieved commercial requirements.
Deep learning is mainly used in the field of character recognition, face technical, semantic analysis, intelligent monitoring. Fast layout is also currently in the smart hardware, education, medical and other industries.
2, data dependencies
Machine learning can adapt to the amount of data, especially the smaller amount of data of the scene. In this case, the traditional machine learning algorithms use rule-making, performance will be better.
Precision depth learning requires a lot of training data, when there are few data volume, depth learning algorithm performance is not good.
3, hardware-dependent
Depth learning algorithm requires a lot of matrix operations, GPU is mainly used to optimize the efficiency matrix operations, so the GPU is the depth of learning to work must hardware.
Machine learning the hardware configuration requirements relative, deep learning is not so high!
4, the training algorithm time
Depth learning algorithm, because there are a lot of parameters included, requires a lot of time training and complete the training time may need to consume two weeks or longer!
Training machine learning consumes relatively little time, requires only a few seconds to several hours.
5, the predicted time
The predicted time of deep learning algorithm, compared to machine learning, requires very little time to run.
6, solution to the problem
Machine learning algorithms to follow standard procedures to solve the problem. It will be split into several parts problems, solve them separately, then then the results are combined to obtain the required answers.
Depth study places centralized way to solve the problem without having to split the issue, calling for a direct end to end solution to the problem.
7, interpretability
Deep learning can reach close to people's standards, but this is still a question. On the mathematical point of view, you can find out what a deep neural network node is activated. But we do not know what model should neurons, we do not know what these nerve cell layer to do together. So I can not explain how the results produced.
Machine learning algorithms are given clear rules, so to explain the reasoning behind this is easy.
8, wherein the processing
Performance machine learning algorithm depends on the extracted features of accuracy, and data processing features, you need more professional knowledge, and very time consuming.
Depth study attempts to acquire high-level features from the data directly, cut depth study of each issue feature extractor design work.
Machine learning and deep learning applications
1, computer vision for license plate recognition and face recognition applications.
2, for applications such as information retrieval search engine, - comprising a text search and image search.
3. Application of Marketing for automatic target recognition and e-mail marketing and other groups.
4, medical diagnostic applications, such as cancer and identify the abnormality detection or the like.
5, natural language processing, such as application sentiment analysis and photo tagging and the like.
6, unmanned.
to sum up
Application of machine learning algorithms in fingerprint recognition, face detection and other areas basically reached the level of commercialization or commercialization of the requirements of a particular scene, but further are exceptionally difficult every front, until deep learning algorithms, artificial intelligence began outbreak areas continue to expand artificial intelligence, such as: unmanned, preventive health care! Depth study appeared very early, but because there was insufficient amount of training data, computing power behind, so the final result is not satisfactory. Depth learning model requires a lot of training data to show a magical effect!