Source: AINLPer micro-channel public number ( click to look at it )
Edit: ShuYini
proofreading: ShuYini
Time: 2020-01-08
The main elements:
1, deep learning (DL) and machine learning link between (ML).
2, the use of deep learning (DL) do natural language processing (NLP) advantage.
Deep Learning (DL) is the relationship between machine learning (ML) of?
Deep Learning (DL) is a branch of machine learning (ML), but the difference between them is what it?
Machine learning is based on the data. Most humans rely on machine learning to identify and describe specific features of the data set. For example, to build a machine learning data scientists solution to the recognized text names, codes may be used to describe the specific features to look for, for example:
1, the case of the target word
2, word on the left and right of the target word
3, the target word in a particular substring, usually indicates a company or person
4, the target word hyphen
and so on.
A typical machine learning solution will eventually be hundreds of thousands or even millions of characteristics of manual design. Then, once the human hand-finished work to identify all of these features, the machine can do it? In this type of solution, the main work is to use a machine learning algorithm to adjust the weight of each feature to optimize the prediction accuracy . The computer is very good at this kind of numerical optimization, but these solutions are still heavily dependent on human thinking and learning problems.
So, how can we without requiring too much manual intervention, to help you learn the machine? We can use to characterize learning. Representing learning, computer identification data of its own characteristics, without the need for manual describes what you're looking.
It represents a simple form of learning, including what you might see in the introductory material in machine learning. Clustering algorithm, such as k-means and expectation maximization, a study showing that the data acquisition and unlabelled packets to look for patterns in the cluster. Dimensionality reduction is a large amount of data having a number of dimensions "flattening" of smaller dimension algorithm, also represents a good example of learning.
Depth study is based on this concept by creating a human brain-like systems using multi-layered study to characterize the performance of this system is superior to other methods of learning. By deep learning, you will enter into a large data set of a model, the model generates a learnable. Then, the model will learn algorithm represented by input to the other layer, a layer using the algorithm input data to generate a new learning FIG. According to the model of the "depth", for a given number of layers, the model This pattern is repeated over and over again. As its data output layer of each subsequent layer before use and then to generate from the input represents their learning. In the graph, it looks like this. This network produces a layered structure. Under this premise larger hierarchical representation, there are several series of deep learning model. In current practice, the depth learning to use most of the neural network. Like artificial intelligence, deep learning is a news and pop culture often misused term. Do not be those who will learn as depth articles related to intelligent computer terminology fooled! If you are not talking about learning neural network or hierarchical representation, probably it is not deep learning.
Why use deep learning?
Deep learning is an exciting natural language processing technology. Previously used manual design characteristics of natural language processing attempts are often too detailed and incomplete. They also took a long time to verify and improve. Deep learning is a relatively fast and flexible enough to adapt quickly to new data. This method avoids the lengthy design and design features manual verification cycles.
Since the depth data learning allows a computer to build their own characteristics, so it is almost common to learn a variety of information of the frame. This includes information about the world of language, visual information and context information.
But the best reason to explore the natural language processing depth study is that it is effective, but also much more effective than researchers had to try other techniques. Since about 2010, the first successful natural language processing, deep learning has made tremendous progress. However, the basic technical depth study first appeared in the 1980s and 1990s. So why do we only started to explore them in the last 10 years?
First, and perhaps most importantly, we now have a lot more data than the 80's and 90's . Popularity and popularity of the Internet means that we have gathered an unprecedented amount of data on almost everything from the products we buy to how we socialize. Internet is a large number of data samples consisting of language, including speaking freely from sources such as Twitter and a blog. When it comes to machine learning, especially in deep learning, it has a large number of data sets is crucial.
At the same time, there has been a faster machine and multi-core cpu and GPU , which helps to support the computing power needed for deep learning. In particular, the depth of learning is very suitable for parallel processing, it is now particularly cheap and efficient.
Finally, the new models, algorithms and ideas to make learning more efficient and flexible depth. This includes better and more flexible intermediate representation learning, more efficient use of learning methods between context and task shifting, and more effective end-to-joint system of learning.
ACED
Attention
More natural language processing knowledge, also please pay attention ** AINLPer ** No public, the best dry instantly delivered.
