What is the relationship between deep learning and AI, and what is learned?

Deep learning is a general term for a class of pattern analysis methods, and in terms of specific research content, it mainly involves three types of methods:

1) A neural network system based on convolutional operations, i.e., convolutional neural networks (CNNs).

2) Autoencoder neural networks based on multilayer neurons, including auto encoder and sparse coding, which have received extensive attention in recent years.

3) The Deep Belief Network (DBN) is pre-trained in the form of multi-layer autoencoder neural network, and then combined with the discrimination information to further optimize the neural network weight.

As a technology to realize machine learning, deep learning has expanded the scope of artificial intelligence, mainly applied to image recognition, speech recognition, and natural language processing. Driving the market beyond the autonomous driving and robotics industries to non-technical industries such as finance, healthcare, retail, and agriculture, AI engineers with deep learning skills have become a hot job for companies of all types. 、

The deep learning created by Zhonggong Education and experts from the Chinese Academy of Sciences is divided into eight stages:

Overview of the first phase of AI and introduction of cutting-edge application results.

The latest applications of deep learning.

Single-layer Deep Learning & Machine Learning.

The relationship and development of artificial intelligence.

The second stage is the principle of neural network and the practical practice of tensorflow.

Gradient descent optimization method.

The basic structure and training process of a feedforward neural network.

Backpropagation algorithms.

Install the tensorflow development environment.

Computational Graph" programming model.

How image recognition works in deep learning.

The third stage is the principle of recurrent neural network and the actual practice of the project.

Language models and word embeddings.

The learning process of word embedding.

The basic structure of a recurrent neural network.

Time series backpropagation algorithm.

The basic structure of a long short-term memory network (LSTM).

LSTM implements language models.

The fourth stage is the principle of generative adversarial network and the actual project practice.

The basic structure and principles of generative adversarial networks (GANs).

The training process of the GAN.

GaN is used for the implementation of ** generation.

The fifth stage of deep learning is distributed processing and project practice.

Multi-GPU parallel implementation.

Distributed and parallel environment construction.

Distributed parallel implementation.

The sixth stage is deep reinforcement learning and project practice.

Introduction to reinforcement learning.

The in-depth decision-making mechanism of the agent (Part I).

The in-depth decision-making mechanism of the agent (middle).

The in-depth decision-making mechanism of the agent (below).

The seventh stage of the license plate recognition project is practiced.

Data set introduction and project requirements analysis.

OpenCV library introduction and license plate positioning.

License plate positioning. License plate recognition.

Participant project case review.

Introduction to the Frontier Technologies of Deep Learning in the Eighth Stage.

An introduction to the cutting-edge technologies of deep learning.

Meta-learning. transfer learning, etc.

Does deep learning have anything to do with the word a, what to learn, well, I don't know about this, you can go to QQ Browser

Deep learningRelationship to machine learning: Machine learning is the foundation of deep learning. InMachine visionand deep learning, where the power of human vision and understanding of visual information can be reproduced or even surpassed.

With the help of deep learning, as part of the debate of machine learning. Another technique in machine cooktop learning is, for example, the "super vector machine". In contrast to deep learning, the functionality must be defined and validated manually

in computer vision.

In the field, if a panda is recognized, the machine will learn by telling the machine various characteristics of the panda, such as nose, eyes, mouth, hair, etc., so that the machine can realize that it is a panda with these characteristics.

Brief introduction. The method of deep learning is to give the machine a **, let the machine extract the features by itself, and then **find out whether it is a panda, if it fails, the neural network passes forward, tells the neural network that there is an error, and re-identifies until the recognition is correct, the most famous is the CNN convolutional neural network that has been popular in recent years.

At its most basic, machine learning uses algorithms to parse data, learn from it, and then make decisions about real-world events. Unlike traditional software programs that are hard-coded to solve specific tasks, machine learning is "trained" on large amounts of data, using a variety of algorithms to learn how to complete a task.

Machine learning is a multidisciplinary discipline.

Probability theory is involved.

Statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in how computers simulate or implement human learning behaviors in order to acquire new knowledge or skills, and to reorganize existing knowledge structures to continuously improve their own performance. It is the core of artificial intelligence, and it is the fundamental way to make computers intelligent.

Deep learning (DL) is a new research direction in the field of machine learning (ML), which was introduced into machine learning to bring it closer to the original goal of artificial intelligence (AI).

Deep learning is the study of the intrinsic rules and representation levels of sample data, and the information obtained during these learning processes is of great help to the interpretation of data such as text, images, and sounds. The ultimate goal is for machines to be able to learn analytically like humans, and to recognize data such as text, images, and sounds. Deep learning is a sophisticated machine learning algorithm that achieves far more performance in speech and image recognition than previous technologies.

Deep learning has achieved a lot in search technology, data mining, machine learning, machine translation, natural language processing, multi** learning, speech, recommendation and personalization techniques, and other related fields. Deep learning enables machines to imitate human activities such as audio-visual and thinking, solves many complex pattern recognition problems, and makes great progress in artificial intelligence-related technologies.

It has been listed as one of the top 10 breakthrough technologies in 2013 by MIT Technology Review (Deep Learning), which exists as a neural network learning algorithm in ML. Artificial intelligence is currently divided into weak artificial intelligence and strong artificial intelligence, and neural networks have become today's DL. In fact, the so-called "artificial intelligence" that can be achieved by current technology is weak AI, which is only used to extract powerful features; The other wants to develop it into a new branch of learning, end-to-end, which may be a breakthrough to achieve strong AI in the future1.

Or to put it another way. Deep learning and AI. When DL was not yet on fire.

DL and ML actually have a subtle relationship, with the rise of computing resources and big data, Ultron is the kind of strong AI (or even boss-level), which is the end-to-end "deep learning idea" I mentioned above. Essentially, AI is a broader concept than deep learning, a technology (I prefer to call it an idea). 2。

Deep learning, on the other hand, is regarded as a feature extractor, which is a technology or idea in AI. Deep Learning & ML

A neural network.

Areas that are layered to understand complex patterns and relationships in data. When the output of one neural network becomes the input of another neural network, effectively superimposing them, the resulting neural network is "deep".

The result is for machines to mimic human neural networks, making simple imitations to analyze data, images, speech, and more.

It's a method of machine learning, and you can visit and learn about it here.

Deep learning, one that is thought of as a feature extractor, is a technology or idea in AI. Deep Learning & ML

Machine learning is the method of implementing artificial intelligence, and deep learning is the technology of realizing machine learning. Machine learning requires human assistance (semi-automated) to implement artificial intelligence, and deep learning fully automates the process.

No, current deep learning is not a strong artificial intelligence. Current deep learning is the modeling of problems and solving specific problems. For example, if you let the autonomous driver play Go, and let the Go AI classify **, this will not work.

The essence of AI is to allow machines to simulate the process of human "learning", and learning is not simply memorizing and storing, but also needs to apply what you already know to the unknown, such as prediction and reasoning. In addition to deep learning of this kindConnectionismSchool, but also a few friendsSymbolismwithEvolutionismI correspondedInductionDeductionEvolutionLearning methods for solving "learning" problems.

SymbolismIt can be understood as storing a piece of knowledge, so that the machine can make deductive reasoning through the knowledge, such as: "it may rain on a cloudy day" and "bring an umbrella when it rains". And today the machine finds that it is a cloudy day, and it can conclude that "you should bring an umbrella on a cloudy day today", then it can advise you to bring an umbrella.

Symbolism believes that everything in the world can be expressed in terms of logic, axioms. But the problem is that if you want to build strong AI, you need a very large knowledge base. Secondly, the fact is that not everything in this world can be expressed in axioms and logic, at least we can't do it now.

Then, only logic and axioms will not work. ConnectionismThis introduces "experience and sensibility". Connectionism uses the method of model (rationality) + data (perceptual) to extract features from data (experience) and find out the laws behind these features.

Finally, these laws are used to make inferences. As long as there is enough data, patterns can be found. For example, distinguish between intervals, slang, and so on.

But the reality is often unsatisfactory, and some laws cannot be found through data, such as chaotic systems, power-law distributions, etc. To give a less appropriate example: through the knowledge of chemistry and physics, we can know the laws of motion and reaction of atoms and molecules, but when they are combined and become a human being, it is difficult for you to prepare the laws of human behavior and thought through the knowledge of chemistry and physics at the molecular level.

SoEvolutionismHe completely abandoned reason, retained only "experience", regardless of whether the world had laws or not, and spoke only through "practice". The most typical example of evolutionism is the genetic algorithm. Let's take a figurative example:

Bacteria live in this world, and the world is changeable (irregular, regular), the dead bacteria are eliminated, and the surviving ones become stronger", this is the answer given by bacteria for "learning".

In the end, you should be able to understand that deep learning is only a combination of "rationality + experience", and its intelligence is far from enough for "learning".

Artificial intelligence was the first to appear, and it is also the largest and outermost concentric circle; followed by machine learning, a little later; The innermost side is deep learning, the core driver of today's artificial intelligence.

In the fifties, artificial intelligence was once extremely favorable. After that, some smaller subsets of artificial intelligence developed. First machine learning, then deep learning. Deep learning, in turn, is a subset of machine learning, and it's having an unprecedented impact.

Some people say that artificial intelligence (AI) is the future, artificial intelligence is science fiction, and artificial intelligence is also a part of our daily lives. These evaluations can be said to be correct, depending on which kind of artificial intelligence you are referring to.

Earlier this year, GoogleDeepmind's AlphaGo defeated South Korea's Go grandmaster Lee Se-yeong Jiudan. In describing Deepmind's victory, artificial intelligence (AI), machine learning, and deep learning are all used. All three played a role in Alphago's defeat of Lee, but they weren't the same thing.

Now, image recognition trained in deep learning can even do better than humans in some scenarios: from identifying cats, to identifying early components of cancer in the blood, to identifying tumors in MRI. Google's AlphaGo first learned how to play Go, and then trained with it to play chess on its own.

The way it trains its own neural network is to play chess with itself constantly, over and over again, and never stop.

Deep learning enables machine learning to enable a wide range of applications and expands the scope of artificial intelligence. Deep learning has done a variety of tasks in a way that makes all of the machine's auxiliary functions possible. Driverless cars, preventative healthcare, and even better movie recommendations are all within reach, or on the verge of being realized.