Machine learning, deep learning and AI: What is the difference

The article is about machine learning, deep learning and AI: What is the difference? When it comes to new data processing techniques, we often hear many different terms. Some people say that they are using machine learning, while others call it artificial intelligence. There are still some people who may claim to be doing deep learning. What does this mean?

Although these terms have relatively specific meanings, there are overlaps and some differences in some aspects, but they are inseparable from big data. Along with the real breakthrough in data processing, it is bound to bring some inevitable hype. A proper understanding of these terms makes it easy for us to use them correctly.

Machine learning

At the most basic level, machine learning refers to any type of computer program that can "learn" by itself without having to be programmed by people.

The term originated from a famous paper “Computational Machinery and Artificial Intelligence” published by Alan Turing in 1950. It was proposed “Can robots think?”

Today, machine learning is a widely used term that covers many types of programs, mostly in big data analysis and data mining. Practical applications Most of the "brains" provided for predictive programs (including spam filtering, product recommendation, and fraud detection) are machine learning algorithms.

Machine learning, deep learning and AI: What is the difference?

▲ linear classification algorithm

Data scientists will be familiar with the differences between supervisory machine learning and unsupervised machine learning, as well as a combination of integrated models and methodologies, and semi-supervised learning combined with supervisory and unsupervised methods.

Supervised learning is the most common technique for training neural networks and decision trees. In supervised learning, the user training program generates answers based on known and tagged data sets.

Unlike supervised learning, data in unsupervised learning does not have any labels or labels. For data sets, unsupervised learning can determine that the data has two different clusters. Unsupervised learning algorithms may divide this data into two different clusters. So it is called a clustering algorithm. Unsupervised learning has a large number of applications. It is used to organize large computer clusters. The second application is the analysis of social networks.

Data scientists can write machine learning algorithms using a range of techniques and languages, including Java, Python, Scala, and more. They can also use a pre-built machine learning framework to speed up the process; Mahout is a popular machine learning framework on Apache Hadoop, and Apache Spark's MLlib library has become a standard.

Deep learning

Deep learning is a form of machine learning that can take advantage of supervised or unsupervised algorithms, or both. Although not necessarily new, deep learning has recently gained popularity, and the most notable applications for accelerating the resolution of certain difficult types of computer problems are computer vision and natural language processing (NLP).

Deep learning is a branch of representation learning (or feature learning) based on machine learning theory. The hierarchical learning process extracts advanced, complex abstractions as data representations, and the deep learning model produces results faster than standard machine learning methods. In simple English, an in-depth learning model will learn its own important features, rather than requiring data scientists to manually select relevant features, such as the consistency of the ears found in cat images.

Machine learning, deep learning and AI: What is the difference?

▲Deep learning pictures of cats

The “deep level” in deep learning comes from the deep learning model, usually the neural network. A Convolutional Neural Network (CNN) can be composed of many levels of models, where each layer takes input from the previous layer, processes it, and daisy-chains it to the next layer. AlphaGo, developed by Google's DeepMind team, defeated the world champion, which many believe is a sign of deep learning.

Machine learning, deep learning and AI: What is the difference?

▲ neural network can have many hidden layers

There are two main reasons why deep learning is so popular today. First, it was found that CNN runs faster on the GPU, such as NVidia's Tesla K80 processor. Second, data scientists realize that the vast database we collect can be used as a large-scale training database, which makes CNN greatly improve the accuracy of computer vision and NLP algorithms.

Most of the progress we've seen in developing autonomous driving can be attributed to the use of CNN's deep learning progress on the GPU, which helps to promote deep learning and further development in the field of artificial intelligence.

Artificial intelligence

Like machine learning and deep learning, artificial intelligence is not "new," but it definitely reflects a revival. The way people use the word is also changing.

When Turing designed his test for the first time, the term artificial intelligence was mainly retained in a technology that could imitate human intelligence. This past is an unreachable thing, just as we talk about time travel today.

Machine learning, deep learning and AI: What is the difference?

▲ short-lived Tay, Microsoft's AI chatbot

Today, artificial intelligence or AI is often used to refer to any type of machine learning program. In this regard, it began to replace “big data” and its linked “advanced analysis” and “predictive analysis”. For those who don't like the word "big data," this might be a good thing.

Some people are worried because of the development of artificial intelligence, Ma Yun stressed that there is no need to worry. "I think that the future machine will be smarter than humans, but it may not be more sensible than humans." Ma said that he hopes that human beings will be technically defeated by machines. But human wisdom is the core, and these machines should be fully utilized to play a role in dealing with diseases and poverty.