Google technical team takes you to practice machine learning with Google

Alan Turing, the father of artificial intelligence, predicted for a long time that "one day people will take their computers for a walk in the park and tell each other that my computer told a very interesting story this morning".

The core of machine learning is, "use algorithms to parse data, learn from it, and then make decisions or predictions about certain things." This means that you don't need to explicitly program computers to perform tasks, but teach computers how to develop algorithms to complete tasks. There are three main types of machine learning, each of which has its advantages and disadvantages: supervised learning, unsupervised learning and reinforcement learning.

Supervised learning

Supervised learning involves labeling data, and computers can use the data provided to identify new samples.

The two main types of supervised learning are classification and regression. In classification, the trained machine will divide a set of data into specific classes. For example, the spam filter of the mailbox analyzes the messages previously marked as spam and compares them with new messages. If a certain percentage is reached, these new messages will be marked as spam and sent to the corresponding folder; Unlike spam, it will be classified as normal and sent to the inbox.

The second is regression. In regression, the machine uses previously marked data to predict the future. For example, weather applications. Using the relevant historical data of weather (i.e. average temperature, humidity and precipitation), the mobile phone's weather application can view the current weather and predict the weather within a certain time range.

Unsupervised learning

In unsupervised learning, data is not labeled. In reality, most of the data are unlabeled, so these algorithms are particularly useful.

Unsupervised learning is divided into clustering and dimension reduction. Clustering is used to group by attributes and behavior objects. This is different from classification because these groups are not available to you. Clustering divides a group into different subgroups (for example, according to age and marital status), and then carries out targeted marketing. On the other hand, dimensionality reduction involves reducing variables in a dataset by finding commonalities. Most data visualization uses dimension reduction to identify trends and rules.

Reinforcement learning

Reinforcement learning uses machine history and experience to make decisions. The classic application of reinforcement learning is games. In contrast to supervised and unsupervised learning, reinforcement learning does not focus on providing "correct" answers or outputs. Instead, it focuses on performance, which is similar to human learning based on positive and negative consequences. If a child encounters a hot stove, he will soon learn not to repeat this action. Also in chess, the computer can learn not to move the king to where the opponent's pieces can reach. According to this principle, machines can finally defeat human players in the game.