3 Minutes to Understand What Machine Learning Is

What is Artificial Intelligence?

Artificial Intelligence (AI), as the name suggests, is how to be wise. To put it simply, artificial intelligence mainly studies how to use the functions of computers to do some tasks that must be performed by humans; in short, it is the process of performing human intelligence through computers can display intelligence similar to that of humans.

What is Machine Learning

Machine learning (ML) is to use algorithms to classify or predict the data collected. In the future, when new data is obtained, the trained model can be used to make predictions. If these performance evaluations can be achieved through using past data to improve is called machine learning.

ML has a wide range of applications, such as recommendation engines, weather forecasting, face recognition, fingerprint recognition, license plate recognition, medical diagnosis assistance, lie detection, document analysis, speech processing, etc.

What Is Deep Learning?

Such deep learning (DL) techniques are called deep neural networks (DNNs). Neural networks are just a way of constructing functions. When we ask questions and prepare a lot of historical data as "archaeological questions", we hope that we can train the neural network to see new questions and answer them correctly: for example, the neural network for dog recognition can be correctly trained after training. Name unseen dogs arranged in layers that loosely mimic the human brain, learning patterns of patterns.

Wondering where is the difference? Let’s understand them one by one.

Machine learning is an architecture included in artificial intelligence. Due to the recent popularity of machine learning, many people misuse artificial intelligence and machine learning. Artificial intelligence is a broad term. As long as it can show intelligent behavior, it can be called artificial intelligence. Even if there are many rule bases behind it, as long as it looks smart, it can also be called artificial intelligence.

There is no one algorithm suitable for all analysis, which is also commonly known as the no free lunch theorem. It is necessary to work hard on the data, and the models used vary according to the data. To discuss the quality of the algorithm, it must be based on specific problem types. But there are some useful models based on rules of thumb, such as Logistic Regression, SVM, Random Forest, and common ones in Deep learning: CNN (image recognition), RNN (text, speech), GAN, etc...

It is difficult to rely solely on the historical data of the stock market, which is commonly known as technical analysis. The reason is because machine learning is a rule behind finding data. If the rule behind it keeps changing, it is basically difficult for a machine to learn something, but it happens that the rules behind the stock market will keep changing. Suppose the machine uses the data of the past ten years to find that as long as a certain K-line rises twice in a row, there is an 80% chance that it will rise for the third time, but it is possible that this rule will fail tomorrow, and even make you lose a lot of money. But it is possible if combined with more information, such as real-time semantic analysis of social networks or financial news, if there is information about an iPhone battery explosion today, the machine learning program can instantly determine the probability of the stock price falling, and make a buy action. But it is also possible that there is negative news, but the price still keeps rising. For example, the negative news of Bitcoin keeps rising, but the price keeps rising, breaking through new historical highs. On the other hand, the price of other cryptocurrencies has no negative news but keeps falling. Therefore, it is quite difficult to use machine learning to predict success in the stock market.

Data science can be called data science as long as it uses data to analyze, and it can only use traditional statistics for analysis and prediction. Learn to equate.

Statistics has many mathematical proofs and assumptions, and it focuses on mathematical interpretability. A lot of statistical concepts are used behind the machine learning model, such as Linear regression, which is also derived from statistics. In addition, in practice, many machine learning relies on empirical rules and results theory to infer. For example, judge which model is better according to the prediction results, rather than prove it by mathematical deduction.

1. Supervised learning
2. Unsupervised learning
3. Semi-supervised learning
4. Reinforcement learning

Deep learning was originally a part of machine learning, but the speed of deep learning was too slow at that time, so it was replaced by SVM and other algorithms. However, due to the growth of GPU hardware performance in recent years, deep learning has overcome previous speed problems and achievements. Obviously, after it became a hot topic, many people discussed deep learning as a separate field from machine learning. And the origin of the name "deep" in deep learning is because there are many layers in the neural network-like hidden layer, which visually looks very deep.