I will write down what I learned as a memorandum.
AI is an abbreviation for Artificial Intelligence, which enables processing of human-like information (simply speaking, a robot).
Machine learning digitizes perceptual information and given information and processes it. In other words, machine learning makes AI what makes it AI.
There are methods such as deep learning in machine learning.
Therefore, roughly speaking, AI (artificial intelligence) <machine learning <deep learning.
There are various methods for machine learning, and it is necessary to use them properly depending on what you want to do. It can be divided into three main types.
Supervised learning is the evaluation of unknown values using the results of analysis of the data that already exists. For example, seasonal ice cream sales forecasts (how much will be sold next month, etc.) I think it's the most familiar one.
"Regression" and "classification" are also typical for supervised learning. ・ Regression: Prediction from continuous numerical values like the previous example of Ike Cream
-Classification is to process which classification a certain data can be divided into. An example is classifying men and women based on specific figures such as height and weight.
Unsupervised learning is a method of classifying by learning to grasp the characteristics from multiple unknown data without a teacher, that is, without answering the given data.
For example, handwriting recognition (finding what the same person wrote).
In 2012, google developed an AI to find cats and it became a hot topic, which uses this unsupervised learning.
It is the same image that a child gradually learns various things as he or she lives.
This is mainly represented by "clustering" and "dimension reduction".
・ Clustering: Finding groups with common characteristics from data that does not have the correct answer as described above.
-Dimension reduction: When dealing with high-dimensional objects, reducing the dimensions makes the data easier to understand (lower dimensions). It's hard to grasp, but it can be said that it is classified only by the necessary data. By doing this, you can improve the accuracy of supervised learning and improve the calculation speed.
You will learn for things that have little or no data. An example is a cleaning robot. The size of the room and the arrangement of things are programmed to learn from the state where there is no data at first, and finally to clean the room efficiently.
It is also a technique that became famous because Alpha Go defeated the best Go player in the world.
Next time, I would like to actually perform simple regression analysis. As for future developments, I will post not only programming-related items such as coding and error handling, but also those related to neuroscience.
Recommended Posts