I tried to organize the terms of Machine Learning that I learned in my own way.
In conventional programming, humans explicitly give commands to personal computers. I was in control of the program. This is called explicit learning.
This technique has its limitations, for example, for many patterns such as spam filters. It is difficult for humans to explicitly logic all spam and prevent it 100%. (Hard) Therefore, "Opportunity learning" is a concept devised by Arthur Samuel in 1959, "A program autonomously learns data from data and phenomena that human beings do not program one by one to create logic."
There are two main learning methods for machine learning.
・ Supervised Learning ・ Unsupervised Learning
There is learning data in advance, and learning is performed using it as a guide. The data to be learned is called the Traning Data Set.
As an example of Supervisored Learning There is Alpha go that can learn hundreds of millions of Go game record data and take the best move with respect to the value of x, isn't it?
To summarize the flow
Pass the value x ↓ Predict the result by comparing the learned data with a by learning by looking at the preset data. ↓ Returns the results obtained by learning
Example) A studied for 7 hours-> Predict the score using the learning data-> A should get 75 points! Predict
I think it looks like this.
Furthermore, Superviored Learning can be roughly divided into 3 weeks.
・ Regression ・ Binary classification ・ Multi-label classification
Predicting results with a wide range of predictions
Example) Predict how many points (0 to 100) you will get by looking at the time you studied
Learning data to classify true / false results
Example) Predict whether to pass or fail the exam by looking at the time spent studying I think it is the simplest learning method in Supervisored Learning.
Multi-label classification
Learning data to classify multiple results
Example) Predict units such as A, B, C, D, E, F by looking at the time spent studying
It is to learn from the value of the request without having the training data in advance.
To give an example of Unsupervised Learning, there is Google News that autonomously groups similar articles. This is because it is difficult to anticipate similar articles in advance and create a dataset Each time, you will see the content of the article, learn the data, and display similar articles.
Next time, I would like to summarize Linear Regression in Superviosrd Learning.
Recommended Posts