Machine learning learning notes

There are several categories of machine learning

1、 Supervised learning (Supervised Learning)：

Train a model on labeled data , The task is to predict the label （ Example ： In the example of house prediction, the prediction of house price . Or predict the category of the picture ）

There is a model that we pay more attention to Self supervised learning algorithm (Self_Supervised_Learning ), It is a kind of supervised learning , Some labels will be generated , But these labels come from the data itself （ Example ： Word insertion ,BERT）

2、 Semi-supervised learning （Semi-Supervised Learning）：

Model training on labeled and unlabeled data , Task 1 of the model ： Learn the model to predict labels （ Try to use information that is not well marked ）, Task 2 ： It is to use the model to label the data without labels （ Example ： Self training ）

3、 Unsupervised learning (Unsupervised)：

Data has no label , The task is not the label of prediction data 【 clustering algorithm 、 Estimated distribution （GAN）】（GAN： Generate a confrontation model ） It is also such a type

4、 Reinforcement learning (Reinforcement)

The model interacts with the environment , Get observation points from the environment before learning , Do some feedback again （ It's more like a human way of learning ）

Reinforcement learning is a hard part , Most of the discussion is about supervised learning and unsupervised learning

The more important thing in the training of supervised learning model

1、 Model （Model）： Predict the output through input ;

2、 Loss function (Loss)： Used to measure the difference between the value predicted by the model and the true value

3、 The goal is (Object)： In model training , Optimization function

4、 Optimize (Optimization)： There are no parameters specified in the model （ Deliberately learned parameters ） Fill in the actual data , Minimize the loss of our objective function ;

Supervise the classification of learning ：

1、 Decision tree （Decision Tree）： Using trees to make decisions ;

2、 Linear model （Linear methods）： The decision is made according to the linear combination of this input ;

3、 Nuclear method （Kernel machines）： The kernel function is used to measure the similarity between two samples .

4、 neural network （Neural Networks）: A multi-layer neural network is used to learn the representation of a feature , So that there will be a good representation in the following linear method .

summary ：

Video learning link ：https://www.bilibili.com/video/BV1Nv411u7bW?spm_id_from=333.999.0.0 Add link description