Basic concepts of machine learning

1. probability

1.1 Marginal probability

1.2 joint probability

1.3 Conditional probability

https://blog.csdn.net/qq_42902997/article/details/123384049

2. sample (sample)、 Example (instance)：

An individual of the object of study . Equivalent to an example in Statistics (example,instance).

3. Understand Cauchy sequences and complete spaces

https://blog.csdn.net/DanielDingshengli/article/details/80492175?spm=1001.2101.3001.6650.5&utm_medium=distribute.pc_relevant.none-task-blog-2%7Edefault%7EBlogCommendFromBaidu%7Edefault-5-80492175-blog-102470627.pc_relevant_blogantidownloadv1&depth_1-utm_source=distribute.pc_relevant.none-task-blog-2%7Edefault%7EBlogCommendFromBaidu%7Edefault-5-80492175-blog-102470627.pc_relevant_blogantidownloadv1&utm_relevant_index=9
Definition of complete space ：

When all Cauchy sequences in space converge , The space is complete .

It can be understood in this way , All Cauchy sequences converge , It means that the whole space is non porous , This means that space is not missing any other elements , So space is complete .

4. Unsupervised learning

https://blog.csdn.net/weixin_40413961/article/details/114556424

5. mode collapse

“mode collapse”( Mode collapse ), It means that the generator produces a single result , Just to get the lowest discriminator loss D_loss, But it ignores the distribution of data sets .

6. variance

variance ： It describes the range of changes in the predicted values , The degree of dispersion , That is, the distance from its expectation . The greater the variance , The distribution of prediction result data is more scattered .

7. Manifold learning

https://blog.csdn.net/weiwei935707936/article/details/109079083

8. log likelihood

https://blog.csdn.net/wydbyxr/article/details/83212703

9. A priori in machine learning 、 Posterior and likelihood

10. KL Divergence

https://zhuanlan.zhihu.com/p/95687720

11. softmax

https://www.jianshu.com/p/695136c5647b

12. Orthogonal initialization （orthogonal initialize）

Orthogonal initialization ：

To solve the gradient disappearance under the depth network 、 The gradient explosion problem , stay RNN Parameter initialization method often used in .
https://blog.csdn.net/shenxiaolu1984/article/details/71508892

13. Inception score

Inception score It is to input the generated image into the classification model pre trained by the real image to measure the distance between it and the distribution of the real image .

14. PSNR

PSNR（ Peak signal to noise ratio ,Peak Signal-to-Noise Ratio）, Used to measure the difference between two images , For example, compressed image and original image , Evaluate compressed image quality ; Restore image and ground truth, Evaluate the performance of restoration algorithm, etc .

15. SSIM

SSIM（ Structural similarity ,Structural Similarity） Based on the assumption that human eyes can extract structured information from images , More in line with human visual perception than the traditional way . Structural similarity , It's a measure of the similarity between two images .

16. ablation study（ Ablation Experiment ）

Model reduction test .

In short, it's
See if the performance will be affected after canceling some modules .
According to Okam's razor law , Simple and complex methods can achieve the same effect , That simple method is more reliable .
actually ablation study It is an experiment designed to study whether some structures proposed in the model are effective .
For example, you put forward the so and so structure , But to determine if this structure is good for the end result , It is necessary to compare the results of the network without the structure with that of the network with the structure , This is it. ablation study.
https://blog.csdn.net/PolarisRisingWar/article/details/123557940

17. Regularization

From the perspective of the purpose of using regularization ： Regularization is to prevent overfitting .
https://blog.csdn.net/qq_53199855/article/details/122417094