일 | 월 | 화 | 수 | 목 | 금 | 토 |
---|---|---|---|---|---|---|
1 | ||||||
2 | 3 | 4 | 5 | 6 | 7 | 8 |
9 | 10 | 11 | 12 | 13 | 14 | 15 |
16 | 17 | 18 | 19 | 20 | 21 | 22 |
23 | 24 | 25 | 26 | 27 | 28 |
- cv
- 생성모델
- userwidget
- ue5.4
- Generative Model
- animation retargeting
- dl
- 딥러닝
- RNN
- multimodal
- motion matching
- Stat110
- ddpm
- BERT
- WBP
- 폰트생성
- Unreal Engine
- Diffusion
- 모션매칭
- deep learning
- 언리얼엔진
- Few-shot generation
- 디퓨전모델
- WinAPI
- NLP
- GAN
- UE5
- CNN
- Font Generation
- 오블완
- Today
- Total
Deeper Learning
[Statistics 110] Universality of Uniform distribution 본문
Universality of Uniform distribution(균등분포의 보편성)
Uniform distribution을 사용하여 임의의 확률분포를 만들어낼 수 있는 성질
condition
$u \sim Unif[0,1]$, $F$는 원하는 분포의 cdf
(단 $F$는 증가함수, 연속함수)
$F$는 cdf이므로 0이상 1이하의 값을 가진다.
$F$를 cdf로 가지는 확률변수 $X$를 구하는 방법
- $X = F^{-1}(u)$ 이면 $X$의 cdf는 $F$를 따름
proof
$P(X \leq x) = P(F^{-1}(u) \leq x)$ (가정에 의해)
양변에 역함수를 적용하면 (앞서 언급한 F에 대한 제약에 의해 부등호의 방향은 동일)
$P(F^{-1}(u)\leq x) = P(u \leq F(x))$
위 식에서 F(x)는 0~1의 값을 가진다. (cdf이기 때문)
0~1의 정의역을 가지는 Uniform distribution에서 길이는 곧 확률이 된다.
따라서 $P(u \leq F(x))$ 자체가 $F(x)$와 동치
$\therefore P(X\leq x) = F(x)$
Usage
uniform 난수 생성 코드만 있어도 다른 분포의 난수를 생성할 수 있음
Gaussian distribution의 cdf의 inverse에 uniform distribution에서 샘플링한 R.V 를 feeding하고 나온 output을 histogram 등을 통해 visualization하면 Gaussian distribution과 비슷함을 알 수 있다.
역으로 X~F를 해당 확률변수의 CDF에 input으로 주면 uniform distribution을 얻을 수 있음
Reference
[0] Harvard Statistics 110, Lecture 8
'Statistics & Math' 카테고리의 다른 글
Gram-Schmidt Process & QR Decomposition (0) | 2022.11.01 |
---|---|
[Statistics 110] Law of the unconscious statistician (LOTUS) (0) | 2022.05.21 |
[Statistics 110] 확률변수와 확률분포 (0) | 2022.03.27 |
Eigenvector & Eigenvalue (0) | 2021.07.20 |
Community Detection (0) | 2021.02.25 |