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 |
- BERT
- 모션매칭
- WBP
- userwidget
- deep learning
- Stat110
- CNN
- multimodal
- dl
- Diffusion
- cv
- RNN
- 폰트생성
- Unreal Engine
- NLP
- ddpm
- 디퓨전모델
- UE5
- 언리얼엔진
- Few-shot generation
- GAN
- motion matching
- WinAPI
- 오블완
- animation retargeting
- 생성모델
- Font Generation
- 딥러닝
- ue5.4
- Generative Model
- Today
- Total
Deeper Learning
[Statistics 110] Lecture 13: Normal Distribution 본문
[Statistics 110] Lecture 13: Normal Distribution
Dlaiml 2022. 12. 10. 17:37Normal Distribution(정규분포)
Central Limit Theorem(중심극한정리)
적당히 많은 수의 i.i.d 확률변수를 더하면 더한 평균값의 분포는 정규분포에 수렴한다
대칭인 종모양의 pdf
평균이 0이고 분산이 1이면 표준정규분포 $Z \sim N(0,1^2)$.
$$ f(z) = ce^{-z^2/2} $$
c는 normalize constant로 적분을 통해 c값을 구해보자
$$ \int_{-\inf}^{\inf} e^{-z^2/2}dz $$
closed-form으로 적분식 풀이가 불가능
$$ \int_{-\inf}^{\inf} e^{-x^2/2}dx\int_{-\inf}^{\inf} e^{-y^2/2}dy \\=\int_{-\inf}^{\inf}\int_{-\inf}^{\inf} e^{-(x^2+y^2)/2}dxdy $$
위처럼 같은식을 2번 곱하는 식으로 바꾸고 전개를 시작
위 식을 $\theta, r$ 을 사용하는 극좌표계로 바꾸면 (r은 jacobian)
$$ \int_{0}^{2\pi}\int_{0}^{\inf} e^{-r^2/2}rdrd\theta \\ =\int_{0}^{2\pi}\int_{0}^{\inf} e^{-u}dud\theta \\ = \int_{0}^{2\pi}1d\theta =2\pi $$
원래 구하려던 적분식을 두번 곱한식의 결과이기 때문에 root를 씌우면 $\sqrt{2\pi}$이므로 정규화 상수 c는 $\frac{1}{\sqrt{2 \pi}}$
즉 PDF는
$$ f(z) = ce^{-z^2/2} = \frac{1}{\sqrt{2 \pi}} e^{-z^2/2} $$
표준정규분포를 따르는 확률변수 Z의 Expected value는 대칭성을 사용하여 쉽게 구할 수 있다
$$ E(X) = \frac{1}{\sqrt{2\pi}}\int_{-\inf}^{inf}ze^{-z^2/2}dz = 0 $$
integral 내부의 수식이 기함수(odd function)이기 때문에 적분값은 0이 된다.
Variance를 구하기 위해 아래 식을 먼저 풀어야 하는데 부분적분을 사용하여 식을 정리하면 1의 값이 나온다
$$ E(X^2) = \frac{1}{\sqrt{2\pi}}\int_{-\inf}^{inf}z^2e^{-z^2/2}dz = \frac{2}{\sqrt{2\pi}}\int_{0}^{inf}(z)(ze^{-z^2/2})dz = \frac{2}{\sqrt{2\pi}}[uv|^{\inf}_{0}\int{0}^{\inf}e^{-z^2/2}dz] = 1 $$
분산은 1 - 0 으로 1이다
Reference
[0] https://www.youtube.com/playlist?list=PL2SOU6wwxB0uwwH80KTQ6ht66KWxbzTIo
'Statistics & Math > Statistics 110: Probability' 카테고리의 다른 글
[Statistics 110] Lecture 15: Midterm Review (0) | 2022.12.17 |
---|---|
[Statistics 110] Lecture 14: Location, Scale, and LOTUS (0) | 2022.12.17 |
[Statistics 110] Lecture 12: Discrete vs. Continuous, the Uniform (0) | 2022.12.10 |
[Statistics 110] Lecture 11: The Poisson distribution (0) | 2022.12.10 |
[Statistics 110] Lecture 10: Expectation Continued (0) | 2022.12.10 |