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 |
- WBP
- WinAPI
- multimodal
- userwidget
- Unreal Engine
- dl
- ddpm
- motion matching
- 생성모델
- Font Generation
- UE5
- Few-shot generation
- 오블완
- cv
- animation retargeting
- RNN
- ue5.4
- Stat110
- NLP
- 언리얼엔진
- 디퓨전모델
- CNN
- 딥러닝
- Diffusion
- Generative Model
- deep learning
- BERT
- 모션매칭
- GAN
- 폰트생성
- Today
- Total
Deeper Learning
[Statistics 110] Lecture 30: Chi-Square, Student-t, Multivariate Normal 본문
[Statistics 110] Lecture 30: Chi-Square, Student-t, Multivariate Normal
Dlaiml 2023. 1. 28. 16:41Chi-Square Distribution(카이제곱 분포)
$V = Z_1^2+...+Z_n^2, \space Z_j \sim N(0,1){i.i.d}$일 때, $V \sim \chi^2{(n)}$.
$\chi^2_{(1)}$은 $Gamma(\frac12,\frac12)$와 같은 분포, $\chi^2$는 $Gamma(\frac n 2,\frac 1 2)$와 같은 분포
t-Distribution(t-분포)
$$ Z \sim N(0,1)\\V\sim \chi ^2_{(n)}\\T = \frac{Z}{\sqrt{V/n}} \sim t_n $$
성질
- Symmetric: $-T \sim t_n$.
- n=1이면 코시분포를 따름
- n이 2이상이면 $E(T) = 0$.
- 정규분포와 비슷하지만 더 heavy tailed(극단적인 값이 더 발생)
- n이 크면 표준정규분포와 매우 비슷함
$n\rightarrow \inf, t_n \rightarrow N(0,1)$을 큰 수의 법칙을 사용해서 증명해보면
$$ T = \frac{Z}{\sqrt{V/n}} \sim t_n\\ $$
에서 $\frac {V_n}n$은 1로 1의 확률로 수렴하므로 $T \rightarrow Z$도 1의 확률로 성립
Multivariate Normal(다변량 정규분포)
Random Vector $\vec X=(X_1,...,X_n)$에서 모든 t에 대한 선형 결합($t_1x_1+...+t_nx_n$)이 정규분포를 따르면, 다변량 정규분포를 따른다.
예시) $Z,W$가 i.i.d $N(0,1)$일 때 $(Z+2W,3Z+5W)$는 다변량 정규분포를 따른다.
임의의 상수 $s,t$에 대해 선형 결합을 하면
$$ s(Z+2W) + t(3Z+5W) = (s+3t)Z + (2s+5t)W \sim N $$
독립인 두 정규분포의 합도 정규분포임을 이전에 증명하였기 때문에 예시에 대한 증명완료
예시) $Z$는 표준정규분포, $S$는 random sign(-1 or 1) 서로 독립일 때, $Z$와 $SZ$는 표준정규분포를 따르지만 둘의 선형결합인 $Z+SZ$는 절반이 0인 분포로 정규분포를 따르지 않는다.
적률생성함수
다변량 정규분포를 따르는 확률변수 $\vec X$의 MGF는 $E(e^ {\vec t\cdot \vec X})$ .
$$ E(e^ {\vec t\cdot \vec X}) =E(e^ {t_1\mu_1+...+t_k\mu_k+\frac 12Var(t_1X_1+...+t_kX_k)}) $$
다변량 정규분포에서는 비상관관계(공분산이 0)이면 독립이다.
Reference
[0] https://www.youtube.com/playlist?list=PL2SOU6wwxB0uwwH80KTQ6ht66KWxbzTIo
'Statistics & Math > Statistics 110: Probability' 카테고리의 다른 글
[Statistics 110] Lecture 32: Markov Chains Continued (0) | 2023.01.28 |
---|---|
[Statistics 110] Lecture 31: Markov Chains (0) | 2023.01.28 |
[Statistics 110] Lecture 29: Law of Large Numbers and Central Limit Theorem (0) | 2023.01.28 |
[Statistics 110] Lecture 28: Inequalities (0) | 2023.01.28 |
[Statistics 110] Lecture 27: Conditional Expectation given an R.V. (0) | 2023.01.28 |