๐Ÿ–ฅ Computer Science/ํ™•๋ฅ ๊ณผ ํ†ต๊ณ„

๊ฐ€๋Šฅ๋„(Likehood) ๊ฐ€๋Šฅ๋„๋Š” ์šฐ๋„๋ผ๊ณ ๋„ ๋ถˆ๋ฆฌ๋ฉฐ, ์–ด๋– ํ•œ ๊ฐ’์ด ๊ด€์ธก๋˜์—ˆ์„ ๋•Œ, ์ด ๊ฐ’์ด ์–ด๋–ค ํ™•๋ฅ  ๋ถ„ํฌ๋กœ๋ถ€ํ„ฐ ์™”์„์ง€์— ๋Œ€ํ•œ ํ™•๋ฅ ์„ ๋‚˜ํƒ€๋‚ด๋Š” ๊ฐ’์ž…๋‹ˆ๋‹ค. ๊ฐ€๋Šฅ๋„ ํ•จ์ˆ˜ (Likedhood Function) n๊ฐœ์˜ ์ž„์˜์˜ ํ‘œ๋ณธ $X_1, X_2, ..., X_n$์— ๋Œ€ํ•œ ๊ด€์ธก๋œ ๊ฐ’๋“ค์˜ ๋ฒกํ„ฐ $x = (x_1,x_2, ... x_n)$ [X1 = x1, X2 = x2, ...]์— ๋Œ€ํ•˜์—ฌ, ์•„๋ž˜ ํ•จ์ˆ˜๊ฐ€ θ์— ๋Œ€ํ•œ ๊ฒฐํ•ฉ๋ถ„ํฌ(joint distribution)์˜ ํ•จ์ˆ˜๋กœ ๊ฐ„์ฃผ๋  ๋•Œ, ์ด๋ฅผ ๊ฐ€๋Šฅ๋„ ํ•จ์ˆ˜(likehood function)(ํ˜น์€ ์šฐ๋„ ํ•จ์ˆ˜๋ผ๊ณ ๋„ ํ•จ)๋ผ๊ณ  ํ•ฉ๋‹ˆ๋‹ค. $$f_n(x|\theta)$$ ์ด๋•Œ x๋Š” ๋ฒกํ„ฐ( $x=(x_1, x_2, ..., x_n)$ )์ด๋ฉฐ, ๊ฐ๊ฐ์˜ x1, x2, ..., xn๋“ค์„ ๋™..
๋ฐ์ดํ„ฐ๋ฅผ ๊ด€์ฐฐํ•˜๊ธฐ ์ „ ๋ชจ์ˆ˜์˜ ๋ถ„ํฌ๋ฅผ ์‚ฌ์ „ ๋ถ„ํฌ(prior distribution)๋ผ๊ณ  ํ•ฉ๋‹ˆ๋‹ค. ๋ฐ์ดํ„ฐ๊ฐ€ ๊ด€์ฐฐ๋œ ํ›„ ์ฃผ์–ด์ง„ ๋ชจ์ˆ˜์— ๋Œ€ํ•œ ์กฐ๊ฑด๋ถ€ ๋ถ„ํฌ(conditional distribution)๋ฅผ ์‚ฌํ›„ ๋ถ„ํฌ(posterior distributuion)๋ผ๊ณ  ํ•ฉ๋‹ˆ๋‹ค. ์ฆ‰ ๋‹ค์Œ๊ณผ ๊ฐ™์Šต๋‹ˆ๋‹ค. $$prior \;\; distribution \;\; \to \;\; f(\theta )$$ $$posterior \;\;distribution \;\; \to \;\; f(\theta \;| \; x_1, x_2, ... \;)$$ Prior Distribution(์‚ฌ์ „ ๋ถ„ํฌ) ๋ชจ์ˆ˜ θ๋ฅผ ๊ฐ€์ง„ ํ†ต๊ณ„์  ๋ชจ๋ธ์ด ์žˆ์„ ๋•Œ, θ๋ฅผ ํ™•๋ฅ ๋ณ€์ˆ˜๋กœ ์ทจ๊ธ‰ํ•˜๊ณ  ๋ฐ์ดํ„ฐ๋ฅผ ๊ด€์ฐฐํ•˜๊ธฐ ์ „์— θ์— ํ• ๋‹นํ•˜๋Š” ๋ถ„ํฌ๋ฅผ ์‚ฌ์ „ ๋ถ„ํฌ๋ผ ํ•ฉ๋‹ˆ๋‹ค. ์ด๋•Œ ๋ชจ์ˆ˜ ๊ณต..
ํ†ต๊ณ„์  ์ถ”์ •(Statistical Inference) ๋ชจ์ง‘๋‹จ์ด ๋งค์šฐ ํฌ๋‹ค๋ฉด ํ•ด๋‹น ์ง‘๋‹จ์˜ ํ‰๊ท , ๋ถ„์‚ฐ, ํ‘œ์ค€ํŽธ์ฐจ, ์ƒ๊ด€๊ณ„์ˆ˜ ๋“ฑ์˜ ํŠน์„ฑ๊ฐ’์„ ๊ตฌํ•˜๊ธฐ๊ฐ€ ๋งค์šฐ ์–ด๋ ต์Šต๋‹ˆ๋‹ค. ์ด๋Ÿฌํ•œ ๊ฒฝ์šฐ ๋ชจ์ง‘๋‹จ์œผ๋กœ๋ถ€ํ„ฐ ์ ๋‹นํ•œ ์ˆ˜์˜ ํ‘œ๋ณธ์„ ์ถ”์ถœํ•˜์—ฌ ํ•ด๋‹น ์ง‘๋‹จ์— ๋Œ€ํ•œ ๋ชจ์ˆ˜๋ฅผ ์ถ”์ •ํ•˜๋Š” ๊ฒƒ์„ ํ†ต๊ณ„์  ์ถ”์ •์ด๋ผ ํ•ฉ๋‹ˆ๋‹ค. ์ด๋•Œ ํ‘œ๋ณธ์€ ์ด๋ฏธ ๊ด€์ฐฐ๋œ ๊ฐ’์ด๋ฉฐ, ๋ชจ์ˆ˜์˜ ๊ฐ’์„ ์–ป๋Š” ๊ณผ์ •์ด๊ธฐ์— ๋ชจ์ˆ˜๋ฅผ ์ผ์ข…์˜ Random Variable๋กœ ์ƒ๊ฐํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์ถ”์ถœ๋œ ํ‘œ๋ณธ X1, ... Xn์€ ๋ชจ์ˆ˜๊ฐ€ ์ฃผ์–ด์กŒ์„ ๋•Œ์˜ Conditional distribution์œผ๋กœ ํ‘œํ˜„ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋ชจ์ˆ˜(Parameter) ๋ชจ์ˆ˜๋Š”๋ชจ์ง‘๋‹จ์˜ ํŠน์„ฑ(ํ‰๊ท , ๋ถ„์‚ฐ, ํ‘œ์ค€ํŽธ์ฐจ, ์ƒ๊ด€๊ณ„์ˆ˜ ๋“ฑ)์„ ๋‚˜ํƒ€๋‚ด๋Š” ๊ฐ’์œผ๋กœ, ํ•ด๋‹น ๊ฐ’๋“ค์€ ๋ชจ์ง‘๋‹จ์„ ์ „์ˆ˜์กฐ์‚ฌ ํ•ด์•ผ๋งŒ ์•Œ ์ˆ˜ ์žˆ๋Š” ๊ฐ’์ž…๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ..
์ค‘์‹ฌ๊ทนํ•œ์ •๋ฆฌ ๋ชจ์ง‘๋‹จ(Population)์ด ํ‰๊ท  $\mu$, ๋ถ„์‚ฐ $\sigma^2$ ์ธ ์ž„์˜์˜ ๋ถ„ํฌ๋ฅผ ์ด๋ฃฐ ๋•Œ, ์ด ๋ชจ์ง‘๋‹จ์œผ๋กœ๋ถ€ํ„ฐ ์ถ”์ถœ๋œ ํ‘œ๋ณธ๋“ค์— ๋Œ€ํ•ด์„œ ํ‘œ๋ณธ์˜ ํฌ๊ธฐ n์ด ์ถฉ๋ถ„ํžˆ ํฌ๋‹ค๋ฉด ํ‘œ๋ณธํ‰๊ท ์€ ๋‹ค์Œ ์ •๊ทœ ๋ถ„ํฌ์— ์ˆ˜๋ ดํ•œ๋‹ค๋Š” ์ •๋ฆฌ์ž…๋‹ˆ๋‹ค. $$\overline{X_n} \; \sim \; N(\;\mu, \; \frac{\sigma^{2}}{n}\; )$$ ๋ฆฐ๋ฐ๋ฒ ๋ฅด๊ทธ-๋ ˆ๋น„(Lindeberg and Levy) ์ค‘์‹ฌ๊ทนํ•œ์ •๋ฆฌ ๋‹ค์Œ๊ณผ ๊ฐ™์ด ์ •์˜๋ฉ๋‹ˆ๋‹ค. ํ‰๊ท  $\mu$, ๋ถ„์‚ฐ $\sigma^2$ ๋ฅผ ๊ฐ€์ง€๋Š” ๋ชจ์ง‘๋‹จ์œผ๋กœ๋ถ€ํ„ฐ ์ž„์˜์ถ”์ถœํ•œ n๊ฐœ์˜ ํ‘œ๋ณธ๋“ค์— ์žˆ์„ ๋•Œ, ํ‘œ๋ณธํ‰๊ท ์€ ๋ชจ๋“  ์‹ค์ˆ˜ x์— ๋Œ€ํ•ด ๋‹ค์Œ์ด ์„ฑ๋ฆฝํ•ฉ๋‹ˆ๋‹ค. $$\displaystyle \lim_{ n\to \infty} P\left ( \frac{\overlin..
ํฐ ์ˆ˜์˜ ๋ฒ•์น™(Law of Large Numbers)์„ ์•Œ์•„๋ณด๊ธฐ ์ „์—, ํ™•๋ฅ ์˜ ์ˆ˜๋ ด(Convergence in Probability)์— ๋Œ€ํ•˜์—ฌ ์•Œ์•„๋ณด๊ฒ ์Šต๋‹ˆ๋‹ค. ํ™•๋ฅ ์˜ ์ˆ˜๋ ด ํ™•๋ฅ ๋ณ€์ˆ˜์˜ ์—ด(sequence) Z1, Z2, ... Zn์ด ๋ชจ๋“  0๋ณด๋‹ค ํฐ ε์— ๋Œ€ํ•˜์—ฌ ๋‹ค์Œ์„ ๋งŒ์กฑํ•  ๋•Œ, $$\displaystyle \lim_{ n\to \infty} P(\;|Z_n - b|< \epsilon \;) = 1$$ ํ™•๋ฅ ๋ณ€์ˆ˜์˜ ์—ด(sequence) Z1, Z2, ... Zn์€ b๋กœ ์ˆ˜๋ ดํ•œ๋‹ค๊ณ  ์ •์˜ํ•˜๋ฉฐ, ๋‹ค์Œ๊ณผ ๊ฐ™์ด ํ‘œ์‹œํ•ฉ๋‹ˆ๋‹ค. $$Z_n \overset{p}{\rightarrow}b$$ ํฐ ์ˆ˜์˜ ๋ฒ•์น™(Law of Large Numbers) ํฐ ์ˆ˜์˜ ๋ฒ•์น™์€ ์ž„์˜์˜ ํ‘œ๋ณธ๋“ค์— ๋Œ€ํ•˜์—ฌ, ํ‘œ๋ณธ์ถ”์ถœํ•œ ํ‘œ๋ณธ๋“ค์˜ ๊ฐœ์ˆ˜๊ฐ€ ์ถฉ๋ถ„ํžˆ ํฌ๋‹ค..
Markov inequality์™€ Chebyshev inequality ๋ชจ๋‘ ๋ชจ์ง‘๋‹จ์˜ ๋ถ„ํฌ๋ฅผ ๋ชจ๋ฅผ ๋•Œ ์‚ฌ์šฉ ๊ฐ€๋Šฅํ•œ ๋ถ€๋“ฑ์‹์ž…๋‹ˆ๋‹ค. Markov’s inequality (๋งˆ๋ฅด์ฝ”๋ธŒ ๋ถ€๋“ฑ์‹) Markov's inequality์€ ์Œ์ด ์•„๋‹Œ ํ™•๋ฅ  ๋ณ€์ˆ˜๊ฐ€ ์–ด๋–ค ์–‘์˜ ์‹ค์ˆ˜ ์ด์ƒ์ผ ํ™•๋ฅ ์˜ ์ƒ๊ณ„๋ฅผ ๋‚˜ํƒ€๋‚ด๋Š” ๋ถ€๋“ฑ์‹์ž…๋‹ˆ๋‹ค. ํ™•๋ฅ ๊ณผ ๊ธฐ๋Œ“๊ฐ’์˜ ๊ด€๊ณ„๋ฅผ ์„ค๋ช…ํ•˜๊ณ , ํ™•๋ฅ  ๋ณ€์ˆ˜์˜ c.d.f์— ๋Œ€ํ•ด ๋Š์Šจํ•˜์ง€๋งŒ ์œ ์šฉํ•œ ํ•œ๊ณ„๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. Markov's inequality์€ ๋‹ค์Œ๊ณผ ๊ฐ™์Šต๋‹ˆ๋‹ค. P(X > 0) = 1 ์„ ๋งŒ์กฑ์‹œํ‚ค๋Š” ํ™•๋ฅ ๋ณ€์ˆ˜ X์— ๋Œ€ํ•˜์—ฌ ๋‹ค์Œ ๋ถ€๋“ฑ์‹์ด ์„ฑ๋ฆฝํ•ฉ๋‹ˆ๋‹ค. $$ P(X \geq t) \leq \frac{E(X)}{t} ,\;\;\;\forall{real \;\;t>0}$$ ์ด๋Š” ๋ถ„์‚ฐ์ด ์—†์–ด๋„ ํ‰๊ท ์œผ๋กœ๋งŒ ์œ ๋„ ๊ฐ€๋Šฅํ•œ ๋ถ€..
ํ‘œ๋ณธํ‰๊ท (Sample Mean) ๋ชจ์ง‘๋‹จ์—์„œ ๋…๋ฆฝ์ (๋ฌด์ž‘์œ„๋กœ)์œผ๋กœ n๊ฐœ์˜ ํ‘œ๋ณธ๋“ค์„ (๋ณต์›)์ถ”์ถœํ•˜์˜€์„ ๋•Œ ํ•ด๋‹น ํ‘œ๋ณธ๋“ค์˜ ํ‰๊ท ์„ ๊ตฌํ•œ ๊ฐ’์„ ์˜๋ฏธํ•ฉ๋‹ˆ๋‹ค. $$\overline{X_n} = \frac{1}{n}\sum^{n}_{i=1}X_i$$ ํ‘œ๋ณธํ‰๊ท ์˜ ํ‰๊ท ๊ณผ ๋ถ„์‚ฐ ๋ชจ์ง‘๋‹จ์˜ ํ‰๊ท ์ด μ, ๋ถ„์‚ฐ์ด σ^2์ผ ๋•Œ, ํ‘œ๋ณธํ‰๊ท ์˜ ํ‰๊ท ๊ณผ ๋ถ„์‚ฐ์€ ๋ชจ์ง‘๋‹จ์˜ ๋ถ„ํฌ์™€ ์ƒ๊ด€์—†์ด ๋‹ค์Œ๊ณผ ๊ฐ™์Šต๋‹ˆ๋‹ค. ํ‘œ๋ณธํ‰๊ท ์˜ ํ‰๊ท  : (๋ชจ์ง‘๋‹จ์˜ ํ‰๊ท ๊ณผ ๋™์ผํ•ฉ๋‹ˆ๋‹ค)$$\mu$$ ํ‘œ๋ณธํ‰๊ท ์˜ ๋ถ„์‚ฐ: $$\frac{\sigma^{2}}{n}$$ ์ฆ๋ช…) ํ‘œ๋ณธํ‰๊ท ์€ ๋‹ค์Œ๊ณผ ๊ฐ™์œผ๋ฏ€๋กœ $$\overline{X_n} = \frac{1}{n}\sum^{n}_{i=1}X_i$$ ํ‘œ๋ณธํ‰๊ท ์˜ ํ‰๊ท ์€ ๋‹ค์Œ๊ณผ ๊ฐ™์Šต๋‹ˆ๋‹ค. $$E(\overline{X_n})=\frac{1}{n}\..
์ •๊ทœ๋ถ„ํฌ (Normal Distribution) ์ •๊ทœ๋ถ„ํฌ๋Š” ์—ฐ์† ํ™•๋ฅ  ๋ถ„ํฌ์— ์†ํ•˜๋ฉฐ ์•„๋ž˜์™€ ๊ฐ™์ด ์ •์˜๋ฉ๋‹ˆ๋‹ค. ํ™•๋ฅ ๋ณ€์ˆ˜ X๊ฐ€ ๋‹ค์Œ p.d.f๋ฅผ ๊ฐ–๋Š” ์—ฐ์† ๋ถ„ํฌ๋ฅผ ๊ฐ–๋Š” ๊ฒฝ์šฐ, ํ‰๊ท (mean) μ ์™€ ๋ถ„์‚ฐ(variance) σ^2์„ ๊ฐ€์ง€๋Š” ์ •๊ทœ๋ถ„ํฌ๋ฅผ ๋”ฐ๋ฆ…๋‹ˆ๋‹ค. $$p.d.f\;\;\; f(\;x\;|\; \mu,\; \sigma^{2}) \;= \;\;\;\;\frac{1}{\sqrt{2\pi}\sigma} \cdot exp(-\frac{(x-\mu)^{2}}{2\sigma^{2}})\;\;\;\;\;\;\;for\; -\infty < x < \infty $$ ๊ทธ๋ฆฌ๊ณ  ์ •๊ทœ๋ถ„ํฌ๋Š” ๋‹ค์Œ๊ณผ ๊ฐ™์Šต๋‹ˆ๋‹ค. $$X \sim N(\mu, \sigma^{2})$$ (์ด๋•Œ ํ‰๊ท  μ๋Š” bounded(-∞
๋ง ๋ž‘
'๐Ÿ–ฅ Computer Science/ํ™•๋ฅ ๊ณผ ํ†ต๊ณ„' ์นดํ…Œ๊ณ ๋ฆฌ์˜ ๊ธ€ ๋ชฉ๋ก (2 Page)