ํ–‰๋ ฌ์˜ โ€œํฌ๊ธฐโ€๋‚˜ โ€œ๊ธธ์ดโ€๋ฅผ ์ˆ˜์น˜์ ์œผ๋กœ ๋‚˜ํƒ€๋‚ด๋Š” ์ฒ™๋„

๋งŒ์กฑํ•ด์•ผ ํ•˜๋Š” ์„ฑ์งˆ

  • ๋น„์Œ์ˆ˜์„ฑ (Non-negativity)
    • ์ด๊ณ ,ย  ์ด๋ฉดย 
  • ๋™์ฐจ์„ฑ (Homogeneity)
    • ์Šค์นผ๋ผย  ์— ๋Œ€ํ•ดย 
  • ์‚ผ๊ฐ ๋ถ€๋“ฑ์‹ (Triangle Inequality)

Induced Norm

ํ–‰๋ ฌ์ด ๋ฒกํ„ฐ์— ์ž‘์šฉํ•  ๋•Œ ๊ทธ ๋ฒกํ„ฐ์˜ ํฌ๊ธฐ๊ฐ€ ์–ผ๋งˆ๋‚˜ ๋ณ€ํ™”ํ•˜๋Š”์ง€ ์ธก์ •ํ•˜๋Š” ๋…ธ๋ฆ„

L1 Norm

์—ด์˜ ์ ˆ๋Œ€๊ฐ’์˜ ํ•ฉ ์ค‘ ๊ฐ€์žฅ ํฐ ๊ฐ’

L2 Norm

ํ–‰๋ ฌ์˜ Eigenvalue ์ค‘ ๊ฐ€์žฅ ํฐ ๊ฐ’

  • ์ŠคํŽ™ํŠธ๋Ÿผ ๋…ธ๋ฆ„(Spectral Norm)์ด๋ผ๊ณ ๋„ ํ•œ๋‹ค.

Infinity Norm

ํ–‰์˜ ์ ˆ๋Œ€๊ฐ’์˜ ํ•ฉ ์ค‘ ๊ฐ€์žฅ ํฐ ๊ฐ’

Frobenius Norm

ํ–‰๋ ฌ์˜ ๊ฐ ์›์†Œ์˜ ์ œ๊ณฑ์˜ ํ•ฉ์˜ ์ œ๊ณฑ๊ทผ

Max Norm

ํ–‰๋ ฌ์˜ ์›์†Œ ์ค‘ ๊ฐ€์žฅ ํฐ ๊ฐ’