์ธ๊ณต์ง€๋Šฅ ์ •๋ฆฌ [๋ณธ๋ก 8] :: ์ธ๊ณต์‹ ๊ฒฝ๋ง ์„ค๊ณ„ ์‹œ ๊ณ ๋ ค์‚ฌํ•ญ ์ •๋ฆฌ!
์ปดํ“จํ„ฐ๊ณผํ•™ (CS)/AI 2020. 2. 29. 19:20

์ธ๊ณต์‹ ๊ฒฝ๋ง ์„ค๊ณ„ ์‹œ ๊ณ ๋ ค์‚ฌํ•ญ Network topology ๋„คํŠธ์›Œํฌ์˜ ๋ชจ์–‘ (feed forward, feed backward) Activation function ์ถœ๋ ฅ์˜ ํ˜•ํƒœ Objectives ๋ถ„๋ฅ˜? ํšŒ๊ท€? Loss function, Error๋กœ ๋‚˜ํƒ€๋‚ผ ์ˆ˜ ์žˆ์Œ Optimizers weight update Generalization Overfitting ๋ฐฉ์ง€ 2. activation function ์ถœ๋ ฅ์˜ ํ˜•ํƒœ ๊ฒฐ์ • 1. one-hot vector ์—ฌ๋Ÿฌ ๊ฐ’ ์ค‘ ํ•˜๋‚˜์˜ ๊ฐ’๋งŒ ์ถœ๋ ฅ ex_ ์ˆซ์ž ์‹๋ณ„ 2. softmax function ํ•ด๋‹น ์ถœ๋ ฅ์ด ๋‚˜์˜ฌ ํ™•๋ฅ ๋กœ ํ‘œํ˜„ 3. objective function ๊ธฐํƒ€ ๋ชฉ์ ํ•จ์ˆ˜ Mean absolute error / mae Mean absolute percentag..

์ธ๊ณต์ง€๋Šฅ ์ •๋ฆฌ [๋ณธ๋ก 7] :: ๊นŠ์–ด์ง„ ์ธ๊ณต์‹ ๊ฒฝ๋ง์˜ ๋ฌธ์ œ์ ?
์ปดํ“จํ„ฐ๊ณผํ•™ (CS)/AI 2020. 2. 29. 19:07

๊นŠ์–ด์ง„ ์ธ๊ณต์‹ ๊ฒฝ๋ง์˜ ๋ฌธ์ œ์  Generalization : training์— ์‚ฌ์šฉ๋˜์ง€ ์•Š์€ data์— ๋Œ€ํ•œ ์„ฑ๋Šฅ Data set Training set training์— ์‚ฌ์šฉํ•˜๋Š” data set Validation set ์ฃผ์–ด์ง„ data set ์ค‘ ๋นผ๋†“์•˜๋‹ค๊ฐ€ ์„ฑ๋Šฅ์„ ๊ฒ€์ฆํ•  ๋•Œ ์‚ฌ์šฉํ•˜๋Š” data set Test set ์ฃผ์–ด์ง€์ง€ ์•Š์•˜๋˜ ์ ‘ํ•œ ์  ์—†๋Š” data set ํ•™์Šต์ด training set์œผ๋กœ ์ง„ํ–‰๋˜๊ธฐ ๋•Œ๋ฌธ์—, ํ•™์Šต์„ ๋ฐ˜๋ณตํ•  ์ˆ˜๋ก training set์— ๋Œ€ํ•œ ์ •ํ™•๋„๋Š” ๋†’์•„์ง€๊ณ  ์˜ค๋ฅ˜์œจ์€ ๋‚ฎ์•„์ง„๋‹ค. ํ•˜์ง€๋งŒ validation set์— ๋Œ€ํ•œ ์˜ค๋ฅ˜์œจ์€ ๋‚ฎ์•„์ง€๋‹ค๊ฐ€ ๋†’์•„์ง€๋Š” ํ˜„์ƒ์„ ๋„๋Š”๋ฐ, ์ด๋Š” ํ•™์Šต์ด ๋„ˆ๋ฌด training set์—๋งŒ ์ ํ•ฉํ•˜๊ฒŒ ๋˜์—ˆ๊ธฐ ๋•Œ๋ฌธ์ด๋‹ค. ์ด๋ฅผ training set์— overfitting(..