$$ \newcommand{\floor}[1]{\left\lfloor #1 \right\rfloor} \newcommand{\ceil}[1]{\left\lceil #1 \right\rceil} \newcommand{\N}{\mathbb{N}} \newcommand{\R}{\mathbb{R}} \newcommand{\Z}{\mathbb{Z}} \newcommand{\Q}{\mathbb{Q}} \newcommand{\C}{\mathbb{C}} \renewcommand{\L}{\mathcal{L}} \newcommand{\x}{\times} \newcommand{\contra}{\scalebox{1.5}{$\lightning$}} \newcommand{\inner}[2]{\left\langle #1 , #2 \right\rangle} \newcommand{\st}{\text{ such that }} \newcommand{\for}{\text{ for }} \newcommand{\Setcond}[2]{ \left\{\, #1 \mid #2 \, \right\}} \newcommand{\setcond}[2]{\Setcond{#1}{#2}} \newcommand{\seq}[1]{ \left\langle #1 \right\rangle} \newcommand{\Set}[1]{ \left\{ #1 \right\}} \newcommand{\set}[1]{ \set{#1} } \newcommand{\sgn}{\text{sign}} \newcommand{\halfline}{\vspace{0.5em}} \newcommand{\diag}{\text{diag}} \newcommand{\legn}[2]{\left(\frac{#1}{#2}\right)} \newcommand{\ord}{\text{ord}} \newcommand{\di}{\mathrel{|}} \newcommand{\gen}[1] \newcommand{\irr}{\mathrm{irr }} \renewcommand{\deg}{\mathrm{deg }} \newcommand{\nsgeq}{\trianglelefteq} \newcommand{\nsg}{\triangleleft} \newcommand{\argmin}{\mathrm{argmin}} \newcommand{\argmax}{\mathrm{argmax}} \newcommand{\minimize}{\mathrm{minimize}} \newcommand{\maximize}{\mathrm{maximize}} \newcommand{\subto}{\mathrm{subject\ to}} \newcommand{\DKL}[2]{D_{\mathrm{KL}}\left(#1 \di\di #2\right)} \newcommand{\ReLU}{\mathrm{ReLU}} \newcommand{\E}{\mathbb{E}} \newcommand{\expect}[1]{\E\left[#1\right]} \newcommand{\expectwith}[2]{\E_{#1}\left[#2\right]} \renewcommand{\P}{\mathbb{P}} \newcommand{\uniform}[2]{\mathrm{Uniform}\left(#1 \dots #2\right)} \newcommand{\gdist}[2]{\mathcal{N}\left(#1, #2\right)} $$

Deep Learning

Deep Learning

Total views

What is This?

์ด ์นดํ…Œ๊ณ ๋ฆฌ์˜ ๊ธ€๋“ค์€ ํฌ๊ฒŒ ์•„๋ž˜ ์„ธ ๊ฐ€์ง€์— ๊ธฐ๋ฐ˜ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

์ œ๊ฐ€ ์ •๋ฆฌํ•œ ๋‚ด์šฉ์€ ๊ณผ๋ชฉ์˜ officialํ•œ ๊ฐ•์˜์ž๋ฃŒ์™€ ์•„๋ฌด๋Ÿฐ ์ƒ๊ด€์ด ์—†์œผ๋ฉฐ, ์ œ๊ฐ€ ์ด๋ฏธ ์ž˜ ์•Œ๊ณ  ์žˆ๋Š” ๋‚ด์šฉ์€ ์Šคํ‚ตํ•œ ๊ฒƒ๋„ ๋งŽ๊ณ , ๋‹ค๋ฅธ ์ž๋ฃŒ๋ฅผ ์ฐพ์•„๋ณด๋ฉฐ ์ถ”๊ฐ€ํ•œ ๋‚ด์šฉ๋„ ๋งŽ์Šต๋‹ˆ๋‹ค. ์ฆ‰, ์—ฌ๊ธฐ ์žˆ๋Š” ํฌ์ŠคํŒ…์€ ๊ณผ๋ชฉ์—์„œ ์‹ค์ œ๋กœ ๊ฐ•์˜๋œ ๋‚ด์šฉ๊ณผ ์ผ์น˜ํ•˜์ง€ ์•Š์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ํŠนํžˆ, ์˜คํƒ€๋‚˜ ์˜ค๋ฅ˜๊ฐ€ ์žˆ๋‹ค๋ฉด ๋งค์šฐ ๋†’์€ ํ™•๋ฅ ๋กœ ์ œ๊ฐ€ ์ดํ•ดํ•œ ๊ฒƒ์„ ์ ๋‹ค๊ฐ€ ์‹ค์ˆ˜ํ•˜์˜€๊ฑฐ๋‚˜ ์ž˜๋ชป ์ดํ•ดํ•˜๊ณ  ์žˆ์„ ๊ฐ€๋Šฅ์„ฑ์ด ๋†’์Šต๋‹ˆ๋‹ค.

  1. Goodfellow, Bengio, Courville์˜ ์ €์„œ Deep Learning ์˜ ๋‚ด์šฉ์„ ๊ณต๋ถ€ํ•˜๊ณ  ์ •๋ฆฌํ•œ ๋‚ด์šฉ
  2. 2021 ๊ฐ€์„ํ•™๊ธฐ์— ์ˆ˜๊ฐ•ํ•˜๋Š” ์ˆ˜์—… ํ•ด์„ํ•™ํŠน๊ฐ• : ์‹ฌ์ธตํ•™์Šต์˜ ์ˆ˜์น˜ํ•ด์„ ์—์„œ ๊ณต๋ถ€ํ•œ ๋‚ด์šฉ (1์˜ ์ฑ…์„ ๊ต์žฌ๋กœ ์‚ฌ์šฉํ•˜๋Š” ์ˆ˜์—…์ž…๋‹ˆ๋‹ค)
  3. 2021 ๊ฐ€์„ํ•™๊ธฐ์— ์ˆ˜๊ฐ•ํ•˜๋Š” ์ˆ˜์—… ์‹ฌ์ธต ์‹ ๊ฒฝ๋ง์˜ ์ˆ˜ํ•™์  ๊ธฐ์ดˆ ์—์„œ ๊ณต๋ถ€ํ•œ ๋‚ด์šฉ (2์— ๋น„ํ•ด ์ˆ˜ํ•™์ ์ธ ๋ฉด์— ์ง‘์ค‘ํ•˜๋Š” ์ˆ˜์—…์ž…๋‹ˆ๋‹ค)

๊ฐ•์˜์™€ ์ฑ…์˜ ๋‚ด์šฉ์„ ์ œ๊ฐ€ ์ ๋‹นํžˆ ์ œ ๊ณต๋ถ€ํ•œ ๋ฐฉํ–ฅ์— ๋งž๊ฒŒ ์žฌ๊ตฌ์„ฑํ•œ ๋ถ€๋ถ„๋“ค์ด ๋งŽ์•„, ์ˆœ์„œ๊ฐ€ ํŠน์ • ์ž๋ฃŒ๋ฅผ ๋”ฐ๋ฅด์ง€ ์•Š์Šต๋‹ˆ๋‹ค.

์žฌ๋ฐŒ๊ฒŒ๋„ ์ œ๊ฐ€ ์ด ๋ธ”๋กœ๊ทธ์— ์“ฐ๊ธฐ ์‹œ์ž‘ํ•œ ๊ธ€์€ -์Šต๋‹ˆ๋‹ค๋กœ, ํ•„๊ธฐ๋…ธํŠธ๋ฅผ $\LaTeX$๋กœ ์ •๋ฆฌํ•˜๊ณ  ๊ทธ๊ฑธ ๋‹ค์‹œ pandoc์œผ๋กœ ๋งˆํฌ๋‹ค์šด ๋ณ€ํ™˜ํ•œ ํฌ์ŠคํŒ…๋“ค์€ -๋‹ค๋กœ ๋๋‚ฉ๋‹ˆ๋‹ค. ์ด๊ฑธ ๋ชจ๋‘ ์–ธ์  ๊ฐ€๋Š” -์Šต๋‹ˆ๋‹ค๋กœ ํ†ต์ผํ•˜๊ฒ ๋‹ค๋Š” ์›๋Œ€ํ•œ ๊ฟˆ์ด ์žˆ์ง€๋งŒ ๋‹น๋ถ„๊ฐ„์€ ์‰ฝ์ง€ ์•Š์„ ์˜ˆ์ •์ž…๋‹ˆ๋‹ค.

์ฐธ๊ณ  3๋ฒˆ ์ˆ˜์—…์˜ ๊ฒฝ์šฐ $\LaTeX$ ๋กœ ํ•„๊ธฐ๋…ธํŠธ๋ฅผ ๊ด€๋ฆฌํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์–ธ์  ๊ฐ€ ์—ฌ๊ฑด์ด ํ—ˆ๋ฝํ•œ๋‹ค๋ฉด ์ด ๋…ธํŠธ๋ฅผ ์—ฌ๊ธฐ์— ์˜ฌ๋ฆดํ…๋ฐ, pandoc์œผ๋กœ ์“ด ํฌ์ŠคํŒ…์€ ๊ทธ์ชฝ์ด ์ข€๋” ๊น”๋”ํ•˜๊ฒŒ ๋ณด์ผ ๊ฒƒ ๊ฐ™์Šต๋‹ˆ๋‹ค. ์ตœ์ข…์ ์œผ๋กœ๋Š” ๋งค์šฐ ์œ ๋ช…ํ•œ Evan Chen Notes ๊ฐ™์€ ๊ฒฐ๊ณผ๋ฌผ์„ ๋ชฉํ‘œํ•˜๊ณ  ์žˆ๋Š”๋ฐ, ๋‹ค์–‘ํ•œ ์ด์Šˆ๋“ค์ด ์žˆ์Šต๋‹ˆ๋‹ค. ๊ฐ€์žฅ ๋จผ์ €, ์ด ์ž๋ฃŒ๋ฅผ ๊ณต๊ฐœํ•ด๋„ ๋˜๋Š”์ง€์˜ ๋ฌธ์ œ๊ฐ€ ์žˆ๋Š”๋ฐ, ์›๋ณธ ๊ฐ•์˜์ž๋ฃŒ๊ฐ€ ๊ต์ˆ˜๋‹˜์˜ ๊ฐœ์ธ ์›น์‚ฌ์ดํŠธ ์— ์–ด๋–ค ๋กœ๊ทธ์ธ์ด๋‚˜ ํ•™๊ต ๊ณ„์ • ์š”๊ตฌ ์—†์ด ์˜ฌ๋ผ์™€ ์žˆ์œผ๋ฏ€๋กœ (์•„๋งˆ๋„) ๊ดœ์ฐฎ์„ ๊ฒƒ์œผ๋กœ ์ƒ๊ฐํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ํŠนํžˆ ๋งˆํฌ๋‹ค์šด ๋ฒ„์ „์€ ์ œ edit์ด ์ข€ ํ—ค๋น„ํ•˜๊ฒŒ ๋“ค์–ด๊ฐˆ ๊ฑฐ๋ผ์„œ ์•„๋งˆ ํฐ ๋ฌธ์ œ๋Š” ์—†๋Š” ๊ฒƒ์œผ๋กœ ์•Œ๊ณ  ์žˆ๋Š”๋ฐ, $\LaTeX$ scribe note๋Š” ๊ฐ•์˜ ํ•„๊ธฐํ•œ ๊ฑฐ๋ผ์„œ ๊ดœ์ฐฎ์€์ง€ ์‚ฌ์‹ค ์ž˜ ๋ชจ๋ฅด๊ฒ ์Šต๋‹ˆ๋‹ค.


Postings

Basics / Theory

MLP, CNN ๋“ฑ ๋ชจ๋ธ์— ๋Œ€ํ•œ ์–˜๊ธฐ๊ฐ€ ์•„๋‹Œ, ์ „์ฒด์ ์ธ ์ด๋ก ์— ๋Œ€ํ•œ ์ด์•ผ๊ธฐ

Shallow Neural Networks

Multi Layer Perceptron

Convolutionary Neural networks

ImageNet Challenge์˜ ์—ญ์‚ฌ๋ฅผ ๋”ฐ๋ผ๊ฐ€๋ฉฐ, ๋ช‡๊ฐ€์ง€ ์„ฑ๊ณต์ ์ธ Image classification ๋ชจ๋ธ๋“ค์— ๋Œ€ํ•ด ๊ณต๋ถ€ํ•ฉ๋‹ˆ๋‹ค.
CIFAR10์—์„œ์˜ ๊ฒฐ๊ณผ ์ •๋ฆฌ

Computer Vision

  • Semantic Segmentation ๊ฐœ์š” : Semantic segmentation ๋ฌธ์ œ ์ •์˜, ๊ฐœ์š”
  • [Fully Convolutional Networks]
  • [Encoder-Decoder, U-Net]
  • [Dilated Convolutions, DeepLab]