post-thumbnail

7. Training Neural Networks โ…ก

์šฐ๋ฆฌ๋Š” ์ง€๋‚œ์‹œ๊ฐ„์— 6๊ฐœ์˜ activation function์„ ๋ฐฐ์› ๋‹ค. ์ด์ค‘์—์„œ Sigmoid์™€ ReLU๋งŒ ๋‹ค์‹œ ๋ด๋ณด์ž!Sigmoid๋Š” ๊ณผ๊ฑฐ์— ์œ ๋ช…ํ–ˆ์ง€๋งŒ Vanishing Gradients์˜ ๋ฌธ์ œ์  ๋•Œ๋ฌธ์— ์ด์ œ๋Š” ์ž˜ ์“ฐ์ง€ ์•Š๋Š”๋‹ค.์ด์ œ๋Š” ReLU๋ฅผ ์“ฐ๋Š” ๊ฒƒ์ด ๊ฐ€์žฅ ์ข‹์€

2022๋…„ 4์›” 28์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
post-thumbnail

6. Training Neural Networks, Part 1

http://cs231n.stanford.edu/slides/2017/cs231n_2017_lecture6.pdfhttps://inhovation97.tistory.com/23One time setupactivation functionspreproce

2022๋…„ 4์›” 8์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
post-thumbnail

6-1. ๋‹ค์ด๋‚˜๋ฏน ํ”„๋กœ๊ทธ๋ž˜๋ฐ ๊ฐœ๋… & ์‹ค์ „ ๋ฌธ์ œ

์ปดํ“จํ„ฐ๋ฅผ ํ™œ์šฉํ•ด๋„ ํ•ด๊ฒฐํ•˜๊ธฐ ์–ด๋ ค์šด ๋ฌธ์ œ์ตœ์ ์˜ ํ•ด๋ฅผ ๊ตฌํ•˜๊ธฐ์— ์‹œ๊ฐ„์ด ๋งค์šฐ ๋งŽ์ด ํ•„์š”๋ฉ”๋ชจ๋ฆฌ ๊ณต๊ฐ„์ด ๋งค์šฐ ๋งŽ์ด ํ•„์š”๐Ÿ’ก ํ•˜์ง€๋งŒ ์–ด๋–ค ๋ฌธ์ œ๋Š” ๋ฉ”๋ชจ๋ฆฌ ๊ณต๊ฐ„์„ ์•ฝ๊ฐ„ ๋” ์‚ฌ์šฉํ•˜๋ฉด ์—ฐ์‚ฐ ์†๋„๋ฅผ ๋น„์•ฝ์ ์œผ๋กœ ์ฆ๊ฐ€์‹œํ‚ฌ ์ˆ˜ ์žˆ๋‹ค.โžก๏ธ ๋‹ค์ด๋‚˜๋ฏน ํ”„๋กœ๊ทธ๋ž˜๋ฐ (๋™์  ๊ณ„ํš๋ฒ•)Q. ๋‹ค์ด๋‚˜๋ฏน ํ”„๋กœ๊ทธ๋ž˜

2022๋…„ 4์›” 6์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
post-thumbnail

5-1. ์ด์ง„ํƒ์ƒ‰ ๊ฐœ๋… & ์‹ค์ „ ๋ฌธ์ œ

์ด๋ฒˆ ์žฅ์—์„œ๋Š” ๋ฆฌ์ŠคํŠธ ๋‚ด์—์„œ ๋ฐ์ดํ„ฐ๋ฅผ ๋งค์šฐ ๋น ๋ฅด๊ฒŒ ํƒ์ƒ‰ํ•˜๋Š” ์ด์ง„ ํƒ์ƒ‰ ์•Œ๊ณ ๋ฆฌ์ฆ˜์— ๋Œ€ํ•ด ๋‹ค๋ฃฌ๋‹ค. ์ด์ง„ ํƒ์ƒ‰์„ ์•Œ์•„๋ณด๊ธฐ ์ „์— ๊ฐ€์žฅ ๊ธฐ๋ณธ ํƒ์ƒ‰ ๋ฐฉ๋ฒ•์ธ ์ˆœ์ฐจ ํƒ์ƒ‰์„ ๊ณต๋ถ€ํ•ด๋ณด์ž! ์ˆœ์ฐจ ํƒ์ƒ‰ (Sequential Search) ๋ฆฌ์ŠคํŠธ ์•ˆ์— ์žˆ๋Š” ํŠน์ •ํ•œ ๋ฐ์ดํ„ฐ๋ฅผ ์ฐพ๊ธฐ ์œ„ํ•ด

2022๋…„ 4์›” 5์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
post-thumbnail

5. Convolutional Neural Networks

1957๋…„ Frank Rosenblatt๊ฐ€ Mark I Perceptron machine์„ ๊ฐœ๋ฐœ์ตœ์ดˆ์˜ ํผ์…‰ํŠธ๋ก  ๊ธฐ๊ณ„1960๋…„ Widrow์™€ Hoff๊ฐ€ Adaline and Madaline ๊ฐœ๋ฐœ์ตœ์ดˆ์˜ Multilayer Perceptron Network1986๋…„ Rume

2022๋…„ 4์›” 4์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€

4. Backpropagation and Neural Networks

๐Ÿ“Œ ๋ฐœํ‘œ ์ž๋ฃŒ ํ‚คํฌ์ธํŠธ ๐Ÿ“Œ Review ๊ฐ„๋‹จํžˆ (Optimization, Gradient Descent) Computational graphs ํ๋ฆ„ loss function input, parameter Gradient ๊ฐ„๋žตํ•œ ์„ค๋ช… matrix

2022๋…„ 3์›” 27์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
post-thumbnail

3. Loss Functions & Optimization

์ฒซ ๋ฒˆ์งธ cat example์„ ๋ณด๋ฉด ์Šค์ฝ”์–ด๊ฐ€ cat๋ณด๋‹ค car๊ฐ€ ๋” ๋†’์€ ๊ฒƒ์„ ๋ณผ ์ˆ˜ ์žˆ๋‹ค. โžก๏ธ bad์„ธ ๋ฒˆ์งธ frog exmaple์„ ๋ณด๋ฉด ์Šค์ฝ”์–ด๊ฐ€ cat, car๋ณด๋‹ค ๋‚ฎ๊ณ  ์Œ์ˆ˜์ธ ๊ฐ’์ด ๋‚˜์˜จ ๊ฒƒ์„ ๋ณผ ์ˆ˜ ์žˆ๋‹ค. โžก๏ธ bady์— ์ด ์ผ€์ด์Šค๊ฐ€ ์–ผ๋งˆ๋‚˜ ๋‚˜์œ์ง€ ๊ฐ’์„ ์ €์žฅ

2022๋…„ 3์›” 16์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
post-thumbnail

์‹ค์Šต1-1. K-Nearest Neighbor classifier

๊ณต์‹์‚ฌ์ดํŠธ) https://cs231n.github.io/assignments2020/assignment1/https://cs231n.github.io/assignments/2021/assignment1_colab.zip์œ„์˜ ๋งํฌ๋ฅผ ํ†ตํ•ด Starter

2022๋…„ 3์›” 16์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
post-thumbnail

2. Image Classification

Image Classification์€ ์ปดํ“จํ„ฐ ๋น„์ „์˜ core taskinput image(ex. cat)pre-determined categories/labels (ex.{dog, cat, truck, palne, ...})์‚ฌ๋žŒ์—๊ฒŒ ์ด ๊ณผ์ •์€ ๋งค์šฐ ์‰ฝ์ง€๋งŒ ์ปดํ“จํ„ฐ์—๊ฒŒ๋Š”

2022๋…„ 3์›” 14์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€
post-thumbnail

1. Introduction

ํฌ์œ ๋ฅ˜์˜ ์‹œ๊ฐ์  ์ฒ˜๋ฆฌ ๋ฉ”์ปค๋‹ˆ์ฆ˜์— ๋Œ€ํ•ด ์—ฐ๊ตฌ์šฐ๋ฆฌ ๋ˆˆ์— ๋ณด์ด๋Š” ์‚ฌ๋ฌผ๋“ค์„ ๊ธฐํ•˜ํ•™์  ๋ชจ์–‘์œผ๋กœ ๋‹จ์ˆœํ™” Vision ์ฑ… โžก๏ธ ์ปดํ“จํ„ฐ ๋น„์ „์ด ์–ด๋–ค ๋ฐฉ์‹์œผ๋กœ ๋ฐœ์ „ํ•ด์•ผ ํ•˜๋Š”์ง€Input Image๊ฐ€ ๋“ค์–ด์™”์„ ๋•Œ ์ด๋ฏธ์ง€์˜ ํŠน์ง•๋“ค์„ ์ถ”์ถœํ•˜๊ณ , ํŠน์ง•์— ๋”ฐ๋ผ depth์™€ surface๋ฅผ ์ถ”์ถœ

2022๋…„ 3์›” 7์ผ
ยท
0๊ฐœ์˜ ๋Œ“๊ธ€