Sign In

ReDON: Recurrent Diffractive Optical Neural Processor with Reconfigurable Self-Modulated Nonlinearity

Created by
  • Haebom
Category
Empty

์ €์ž

Ziang Yin, Qi Jing, Raktim Sarma, Rena Huang, Yu Yao, Jiaqi Gu

๐Ÿ’ก ๊ฐœ์š”

๋ณธ ๋…ผ๋ฌธ์€ ๊ณ ์ •๋œ ํšŒ์ ˆ ๋งˆ์Šคํฌ์˜ ํ•œ๊ณ„๋ฅผ ๊ทน๋ณตํ•˜๊ธฐ ์œ„ํ•ด ์žฌ๊ตฌ์„ฑ ๊ฐ€๋Šฅํ•œ ์ˆœํ™˜ ์ž๊ธฐ ๋ณ€์กฐ ๋น„์„ ํ˜•์„ฑ์„ ๊ฐ–์ถ˜ ์ƒˆ๋กœ์šด ๊ด‘ ์‹ ๊ฒฝ๋ง ํ”„๋กœ์„ธ์„œ์ธ ReDON์„ ์ œ์•ˆํ•œ๋‹ค. ReDON์€ ๊ด‘ํ•™ ์˜์—ญ์—์„œ ์ „ํŒŒ๋˜๋Š” ์‹ ํ˜ธ ์ผ๋ถ€๋ฅผ ๊ฐ์ง€ํ•˜์—ฌ ๊ฐ€๋ณ๊ณ  ํŒŒ๋ผ๋ฏธํ„ฐํ™”๋œ ํ•จ์ˆ˜๋ฅผ ํ†ตํ•ด ์œ„์ƒ ๋˜๋Š” ๊ฐ•๋„๋ฅผ ๋™์ ์œผ๋กœ ์กฐ์ ˆํ•จ์œผ๋กœ์จ ํšจ์œจ์ ์ธ ๋น„์„ ํ˜•์„ฑ์„ ๊ตฌํ˜„ํ•œ๋‹ค. ์ด๋ฅผ ํ†ตํ•ด ๊ธฐ์กด DONN ๋Œ€๋น„ ์ด๋ฏธ์ง€ ์ธ์‹ ๋ฐ ๋ถ„ํ•  ์ž‘์—…์—์„œ ์ •ํ™•๋„๋ฅผ ์ตœ๋Œ€ 20% ํ–ฅ์ƒ์‹œํ‚ค๋Š” ์„ฑ๊ณผ๋ฅผ ๋‹ฌ์„ฑํ–ˆ๋‹ค.

๐Ÿ”‘ ์‹œ์‚ฌ์  ๋ฐ ํ•œ๊ณ„

โ€ข
๊ธฐ์กด DONN์˜ ๊ณ ์ •๋œ ๊ตฌ์กฐ์™€ ์ œํ•œ์ ์ธ ๋น„์„ ํ˜•์„ฑ์„ ๊ทน๋ณตํ•˜๋Š” ์žฌ๊ตฌ์„ฑ ๊ฐ€๋Šฅํ•œ ๊ด‘ํ•™ ์ปดํ“จํŒ… ์•„ํ‚คํ…์ฒ˜์˜ ๊ฐ€๋Šฅ์„ฑ์„ ์ œ์‹œํ•œ๋‹ค.
โ€ข
GLU์™€ ์œ ์‚ฌํ•œ ๊ฒŒ์ดํŠธ ๋ฉ”์ปค๋‹ˆ์ฆ˜์„ ํ†ตํ•ด ์ตœ์†Œํ•œ์˜ ์˜ค๋ฒ„ํ—ค๋“œ๋กœ ๊ฐ•๋ ฅํ•œ ๋น„์„ ํ˜• ํ‘œํ˜„ ๋Šฅ๋ ฅ์„ ํ™•๋ณดํ•จ์œผ๋กœ์จ ๋น„-ํฐ ๋…ธ์ด๋งŒ ์•„ํ‚คํ…์ฒ˜์˜ ์ปดํ“จํŒ… ์„ฑ๋Šฅ์„ ํ–ฅ์ƒ์‹œ์ผฐ๋‹ค.
โ€ข
๋‹ค์–‘ํ•œ ์ž๊ธฐ ๋ณ€์กฐ ์„ค์ •์— ๋Œ€ํ•œ ์ฒด๊ณ„์ ์ธ ๋ถ„์„์„ ํ†ตํ•ด ํ•˜๋“œ์›จ์–ด ํšจ์œจ์„ฑ๊ณผ ๊ณ„์‚ฐ ํ‘œํ˜„๋ ฅ ๊ฐ„์˜ ํŠธ๋ ˆ์ด๋“œ์˜คํ”„๋ฅผ ์ดํ•ดํ•˜๋Š” ๋ฐ ๊ธฐ์—ฌํ•œ๋‹ค.
โ€ข
๋ณธ ์—ฐ๊ตฌ์—์„œ ์ œ์•ˆ๋œ ์ž๊ธฐ ๋ณ€์กฐ ๋ฐฉ์‹์˜ ์•ˆ์ •์„ฑ, ํ™•์žฅ์„ฑ, ๊ทธ๋ฆฌ๊ณ  ๋ณต์žกํ•œ ๋ชจ๋ธ์— ๋Œ€ํ•œ ์ ์šฉ ๊ฐ€๋Šฅ์„ฑ์€ ํ–ฅํ›„ ์ถ”๊ฐ€์ ์ธ ์—ฐ๊ตฌ๊ฐ€ ํ•„์š”ํ•˜๋‹ค.
๐Ÿ‘
Made with Slashpage