Sign In

PRECEPT: Planning Resilience via Experience, Context Engineering & Probing Trajectories A Unified Framework for Test-Time Adaptation with Compositional Rule Learning and Pareto-Guided Prompt Evolution

Created by
  • Haebom
Category
Empty

์ €์ž

Arash Shahmansoori

๐Ÿ’ก ๊ฐœ์š”

๋ณธ ๋…ผ๋ฌธ์€ LLM ์—์ด์ „ํŠธ๊ฐ€ ์กฐ๊ฑด ์ˆ˜๊ฐ€ ๋Š˜์–ด๋‚จ์— ๋”ฐ๋ผ ์ง€์‹ ๊ฒ€์ƒ‰ ์„ฑ๋Šฅ์ด ์ €ํ•˜๋˜๊ณ , ํ•™์Šต๋œ ๊ทœ์น™์„ ์‹ ๋ขฐ์„ฑ ์žˆ๊ฒŒ ์กฐํ•ฉํ•˜๊ธฐ ์–ด๋ ค์šฐ๋ฉฐ, ์˜ค๋ž˜๋˜๊ฑฐ๋‚˜ ์ ๋Œ€์ ์ธ ์ง€์‹์„ ํƒ์ง€ํ•˜๋Š” ๋ฉ”์ปค๋‹ˆ์ฆ˜์ด ๋ถ€์กฑํ•œ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๊ธฐ ์œ„ํ•œ ํ†ตํ•ฉ ํ”„๋ ˆ์ž„์›Œํฌ PRECEPT๋ฅผ ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค. PRECEPT๋Š” ๊ฒฐ์ •๋ก ์  ๊ทœ์น™ ๊ฒ€์ƒ‰, ์ถฉ๋Œ ๊ฐ์ง€ ๋ฉ”๋ชจ๋ฆฌ, ๊ทธ๋ฆฌ๊ณ  ํŒŒ๋ ˆํ†  ์ตœ์ ์„ ํ™œ์šฉํ•œ ํ”„๋กฌํ”„ํŠธ ์ง„ํ™” ๋ฃจํ”„๋กœ ๊ตฌ์„ฑ๋˜์–ด ํ…Œ์ŠคํŠธ ํƒ€์ž„ ์ ์‘ ๋Šฅ๋ ฅ์„ ํ–ฅ์ƒ์‹œํ‚ต๋‹ˆ๋‹ค.

๐Ÿ”‘ ์‹œ์‚ฌ์  ๋ฐ ํ•œ๊ณ„

โ€ข
๊ฒฐ์ •๋ก ์  ๊ทœ์น™ ๊ฒ€์ƒ‰: ๊ธฐ์กด์˜ ๋ถ€๋ถ„ ๋งค์นญ ์˜ค๋ฅ˜๋ฅผ ์ œ๊ฑฐํ•˜๊ณ  ๊ณ„์ธต์  ์˜๋ฏธ ๊ตฌ์กฐ๋ฅผ ํ†ตํ•ด ๊ทœ์น™์˜ ์กฐํ•ฉ์  ๊ตฌ์„ฑ์„ ์ง€์›ํ•ฉ๋‹ˆ๋‹ค.
โ€ข
์ถฉ๋Œ ๊ฐ์ง€ ๋ฉ”๋ชจ๋ฆฌ: ์ •์ -๋™์  ์ง€์‹ ๊ฐ„์˜ ๋ถˆ์ผ์น˜๋ฅผ ํ•ด๊ฒฐํ•˜๊ณ  ์ง€์†์ ์ธ ์ง€์‹ ๋ณ€ํ™”์— ๋Œ€ํ•œ ์ ์‘์„ ๊ฐ€๋Šฅํ•˜๊ฒŒ ํ•ฉ๋‹ˆ๋‹ค.
โ€ข
ํŒŒ๋ ˆํ†  ์ตœ์  ๊ธฐ๋ฐ˜ ํ”„๋กฌํ”„ํŠธ ์ง„ํ™”: ํ”„๋กฌํ”„ํŠธ์˜ ํšจ์œจ์„ฑ์„ end-to-end ํ‰๊ฐ€๋ฅผ ํ†ตํ•ด ์ง€์†์ ์œผ๋กœ ๊ฐœ์„ ํ•ฉ๋‹ˆ๋‹ค.
โ€ข
ํ•œ๊ณ„์ : ๋ณธ ์—ฐ๊ตฌ๋Š” ์ œ์‹œ๋œ ํ”„๋ ˆ์ž„์›Œํฌ์˜ ์„ฑ๋Šฅ์„ ์‹คํ—˜์ ์œผ๋กœ ์ž…์ฆํ–ˆ์ง€๋งŒ, ์‹ค์ œ ๋ณต์žกํ•˜๊ณ  ๋™์ ์ธ ํ™˜๊ฒฝ์—์„œ์˜ ์ผ๋ฐ˜ํ™” ์„ฑ๋Šฅ๊ณผ ํ™•์žฅ์„ฑ์— ๋Œ€ํ•œ ์ถ”๊ฐ€์ ์ธ ๊ฒ€์ฆ์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.
๐Ÿ‘
Made with Slashpage