Sign In

MemQ: Integrating Q-Learning into Self-Evolving Memory Agents over Provenance DAGs

์ž‘์„ฑ์ž
  • Haebom
์นดํ…Œ๊ณ ๋ฆฌ
Empty

์ €์ž

Junwei Liao, Haoting Shi, Ruiwen Zhou, Jiaqian Wang, Shengtao Zhang, Wei Zhang, Weinan Zhang, Ying Wen, Zhiyu Li, Feiyu Xiong, Bo Tang, Muning Wen

๐Ÿ’ก ๊ฐœ์š”

๊ธฐ์กด์˜ LLM ์—์ด์ „ํŠธ๋“ค์€ ๊ฐ ๊ธฐ์–ต์„ ๋…๋ฆฝ์ ์œผ๋กœ ์ฒ˜๋ฆฌํ•˜์—ฌ ๊ธฐ์–ต ๊ฐ„์˜ ์˜์กด์„ฑ ์‚ฌ์Šฌ์„ ๊ฐ„๊ณผํ–ˆ์Šต๋‹ˆ๋‹ค. ๋ณธ ๋…ผ๋ฌธ์€ provenance DAG๋ฅผ ํ†ตํ•ด ๊ธฐ์–ต ๊ฐ„์˜ ์‹ ์šฉ ์ „๋‹ฌ์„ ์ถ”์ ํ•˜๋Š” MemQ๋ฅผ ์ œ์•ˆํ•˜๋ฉฐ, ์ด๋Š” ์‹œ๊ฐ„์  ๊ฑฐ๋ฆฌ๊ฐ€ ์•„๋‹Œ ๊ตฌ์กฐ์  ๊ทผ์ ‘์„ฑ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ์‹ ์šฉ์„ ๋ถ„๋ฐฐํ•ฉ๋‹ˆ๋‹ค. MemQ๋Š” 6๊ฐ€์ง€ ๋‹ค์–‘ํ•œ ๋ฒค์น˜๋งˆํฌ์—์„œ ์ตœ๊ณ  ์„ฑ๋Šฅ์„ ๋‹ฌ์„ฑํ–ˆ์œผ๋ฉฐ, ํŠนํžˆ ๋‹ค๋‹จ๊ณ„ ์ž‘์—…์—์„œ ํฐ ์„ฑ๋Šฅ ํ–ฅ์ƒ์„ ๋ณด์˜€์Šต๋‹ˆ๋‹ค.

๐Ÿ”‘ ์‹œ์‚ฌ์  ๋ฐ ํ•œ๊ณ„

โ€ข
๊ธฐ์–ต ๊ฐ„์˜ ๋ณต์žกํ•œ ์˜์กด์„ฑ์„ ํšจ๊ณผ์ ์œผ๋กœ ๋ชจ๋ธ๋งํ•˜์—ฌ LLM ์—์ด์ „ํŠธ์˜ ํ•™์Šต ๋ฐ ์ถ”๋ก  ๋Šฅ๋ ฅ์„ ํ–ฅ์ƒ์‹œํ‚ฌ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
โ€ข
Exogenous-Context MDP ํ”„๋ ˆ์ž„์›Œํฌ๋Š” ๊ธฐ์–ต ๊ด€๋ฆฌ ๋ฌธ์ œ๋ฅผ ๊ตฌ์กฐํ™”ํ•˜๊ณ  ๋ถ„์„ํ•˜๋Š” ๋ฐ ์œ ์šฉํ•œ ํ‹€์„ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.
โ€ข
$\gamma$์™€ $\lambda$์™€ ๊ฐ™์€ ํŒŒ๋ผ๋ฏธํ„ฐ๊ฐ€ EC-MDP ๊ตฌ์กฐ์™€ ์–ด๋–ป๊ฒŒ ์ƒํ˜ธ์ž‘์šฉํ•˜๋Š”์ง€์— ๋Œ€ํ•œ ์‹ฌ์ธต์ ์ธ ์ดํ•ด๋ฅผ ๋ฐ”ํƒ•์œผ๋กœ ํ–ฅํ›„ ์—ฐ๊ตฌ ๋ฐฉํ–ฅ์„ ์ œ์‹œํ•ฉ๋‹ˆ๋‹ค.
โ€ข
ํ˜„์žฌ๋Š” ์ฝ”๋“œ ๊ณต๊ฐœ๊ฐ€ ์˜ˆ์ •๋˜์–ด ์žˆ์œผ๋ฉฐ, ์‹ค์ œ ์‹œ์Šคํ…œ ์ ์šฉ ๋ฐ scalability์— ๋Œ€ํ•œ ์ถ”๊ฐ€์ ์ธ ์—ฐ๊ตฌ๊ฐ€ ํ•„์š”ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
๐Ÿ‘
Made with Slashpage