KOINEU Logo
ํด๋ผ์šฐ๋“œ ์—†์ด๋„ ๊ฐ€๋Šฅํ•œ ์ธ์‹œ๋˜ํŠธ ์ž๋™ ๋ถ„๋ฅ˜ ๋กœ์ปฌ LLM ์„ฑ๋Šฅ ํ‰๊ฐ€

ํด๋ผ์šฐ๋“œ ์—†์ด๋„ ๊ฐ€๋Šฅํ•œ ์ธ์‹œ๋˜ํŠธ ์ž๋™ ๋ถ„๋ฅ˜ ๋กœ์ปฌ LLM ์„ฑ๋Šฅ ํ‰๊ฐ€

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ชฉ์  ๋ฌธ์ œ ์ธ์‹ : ํด๋ผ์šฐ๋“œ ๊ธฐ๋ฐ˜ LLM์€ ๋น„์šฉยท์ง€์—ฐยท๋ฐ์ดํ„ฐ ์œ ์ถœ ์œ„ํ—˜์ด ์žˆ์–ด, ์˜จโ€‘ํ”„๋ ˆ๋ฏธ์Šค(์˜จํ”„๋ ˆ๋ฏธ์Šค) ํ™˜๊ฒฝ์—์„œ์˜ ํ™œ์šฉ ๊ฐ€๋Šฅ์„ฑ์„ ๊ฒ€์ฆํ•  ํ•„์š”๊ฐ€ ์žˆ๋‹ค. ๋ชฉํ‘œ : ๋กœ์ปฌ์—์„œ ์‹คํ–‰ ๊ฐ€๋Šฅํ•œ SLM์ด ๋ณด์•ˆ ์ธ์‹œ๋˜ํŠธ ์ž๋™ ๋ถ„๋ฅ˜ ์ž‘์—…์„ ์ถฉ๋ถ„ํžˆ ์ˆ˜ํ–‰ํ•  ์ˆ˜ ์žˆ๋Š”์ง€, ํŠนํžˆ temperature ํ•˜์ดํผํŒŒ๋ผ๋ฏธํ„ฐ์™€ ๋ชจ๋ธ ๊ทœ๋ชจยทํ•˜๋“œ์›จ์–ด ์‚ฌ์–‘์ด ์„ฑ๋Šฅ์— ๋ฏธ์น˜๋Š” ์˜ํ–ฅ์„ ์ •๋Ÿ‰์ ์œผ๋กœ ํ‰๊ฐ€ํ•œ๋‹ค. 2. ์‹คํ—˜ ์„ค๊ณ„ | ์š”์†Œ | ์ƒ์„ธ ๋‚ด์šฉ | | | | | ๋ฐ์ดํ„ฐ | CSIRT ์‹ค์ œ ์ธ์‹œ๋˜ํŠธ 194๊ฑด ์ค‘ 6๊ฐœ ์นดํ…Œ๊ณ ๋ฆฌยท๊ฐ 4๊ฑด์”ฉ(์ด 24๊ฑด) ์„ ํƒ, ์ „๋ฌธ

AI Autonomy Coefficient ($ฮฑ$): Defining Boundaries for Responsible AI Systems

AI Autonomy Coefficient ($ฮฑ$): Defining Boundaries for Responsible AI Systems

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ฌธ์ œ ์ •์˜ HITL์˜ ์˜ค์šฉ : ๊ธฐ์กด HITL์€ ํ’ˆ์งˆ ๋ณด์ฆยท์•ˆ์ „์„ฑ ํ™•๋ณด๋ฅผ ์œ„ํ•œ ๋ณด์กฐ์  ๋ฉ”์ปค๋‹ˆ์ฆ˜์œผ๋กœ ์„ค๊ณ„๋์ง€๋งŒ, ์ƒ์—…์  ์••๋ฐ• ์†์—์„œ โ€œํ•„์ˆ˜์ ์ธ ์ธ๊ฐ„ ์ž‘์—…โ€์œผ๋กœ ์ „๋ฝํ•˜๊ณ  ์žˆ๋‹ค. HISOAI ๊ฐœ๋… : ์ธ๊ฐ„์ด ์‹ค์ œ AI ๊ธฐ๋Šฅ์„ ๋Œ€์ฒดํ•˜๋Š” ๊ตฌ์กฐ์  ์˜์กด์„ฑ์„ ๋ช…ํ™•ํžˆ ์ •์˜ํ•จ์œผ๋กœ์จ, ์œค๋ฆฌ์ ยท๊ฒฝ์ œ์  ๋น„์šฉ(โ€˜ghost workโ€™, ๊ณผ์žฅ๋œ ์ž๋™ํ™” ์ฃผ์žฅ)์„ ์ •๋Ÿ‰ํ™”ํ•œ๋‹ค. ์ด๋Š” ๊ธฐ์กด ๋ฌธํ—Œ์—์„œ ๋‹ค๋ฃจ์ง€ ์•Š์•˜๋˜ ๊ตฌ์กฐ์  ํˆฌ๋ช…์„ฑ ๋ฌธ์ œ๋ฅผ ์ œ๊ธฐํ•œ๋‹ค. 2. ํ•ต์‹ฌ ๊ธฐ์—ฌ | ๊ตฌ๋ถ„ | ๋‚ด์šฉ | | | | | ๊ฐœ๋…์  | HISOAI๋ผ๋Š” ์ƒˆ๋กœ์šด ์‹คํŒจ ๋ชจ๋“œ ์ •์˜ |

System
Fail Fast, Win Big: Rethinking the Drafting Strategy in Speculative Decoding via Diffusion LLMs

Fail Fast, Win Big: Rethinking the Drafting Strategy in Speculative Decoding via Diffusion LLMs

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ Diffusion LLM(dLLM) ์€ ํ† ํฐ์„ ๋™์‹œ์— โ€œunmaskโ€ ํ•˜๋Š” ๋น„์ž์œจ์ (semiautoregressive) ๋ฐฉ์‹์œผ๋กœ, ํ•œ ๋ฒˆ์˜ forward pass์— ๋‹ค์ˆ˜ ํ† ํฐ์„ ์ƒ์„ฑํ•œ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์กฐ๊ฑด๋ถ€ ๋…๋ฆฝ์„ฑ ๊ฐ€์ • ๋•Œ๋ฌธ์— ํ† ํฐ ๊ฐ„ ์ƒํ˜ธ ์˜์กด์„ฑ์„ ์ถฉ๋ถ„ํžˆ ๋ฐ˜์˜ํ•˜์ง€ ๋ชปํ•ด ํ’ˆ์งˆ์ด ๋–จ์–ด์ง€๋Š” ๊ฒƒ์ด ์ผ๋ฐ˜์ ์ธ ๋ฌธ์ œ๋‹ค. Speculative Decoding ์€ ๊ฐ€๋ฒผ์šด ์ดˆ์•ˆ(drafter) ๋ชจ๋ธ์ด ํ›„๋ณด ํ† ํฐ์„ ์ œ์‹œํ•˜๊ณ , ๋ฌด๊ฑฐ์šด ๋ชฉํ‘œ ๋ชจ๋ธ์ด ์ด๋ฅผ ๊ฒ€์ฆํ•˜๋Š” ๋ฐฉ์‹์œผ๋กœ, ์ดˆ์•ˆ ๊ธธ์ด๊ฐ€ ๊ธธ์ˆ˜๋ก ๊ฐ€์† ํšจ๊ณผ๊ฐ€ ์ปค์ง€์ง€๋งŒ ๋ฆฌ์ ์…˜ ํ™•๋ฅ ๋„ ๊ธฐํ•˜๊ธ‰

UCCL-EP: Portable Expert-Parallel Communication

UCCL-EP: Portable Expert-Parallel Communication

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ฌธ์ œ ์ •์˜ MoE ๋ชจ๋ธ ์€ ์ˆ˜๋ฐฑ ๊ฐœ์˜ ์ „๋ฌธ๊ฐ€๋ฅผ GPU์— ๋ถ„์‚ฐ ๋ฐฐ์น˜ํ•˜๊ณ , ํ† ํฐ๋‹น ๋ช‡ ๊ฐœ๋งŒ ์„ ํƒํ•ด ํ™œ์„ฑํ™”ํ•จ์œผ๋กœ์จ ํŒŒ๋ผ๋ฏธํ„ฐ ์šฉ๋Ÿ‰์€ ํฌ๊ฒŒ ๋Š˜๋ฆฌ๋ฉด์„œ๋„ ์—ฐ์‚ฐ ๋น„์šฉ์€ ๋‚ฎ๊ฒŒ ์œ ์ง€ํ•œ๋‹ค. EP ํ™˜๊ฒฝ์—์„œ๋Š” ํ† ํฐโ€‘๋ ˆ๋ฒจ (โ‰ˆ7 KB) ๋ฏธ์„ธ ์ „์†ก ์ด ๋นˆ๋ฒˆํžˆ ๋ฐœ์ƒํ•˜๊ณ , ๋ผ์šฐํŒ… ๋Œ€์ƒ์ด ์‹คํ–‰ ์‹œ์ ์— ๊ฒฐ์ •๋œ๋‹ค. ๋”ฐ๋ผ์„œ ๋‚ฎ์€ ๋ ˆ์ดํ„ด์‹œยท๋†’์€ ํŠธ๋ž˜ํ”ฝ ์„ ๋™์‹œ์— ๋งŒ์กฑํ•ด์•ผ ํ•œ๋‹ค. ๊ธฐ์กด DeepEP์€ GPUโ€‘initiated RDMA (IBGDA)๋ฅผ ์ด์šฉํ•ด GPU๊ฐ€ ์ง์ ‘ NIC์— ์“ฐ๋Š” ๋ฐฉ์‹์œผ๋กœ ์ตœ๊ณ ์˜ ์„ฑ๋Šฅ์„ ๋‹ฌ์„ฑํ–ˆ์ง€๋งŒ, 1. GPUโ€‘NIC ์ธํ„ฐํŽ˜์ด์Šค๊ฐ€ ๋ฒค

Beyond Vision: Contextually Enriched Image Captioning with Multi-Modal Retrieval

Beyond Vision: Contextually Enriched Image Captioning with Multi-Modal Retrieval

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ๋ฌธ์ œ ์ •์˜ : ๊ธฐ์กด ์ด๋ฏธ์ง€ ์บก์…”๋‹ ๋ชจ๋ธ์€ ์‹œ๊ฐ์  ์š”์†Œ์—๋งŒ ์˜์กดํ•ด ์‚ฌ๊ฑด ๋ฐฐ๊ฒฝยท์‹œ๊ฐ„ยท๊ฒฐ๊ณผยท๊ณ ์œ ๋ช…์‚ฌ ๋“ฑ ๋น„์‹œ๊ฐ์  ์ •๋ณด๋ฅผ ์ œ๊ณตํ•˜์ง€ ๋ชปํ•œ๋‹ค. ์ด๋Š” ํŠนํžˆ ๋‰ด์Šค ์‚ฌ์ง„ยท์—ญ์‚ฌ ๊ธฐ๋กยท๊ต์œก ์ฝ˜ํ…์ธ ์™€ ๊ฐ™์ด โ€œ์™œ ์ด ์‚ฌ์ง„์ด ์ค‘์š”ํ•œ๊ฐ€?โ€๋ฅผ ์„ค๋ช…ํ•ด์•ผ ํ•˜๋Š” ๋„๋ฉ”์ธ์—์„œ ํฐ ์ œ์•ฝ์ด ๋œ๋‹ค. ์„ ํ–‰ ์—ฐ๊ตฌ์™€ ์ฐจ๋ณ„์  : ๊ธฐ์กด ์ง€์‹โ€‘์ฆ๊ฐ• ์บก์…”๋‹(์˜ˆ: Knowledgeโ€‘Enhanced Captioning, Retrievalโ€‘Augmented Generation)์€ ์ฃผ๋กœ ์งง์€ ๋ฌธ์žฅ ์ˆ˜์ค€์— ๋จธ๋ฌผ๋ฉฐ, ์ด๋ฏธ์ง€์™€ ํ…์ŠคํŠธ ๊ฐ„ ์ง์ ‘์ ์ธ ์ •๋ ฌ์— ์ดˆ์ ์„ ๋งž์ถ˜๋‹ค. ๋ณธ

Faith Lens ๋Œ€๊ทœ๋ชจ ์–ธ์–ด ๋ชจ๋ธ ์‹ ๋ขฐ์„ฑ ํ™˜๊ฐ ํƒ์ง€๋ฅผ ์œ„ํ•œ ๋น„์šฉ ํšจ์œจ์  ๋ชจ๋ธ

Faith Lens ๋Œ€๊ทœ๋ชจ ์–ธ์–ด ๋ชจ๋ธ ์‹ ๋ขฐ์„ฑ ํ™˜๊ฐ ํƒ์ง€๋ฅผ ์œ„ํ•œ ๋น„์šฉ ํšจ์œจ์  ๋ชจ๋ธ

1. ์ฃผ์š” ๊ธฐ์—ฌ | ๋ฒˆํ˜ธ | ๋‚ด์šฉ | ์˜์˜ | | | | | | โ‘  | ์„ค๋ช…์ด ํฌํ•จ๋œ ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ ๋ฅผ ๊ณ ์„ฑ๋Šฅ LLM(DeepSeekโ€‘V3.2โ€‘Think)์œผ๋กœ ์ž๋™ ์ƒ์„ฑ | ๋ผ๋ฒจ๋งŒ ์žˆ๋Š” ๊ธฐ์กด ํ•ฉ์„ฑ ๋ฐ์ดํ„ฐ์™€ ๋‹ฌ๋ฆฌ, ์„ค๋ช…๊นŒ์ง€ ์ œ๊ณตํ•ด ํ•™์Šต ์‹œ ์„ค๋ช… ๋Šฅ๋ ฅ์„ ๋‚ด์žฌํ™” | | โ‘ก | ๋‹ค์ค‘ ๊ธฐ์ค€ ํ•„ํ„ฐ๋ง (๋ผ๋ฒจ ์ •ํ™•์„ฑ, ์„ค๋ช… ํ’ˆ์งˆ, ๋ฐ์ดํ„ฐ ๋‹ค์–‘์„ฑ) | ์ €ํ’ˆ์งˆยทํŽธํ–ฅ ๋ฐ์ดํ„ฐ๊ฐ€ ๋ชจ๋ธ์— ํ•™์Šต๋˜๋Š” ๊ฒƒ์„ ๋ฐฉ์ง€, ์ „๋ฐ˜์ ์ธ ์ผ๋ฐ˜ํ™” ํ–ฅ์ƒ | | โ‘ข | ๊ทœ์น™ ๊ธฐ๋ฐ˜ RL (์˜ˆ์ธก ์ •ํ™•๋„ + ์„ค๋ช… ํ’ˆ์งˆ + ํฌ๋งท ๋ณด์ƒ) | ๋‹จ์ˆœ SFT๋งŒ์œผ๋กœ๋Š” ์–ป๊ธฐ ํž˜๋“  ์„ค๋ช…์˜ ์‹ค์šฉ์„ฑ(์ดˆ๋ณด

๊ณ„์ธต์  ๊ต์œก ๊ฐ๋…์„ ํ†ตํ•œ ์ €๋น„์šฉ LLM ํŠœํ„ฐ์˜ ์‹ ๋ขฐ์„ฑ ๊ฐ•ํ™”

๊ณ„์ธต์  ๊ต์œก ๊ฐ๋…์„ ํ†ตํ•œ ์ €๋น„์šฉ LLM ํŠœํ„ฐ์˜ ์‹ ๋ขฐ์„ฑ ๊ฐ•ํ™”

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ฌธ์ œ ์ •์˜ ๊ต์œก ํ˜„์žฅ์˜ ๊ธด๊ธ‰์„ฑ : ์ „ ์„ธ๊ณ„์ ์œผ๋กœ ๊ต์‚ฌ ๋ถ€์กฑ์ด ์‹ฌํ™”๋˜๊ณ  ์žˆ์–ด ์ž๋™ ํŠœํ„ฐ๋ง ์‹œ์Šคํ…œ์— ๋Œ€ํ•œ ์ˆ˜์š”๊ฐ€ ๊ธ‰์ฆํ•˜๊ณ  ์žˆ๋‹ค. LLM์˜ ํ•œ๊ณ„ : ๊ธฐ์กด ๋‹จ์ผ ์—์ด์ „ํŠธ ๋ฐฉ์‹์€ ์ƒ์„ฑ ๊ณผ ํ‰๊ฐ€ ๋ฅผ ๋™์ผ ๋ชจ๋ธ์— ๋งก๊ฒจ ํ™•์ฆ ํŽธํ–ฅ ๊ณผ ์‹œ์ฝ”ํŒ์‹œ ๋ฅผ ์ดˆ๋ž˜ํ•œ๋‹ค. ์ด๋Š” ํ•™์ƒ์˜ ์˜ค๊ฐœ๋…์„ ๊ต์ •ํ•˜์ง€ ๋ชปํ•˜๊ณ  ์˜คํžˆ๋ ค ๊ฐ•ํ™”์‹œํ‚ฌ ์œ„ํ—˜์ด ์žˆ๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด โ€“ HPO ํ”„๋ ˆ์ž„์›Œํฌ | ๋‹จ๊ณ„ | ์—ญํ•  | ์ฃผ์š” ๊ธฐ๋Šฅ | | | | | | Phase 1 โ€“ Intelligence Distillation | Conceptual Analyst , Beh

์Œ์„ฑ ๋ถ„์„ ๊ธฐ๋ฐ˜ ๊ทผ์œ„์ถ•์„ฑ ์ธก์‚ญ๊ฒฝํ™”์ฆ ์ค‘์ฆ๋„ ํ†ตํ•ฉ ๋ถ„๋ฅ˜ ์—ฐ๊ตฌ

์Œ์„ฑ ๋ถ„์„ ๊ธฐ๋ฐ˜ ๊ทผ์œ„์ถ•์„ฑ ์ธก์‚ญ๊ฒฝํ™”์ฆ ์ค‘์ฆ๋„ ํ†ตํ•ฉ ๋ถ„๋ฅ˜ ์—ฐ๊ตฌ

1. ๋ฐ์ดํ„ฐ์™€ ๋ฌธ์ œ ์„ค์ • ๋ฐ์ดํ„ฐ ๊ทœ๋ชจยท๋ถˆ๊ท ํ˜• : ํ›ˆ๋ จ 219๋ช…, ๊ฒ€์ฆ 53๋ช…์œผ๋กœ ์ „์ฒด ์ƒ˜ํ”Œ์ด ์ ๊ณ , ํด๋ž˜์Šค 1(๊ฐ€์žฅ ์‹ฌ๊ฐ) 4๋ช…, ํด๋ž˜์Šค 5(์ •์ƒ) 86๋ช… ๋“ฑ ์‹ฌ๊ฐํ•œ ํด๋ž˜์Šค ๋ถˆ๊ท ํ˜•์ด ์กด์žฌํ•œ๋‹ค. ์ด๋Š” ๋ชจ๋ธ์ด ์†Œ์ˆ˜ ํด๋ž˜์Šค์— ๊ณผ์†Œ ์ ํ•ฉ(overโ€‘fitting)ํ•˜๊ฑฐ๋‚˜ ๋‹ค์ˆ˜ ํด๋ž˜์Šค์— ํŽธํ–ฅ๋  ์œ„ํ—˜์„ ๋†’์ธ๋‹ค. ๋…ผ๋ฌธ์—์„œ๋Š” ๊ฐ€์ค‘์น˜ ์†์‹ค, ๋ฐ์ดํ„ฐ ์ฆ๊ฐ•, ๊ณ„์ธตํ˜• ๋ถ„๋ฅ˜ ๋“ฑ์œผ๋กœ ์™„ํ™”ํ•˜๋ ค ํ–ˆ์ง€๋งŒ, ๋ณด๋‹ค ์ฒด๊ณ„์ ์ธ ๋ฆฌ์ƒ˜ํ”Œ๋ง(SMOTE ๋“ฑ) ํ˜น์€ costโ€‘sensitive learning ์ด ์ถ”๊ฐ€๋  ์—ฌ์ง€๊ฐ€ ์žˆ๋‹ค. ๋‹ค์ค‘ ๋ฐœํ™” ํ™œ์šฉ : ํ™”์ž๋‹น 8๊ฐœ์˜ ๋ฐœํ™”(5๋ชจ์Œ +

์‹ค์‹œ๊ฐ„ ๋น„๋””์˜ค ๊ธฐ๋ฐ˜ 2D ๋™์ž‘ ๋ชจ๋ฐฉ์„ ํ†ตํ•œ ๋‹ค์ค‘ ์บ๋ฆญํ„ฐ ์ œ์–ด ํ•™์Šต

์‹ค์‹œ๊ฐ„ ๋น„๋””์˜ค ๊ธฐ๋ฐ˜ 2D ๋™์ž‘ ๋ชจ๋ฐฉ์„ ํ†ตํ•œ ๋‹ค์ค‘ ์บ๋ฆญํ„ฐ ์ œ์–ด ํ•™์Šต

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ MoCap ๋น„์šฉ ๋ฌธ์ œ : ๊ณ ํ’ˆ์งˆ 3D ๋ชจ์…˜ ์บก์ฒ˜๋Š” ์žฅ๋น„ยท์ธ๋ ฅยท์‹œ๊ฐ„์ด ๋งŽ์ด ์†Œ์š”๋ผ ํ™•์žฅ์„ฑ์ด ๋‚ฎ๋‹ค. ๋น„๋””์˜ค ๋ฐ์ดํ„ฐ์˜ ํ’๋ถ€ํ•จ : ์ธํ„ฐ๋„ท์— ์กด์žฌํ•˜๋Š” โ€˜inโ€‘theโ€‘wildโ€™ ์˜์ƒ์€ ๋ฌดํ•œ์— ๊ฐ€๊น๊ณ , 2D ํ‚คํฌ์ธํŠธ ์ถ”์ถœ์€ ํ˜„์žฌ ๊ฑฐ์˜ ๋น„์šฉ์ด ๋“ค์ง€ ์•Š๋Š”๋‹ค. ๊ธฐ์กด 3D ์žฌ๊ตฌ์„ฑ ํ•œ๊ณ„ : 2Dโ†’3D ๋ณ€ํ™˜์€ ๊ทผ๋ณธ์ ์ธ ๊นŠ์ด ๋ถˆํ™•์‹ค์„ฑ์„ ๋‚ดํฌํ•˜๊ณ , ํŠนํžˆ ์ธ๊ฐ„โ€‘๋ฌผ์ฒด ์ƒํ˜ธ์ž‘์šฉ(HOI)์ด๋‚˜ ๋น„์ธ๊ฐ„(๋™๋ฌผ) ๋™์ž‘์—์„œ๋Š” ์‚ฌ์ „ ํ•™์Šต๋œ 3D ํ”„๋ผ์ด์–ด๊ฐ€ ๋ถ€์กฑํ•ด ๋ฌผ๋ฆฌ์  ํƒ€๋‹น์„ฑ์ด ๋–จ์–ด์ง„๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด | ์š”์†Œ | ์„ค๋ช… | ์žฅ์  | | | | |

์—์ด์ „ํŠธ ์‹œ์Šคํ…œ ์Šค์ผ€์ผ๋ง ์›๋ฆฌ ๋‹ค์ค‘ ์—์ด์ „ํŠธ ํ˜‘์—…๊ณผ ๋ชจ๋ธ ๋Šฅ๋ ฅ์˜ ์ •๋Ÿ‰์  ๋ถ„์„

์—์ด์ „ํŠธ ์‹œ์Šคํ…œ ์Šค์ผ€์ผ๋ง ์›๋ฆฌ ๋‹ค์ค‘ ์—์ด์ „ํŠธ ํ˜‘์—…๊ณผ ๋ชจ๋ธ ๋Šฅ๋ ฅ์˜ ์ •๋Ÿ‰์  ๋ถ„์„

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ์—์ด์ „ํŠธ ์‹œ์Šคํ…œ์˜ ๊ธ‰์„ฑ์žฅ : LLM ๊ธฐ๋ฐ˜ ์—์ด์ „ํŠธ๋Š” ์ฝ”๋“œ ์ƒ์„ฑ, ์›น ๋ธŒ๋ผ์šฐ์ง•, ์˜๋ฃŒ ์˜์‚ฌ๊ฒฐ์ • ๋“ฑ ๋‹ค์–‘ํ•œ ๋ถ„์•ผ์—์„œ ์‹ค์šฉํ™”๋˜๊ณ  ์žˆ๋‹ค. ์ •๋Ÿ‰์  ์„ค๊ณ„ ๊ฐ€์ด๋“œ ๋ถ€์žฌ : ๊ธฐ์กด ์—ฐ๊ตฌ๋Š” โ€œ์—์ด์ „ํŠธ ์ˆ˜๋ฅผ ๋Š˜๋ฆฌ๋ฉด ๋œ๋‹คโ€๋Š” ๊ฒฝํ—˜์  ์ฃผ์žฅ์— ๋จธ๋ฌผ๋Ÿฌ, ์‹ค์ œ ๋ฐฐํฌ ์‹œ ๋น„์šฉยท์„ฑ๋Šฅ ํŠธ๋ ˆ์ด๋“œ์˜คํ”„๋ฅผ ์˜ˆ์ธกํ•˜๊ธฐ ์–ด๋ ต๋‹ค. ๋ณธ ์—ฐ๊ตฌ์˜ ์ฐจ๋ณ„์  : ์—์ด์ „ํŠธ ํ‰๊ฐ€ ์ •์˜, ํ†ต์ œ๋œ ์‹คํ—˜ ์„ค๊ณ„, ๊ทธ๋ฆฌ๊ณ  ํ˜‘์—… ๋ฉ”ํŠธ๋ฆญ ๊ธฐ๋ฐ˜ ์˜ˆ์ธก ๋ชจ๋ธ ์„ ํ†ตํ•ด โ€œ์–ธ์ œ, ์–ด๋–ป๊ฒŒ, ์–ด๋А ์ •๋„โ€ ํ˜‘์—…์ด ์œ ๋ฆฌํ•œ์ง€๋ฅผ ๊ณผํ•™์ ์œผ๋กœ ๊ทœ๋ช…ํ•œ๋‹ค. 2. ์‹คํ—˜ ์„ค๊ณ„ ๋ฐ ๋ฐฉ๋ฒ•๋ก  | ์š”์†Œ | ๋‚ด์šฉ |

No Image

2512.11807

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ฌธ์ œ ์ œ๊ธฐ Faizal et al. (2025) ์€ Gรถdelโ€‘๋ถˆ์™„์ „์„ฑ, Tarskiโ€‘๋ถˆ๊ฐ€์ •์˜์„ฑ, Chaitinโ€‘์ •๋ณด ํ•œ๊ณ„ ๋“ฑ์„ ๋ฌผ๋ฆฌํ•™์— ์ ์šฉํ•ด โ€œ์šฐ์ฃผ๋ฅผ ์™„์ „ํ•˜๊ฒŒ ๊ธฐ์ˆ ํ•˜๋Š” ํ˜•์‹์  ์ด๋ก ์€ ์กด์žฌํ•  ์ˆ˜ ์—†์œผ๋ฉฐ, ๋”ฐ๋ผ์„œ ์šฐ์ฃผ ์‹œ๋ฎฌ๋ ˆ์ด์…˜๋„ ๋ถˆ๊ฐ€๋Šฅํ•˜๋‹คโ€๊ณ  ์ฃผ์žฅํ•œ๋‹ค. Redden์€ ์ด ์ฃผ์žฅ์— ๋‘ ๊ฐ€์ง€ ํ•ต์‹ฌ ์˜ค๋ฅ˜ ๊ฐ€ ์žˆ๋‹ค๊ณ  ์ง€์ ํ•œ๋‹ค. 1. ์ฆ๋ช… ๊ฐ€๋Šฅ์„ฑ ๊ณผ ๊ณ„์‚ฐ ์‹คํ–‰ ์„ ๋™์ผ์‹œํ•จ. 2. ๋ฌผ๋ฆฌ ๋ฒ•์น™์ด ๋ถˆ๊ฐ€๊ฒฐ์ • ๋ฌธ์ œ ์˜ ํ•ด๋‹ต์— ์˜์กดํ•œ๋‹ค๋Š” ์ „์ œ ์—†์ด ๊ฒฐ๋ก ์„ ๋„์ถœํ•จ. 2. ํ•ต์‹ฌ ๊ฐœ๋… ์ •๋ฆฌ | ๊ตฌ๋ถ„ | ์ •์˜ | ๋…ผ๋ฌธ์—์„œ์˜ ์—ญํ•  | | |

DoVer: Intervention-Driven Auto Debugging for LLM Multi-Agent Systems

DoVer: Intervention-Driven Auto Debugging for LLM Multi-Agent Systems

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ฌธ์ œ ์ •์˜ ๋‹ค์ค‘โ€‘์—์ด์ „ํŠธ LLM ์‹œ์Šคํ…œ ์€ ์—ฌ๋Ÿฌ ์—ญํ• ์˜ ์—์ด์ „ํŠธ๊ฐ€ ํ˜‘์—…ํ•ด ๋ณต์žกํ•œ ์ž‘์—…์„ ์ˆ˜ํ–‰ํ•œ๋‹ค. ์‹คํŒจ๋Š” ์˜ˆ์™ธยทํฌ๋ž˜์‹œ๊ฐ€ ์•„๋‹ˆ๋ผ โ€œ์ •์ƒ ์‹คํ–‰ ํ›„ ๋ถ€์ •ํ™•ํ•œ ๊ฒฐ๊ณผโ€ ํ˜•ํƒœ์ด๋ฉฐ, ์ด๋Š” ๊ธด ํ…์ŠคํŠธ ๋กœ๊ทธ ์™€ ๋ถ„๊ธฐ๋œ ์ƒํ˜ธ์ž‘์šฉ ๋•Œ๋ฌธ์— ์›์ธ ํŒŒ์•…์ด ์–ด๋ ค์›Œ์ง„๋‹ค. ๊ธฐ์กด ๋กœ๊ทธโ€‘๊ธฐ๋ฐ˜ ๊ท€์† ์€ LLM์—๊ฒŒ ์ „์ฒด ๋กœ๊ทธ๋ฅผ ์ฃผ๊ณ  โ€œ๋ˆ„๊ฐ€ ์–ธ์ œ ์˜ค๋ฅ˜๋ฅผ ์ผ์œผ์ผฐ๋Š”๊ฐ€โ€๋ฅผ ๋ฌป๋Š” ๋ฐฉ์‹์ธ๋ฐ, ์ •๋‹ต ๋ผ๋ฒจ์ด ๋ถˆํ™•์‹ค ํ•˜๊ณ  ๋‹ค์ค‘ ์›์ธ ์ด ์กด์žฌํ•  ์ˆ˜ ์žˆ์–ด ๊ทผ๋ณธ์ ์ธ ํ•œ๊ณ„๊ฐ€ ์žˆ๋‹ค. 2. DoVer ํ”„๋ ˆ์ž„์›Œํฌ ํ•ต์‹ฌ ์•„์ด๋””์–ด | ๋‹จ๊ณ„ | ์„ค๋ช… | | | | | Hypothes

System
Energy-Aware Data-Driven Model Selection in LLM-Orchestrated AI Systems

Energy-Aware Data-Driven Model Selection in LLM-Orchestrated AI Systems

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ LLMโ€‘์˜ค์ผ€์ŠคํŠธ๋ ˆ์ด์…˜ ์€ ์ตœ๊ทผ ์—์ด์ „ํŠธํ˜• AI ์‹œ์Šคํ…œ์—์„œ ํ•ต์‹ฌ ๊ธฐ์ˆ ๋กœ ๋ถ€์ƒํ–ˆ์œผ๋ฉฐ, ๋ชจ๋ธยทํˆด ์„ ํƒ, ์ž‘์—… ๊ณ„ํš, ๊ฒฐ๊ณผ ํ†ตํ•ฉ์„ LLM์—๊ฒŒ ์œ„์ž„ํ•œ๋‹ค. ๊ธฐ์กด ์ ‘๊ทผ์€ ์ •์„ฑ์  ๋ฉ”ํƒ€๋ฐ์ดํ„ฐ (๋ชจ๋ธ ์นด๋“œ, ์ธ๊ธฐ ์ง€ํ‘œ)๋งŒ์„ ์ž…๋ ฅ์œผ๋กœ ์‚ฌ์šฉํ•ด, ์‹ค์ œ ํƒœ์Šคํฌโ€‘๋ ˆ๋ฒจ ์„ฑ๋Šฅ (์ •ํ™•๋„, ์ง€์—ฐ, ์—๋„ˆ์ง€)๊ณผ์˜ ๋ถˆ์ผ์น˜๋ฅผ ์ดˆ๋ž˜ํ•œ๋‹ค. AI ๋ฐ์ดํ„ฐ์„ผํ„ฐ์˜ ์—๋„ˆ์ง€ ์ง€์† ๊ฐ€๋Šฅ์„ฑ ๋ฌธ์ œ๊ฐ€ ๋Œ€๋‘๋˜๋Š” ์ƒํ™ฉ์—์„œ, ๋ชจ๋ธ ์„ ํƒ ๋‹จ๊ณ„์—์„œ์˜ ์—๋„ˆ์ง€ ์ตœ์ ํ™”๋Š” ์•„์ง ์ถฉ๋ถ„ํžˆ ํƒ๊ตฌ๋˜์ง€ ์•Š์€ ์˜์—ญ์ด๋‹ค. 2. ์‹คํ—˜์  ๋ถ„์„ (Section III) ํ‰๊ฐ€ ํ”Œ๋žซํผ : JARVIS(Hu

Data System Model
๋‹ค์ค‘๋ชจ๋‹ฌ ์ž ์žฌ ํ† ํฐ์œผ๋กœ ๊ฐ•ํ™”๋œ ๊ณต๊ฐ„ ์ถ”๋ก 

๋‹ค์ค‘๋ชจ๋‹ฌ ์ž ์žฌ ํ† ํฐ์œผ๋กœ ๊ฐ•ํ™”๋œ ๊ณต๊ฐ„ ์ถ”๋ก 

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ์‹œ๊ฐยท๊ณต๊ฐ„ ์ถ”๋ก ์˜ ๋‚œ์ œ : ์ธ๊ฐ„์€ ์‹œ๊ฐ ํผ์ฆ์„ ํ’€ ๋•Œ ๋‚ด๋ถ€์ ์œผ๋กœ โ€œ์ด๋ฏธ์ง€โ€์™€ โ€œ์–ธ์–ดโ€๋ฅผ ์˜ค๊ฐ€๋ฉฐ ์‚ฌ๊ณ ํ•˜์ง€๋งŒ, ํ˜„์žฌ ๋ชจ๋ธ์€ ํ…์ŠคํŠธ CoT์— ๋น„ํ•ด ์‹œ๊ฐ ์ •๋ณด๋ฅผ ์ถฉ๋ถ„ํžˆ ํ™œ์šฉํ•˜์ง€ ๋ชปํ•œ๋‹ค. ๊ธฐ์กด ์ ‘๊ทผ๋ฒ•์˜ ํ•œ๊ณ„ : ์ „๋ฌธ ๋„๊ตฌ ์˜์กด (crop, sketch ๋“ฑ) โ†’ ํŒŒ์ดํ”„๋ผ์ธ์ด ๋ณต์žกํ•˜๊ณ  ์˜ค๋ฅ˜์— ์ทจ์•ฝ. ๋Œ€๊ทœ๋ชจ ์ด๋ฏธ์ง€ ์ƒ์„ฑ (Unified VLM) โ†’ ํ•™์Šตยท์ถ”๋ก  ๋น„์šฉ์ด ๋น„๊ฒฝ์ œ์ . ๋ชจ๋‹ฌ ์ „์šฉ ํ† ํฐ (์‹œ๊ฐ ํ† ํฐ, ์—ฐ์† ์ž„๋ฒ ๋”ฉ) โ†’ ์ž‘์—…๋ณ„ ๋งž์ถค ๋ฐ์ดํ„ฐ ํ•„์š”, ์ผ๋ฐ˜ํ™” ์–ด๋ ค์›€. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด โ€“ ยulโ€‘Tokens ๋ชจ๋‹ฌ๋ฆฌํ‹ฐโ€‘ag

๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ LLM ํƒˆ์˜ฅ์˜ ์ƒˆ๋กœ์šด ์ง€ํ‰: ์ด๋ฏธ์ง€ ์Šคํ…Œ๊ฐ€๋…ธ๊ทธ๋ž˜ํ”ผ๋ฅผ ํ™œ์šฉํ•œ ์ด์ค‘ ์€๋‹‰ ๊ณต๊ฒฉ

๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ LLM ํƒˆ์˜ฅ์˜ ์ƒˆ๋กœ์šด ์ง€ํ‰: ์ด๋ฏธ์ง€ ์Šคํ…Œ๊ฐ€๋…ธ๊ทธ๋ž˜ํ”ผ๋ฅผ ํ™œ์šฉํ•œ ์ด์ค‘ ์€๋‹‰ ๊ณต๊ฒฉ

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ LLM์˜ ๊ธ‰๋ถ€์ƒ : GPTโ€‘4o, Gemini ๋“ฑ์€ ํ…์ŠคํŠธยท์ด๋ฏธ์ง€ยท์Œ์„ฑ๊นŒ์ง€ ์ฒ˜๋ฆฌํ•ด ์—์ด์ „ํŠธํ˜• AI์˜ ํ•ต์‹ฌ์ด ๋˜๊ณ  ์žˆ๋‹ค. ๋ณด์•ˆยท์•ˆ์ „์„ฑ ๋ฌธ์ œ : ์ •๋ ฌ(SFT, RLHF, RLAIF)์—๋„ ๋ถˆ๊ตฌํ•˜๊ณ  ํ…์ŠคํŠธ ๊ธฐ๋ฐ˜ ํƒˆ์˜ฅ์ด ๊ฐ€๋Šฅํ•จ์ด ์ž…์ฆ๋ผ ์™”๋‹ค. ์ƒ์šฉ ์‹œ์Šคํ…œ์˜ ๋ฐฉ์–ด ํ•œ๊ณ„ : ํ…์ŠคํŠธยท์ด๋ฏธ์ง€ ํ•„ํ„ฐ๋Š” โ€œ๋ช…์‹œ์  ์•…์„ฑ ์ฝ˜ํ…์ธ โ€๋ฅผ ํƒ์ง€ํ•˜๋„๋ก ์„ค๊ณ„๋ผ ์žˆ์–ด, ์€๋‹‰๋œ(๋น„๊ฐ€์‹œ์ ) ์•…์„ฑ ์ •๋ณด๋ฅผ ๋†“์น˜๊ธฐ ์‰ฝ๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด โ€“ โ€œ์ด์ค‘ ์Šคํ…Œ๊ฐ€๋…ธ๊ทธ๋ž˜ํ”ผโ€ | ๋‹จ๊ณ„ | ๋‚ด์šฉ | ํ•ต์‹ฌ ๊ธฐ์ˆ  | | | | | | 1. ์•…์˜์  ์ฟผ๋ฆฌ ์ธ์ฝ”๋”ฉ

๋ฉ€ํ‹ฐ์ด์„ฑ ํ†ตํ•ฉ ํŒ๋ณ„ ์ถ”๋ก  ํ”„๋ ˆ์ž„์›Œํฌ MIND

๋ฉ€ํ‹ฐ์ด์„ฑ ํ†ตํ•ฉ ํŒ๋ณ„ ์ถ”๋ก  ํ”„๋ ˆ์ž„์›Œํฌ MIND

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ CoT(MCoT) ์—ฐ๊ตฌ๋Š” ๋‹จ๊ณ„๋ณ„ ์‚ฌ๊ณ  ํ๋ฆ„์„ ์œ ๋„ํ•ด ๋…ผ๋ฆฌ ํˆฌ๋ช…์„ฑ์„ ๋†’์˜€์ง€๋งŒ, ๋‹จ์ผ ๊ทผ๊ฑฐ(supervision) ์— ์˜์กดํ•ด ๋‹ค์–‘์„ฑยท์ž๊ธฐ ๊ต์ • ๋Šฅ๋ ฅ ์ด ๋ถ€์กฑํ–ˆ๋‹ค. ์ธ๊ฐ„์€ ๋‹ค์–‘ํ•œ ๊ทผ๊ฑฐ๋ฅผ ๋น„๊ตยท๋ฐ˜์ฆ ํ•˜๋ฉฐ ์‚ฌ๊ณ ๋ฅผ ๋‹ค๋“ฌ๋Š”๋‹ค. ์ €์ž๋“ค์€ ์ด๋ฅผ ๋ชจ๋ธ์— ์ ์šฉํ•˜๊ธฐ ์œ„ํ•ด ๋‹ค์ค‘โ€‘๊ทผ๊ฑฐ(positive + negative) ํ•™์Šต์„ ์„ค๊ณ„ํ–ˆ๋‹ค. 2. ํ•ต์‹ฌ ๊ธฐ๋ฒ• | ๊ตฌ์„ฑ ์š”์†Œ | ํ•ต์‹ฌ ์•„์ด๋””์–ด | ๊ตฌํ˜„ ๋ฐฉ์‹ | | | | | | RAD | ๊ธ์ •ยท๋ถ€์ • ๊ทผ๊ฑฐ๋ฅผ ๋™์‹œ ์ƒ์„ฑ ํ•ด ๋ฐ์ดํ„ฐ ๋‹ค์–‘์„ฑ ํ™•๋ณด | LLM ๊ธฐ๋ฐ˜ Coโ€‘generation

์‹œ๊ฐ„ ์˜ˆ์ธก์„ ์œ„ํ•œ ํ†ตํ•ฉ ์ธ์ฝ”๋” ๋””์ฝ”๋” ํ”„๋ ˆ์ž„์›Œํฌ TIMEPERCEIVER

์‹œ๊ฐ„ ์˜ˆ์ธก์„ ์œ„ํ•œ ํ†ตํ•ฉ ์ธ์ฝ”๋” ๋””์ฝ”๋” ํ”„๋ ˆ์ž„์›Œํฌ TIMEPERCEIVER

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ์‹œ๊ณ„์—ด ์˜ˆ์ธก์€ ๋‚ ์”จ, ์ „๋ ฅ, ๊ตํ†ต ๋“ฑ ์‹ค์ƒํ™œ์— ํ•„์ˆ˜์ ์ธ ๋ฌธ์ œ์ด๋ฉฐ, ๋ณต์žกํ•œ ์‹œ๊ณ„์—ดยท๋‹ค๋ณ€๋Ÿ‰ ์ƒํ˜ธ์ž‘์šฉ์„ ํšจ๊ณผ์ ์œผ๋กœ ๋ชจ๋ธ๋งํ•˜๋Š” ๊ฒƒ์ด ํ•ต์‹ฌ์ด๋‹ค. ์ตœ๊ทผ Transformer ๊ธฐ๋ฐ˜ ๋ชจ๋ธ, CNN, MLP, SSM ๋“ฑ ๋‹ค์–‘ํ•œ ์ธ์ฝ”๋” ์„ค๊ณ„๊ฐ€ ์ œ์•ˆ๋์ง€๋งŒ, 1) ๋””์ฝ”๋” ์„ค๊ณ„๊ฐ€ ๋‹จ์ˆœ (์„ ํ˜• ํˆฌ์‚ฌ) 2) ํ•™์Šต ์ „๋žต์ด ์‚ฌ์ „ํ•™์Šตโ€‘๋ฏธ์„ธ์กฐ์ • ํ˜•ํƒœ๋กœ ์ธ์ฝ”๋”์™€ ๋ถˆ์ผ์น˜ ์™€ ๊ฐ™์€ ๊ตฌ์กฐโ€‘ํ•™์Šต ๋ถˆ์ผ์น˜๊ฐ€ ์กด์žฌํ•œ๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด ์˜ˆ์ธก ๋ชฉํ‘œ์˜ ์ผ๋ฐ˜ํ™” : ์ž…๋ ฅ ์ธ๋ฑ์Šค I์™€ ๋ชฉํ‘œ ์ธ๋ฑ์Šค J๋ฅผ ์ž์œ ๋กญ๊ฒŒ ์ง€์ •ํ•จ์œผ๋กœ์จ extrapolation, inter

ํ•ญ๊ณต ์—ฐ๋ฃŒํƒฑํฌ ์ด๋ฌผ์งˆ ํƒ์ง€๋ฅผ ์œ„ํ•œ ํ•ฉ์„ฑ ์‹ค์ œ ๋ฐ์ดํ„ฐ์…‹ FOD S2R

ํ•ญ๊ณต ์—ฐ๋ฃŒํƒฑํฌ ์ด๋ฌผ์งˆ ํƒ์ง€๋ฅผ ์œ„ํ•œ ํ•ฉ์„ฑ ์‹ค์ œ ๋ฐ์ดํ„ฐ์…‹ FOD S2R

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ์•ˆ์ „์„ฑ : ์—ฐ๋ฃŒํƒฑํฌ ๋‚ด๋ถ€์— ์ž‘์€ ๊ธˆ์† ํŒŒํŽธยท๋ณผํŠธยท๊ณต๊ตฌ ๋“ฑ์ด ์œ ์ž…๋˜๋ฉด ์—ฐ๋ฃŒ ์˜ค์—ผยท์—”์ง„ ๊ณ ์žฅ ๋“ฑ ์น˜๋ช…์  ์‚ฌ๊ณ  ์œ„ํ—˜์ด ์žˆ๋‹ค. ๋ฐ์ดํ„ฐ ๋ถ€์กฑ : ๊ธฐ์กด FOD ๋ฐ์ดํ„ฐ์…‹์€ ํ™œ์ฃผ๋กœยท๊ณตํ•ญ ์ฃผ๋ณ€ ๋“ฑ ๊ฐœ๋ฐฉํ˜• ํ™˜๊ฒฝ์— ๊ตญํ•œ๋ผ, ์กฐ๋ช…ยท๋ฐ˜์‚ฌยท๋ณต์žกํ•œ ๋‚ด๋ถ€ ๊ตฌ์กฐ๋ฅผ ๋ฐ˜์˜ํ•˜์ง€ ๋ชปํ•œ๋‹ค. ์ˆ˜์ง‘ยท๋ผ๋ฒจ๋ง ๋น„์šฉ : ์‹ค์ œ ์—ฐ๋ฃŒํƒฑํฌ๋Š” ๊ณ ์••ยท๊ณ ์œ„ํ—˜ ํ™˜๊ฒฝ์ด๋ฏ€๋กœ ์ด๋ฏธ์ง€ ํš๋“ยท์ฃผ์„ ์ž‘์—…์ด ๋งค์šฐ ์ œํ•œ์ ์ด๋‹ค. 2. ๋ฐ์ดํ„ฐ์…‹ ์„ค๊ณ„ (FODโ€‘S2R) | ๊ตฌ๋ถ„ | ์ด๋ฏธ์ง€ ์ˆ˜ | ํด๋ž˜์Šค ์ˆ˜ | ํ‰๊ท  ์ฃผ์„ ์ˆ˜ | ์ฃผ์š” ๋ณ€์ด ์š”์†Œ | | | | | | | | ์‹ค์ œ(Real

A-QCF-Net: An Adaptive Quaternion Cross-Fusion Network for Multimodal Liver Tumor Segmentation from Unpaired Datasets

A-QCF-Net: An Adaptive Quaternion Cross-Fusion Network for Multimodal Liver Tumor Segmentation from Unpaired Datasets

| ๊ตฌ๋ถ„ | ๋‚ด์šฉ ๋ฐ ํ‰๊ฐ€ | | | | | ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ | โ€ข CT์™€ MRI๋Š” ๊ฐ๊ฐ ๊ตฌ์กฐ์ ยท์กฐ์ง์  ์ •๋ณด๋ฅผ ์ œ๊ณตํ•˜์ง€๋งŒ, ๋‹จ์ผ ๋ชจ๋‹ฌ๋งŒ์œผ๋กœ๋Š” ๊ฐ„์•” ์ข…์–‘์„ ์™„์ „ํ•˜๊ฒŒ ํŒŒ์•…ํ•˜๊ธฐ ์–ด๋ ต๋‹ค.<br>โ€ข ๊ธฐ์กด ๋‹ค์ค‘๋ชจ๋‹ฌ ์—ฐ๊ตฌ๋Š” ๋Œ€๋ถ€๋ถ„ ์ง์„ ์ด๋ฃฌ ์ •๋ ฌ ๋ฐ์ดํ„ฐ ์— ์˜์กดํ•ด ์™”์œผ๋ฉฐ, ์‹ค์ œ ์ž„์ƒ์—์„œ๋Š” ์ด๋Ÿฐ ๋ฐ์ดํ„ฐ๊ฐ€ ๋“œ๋ฌผ๋‹ค. ๋”ฐ๋ผ์„œ ๋น„์—ฐ๊ฒฐ ๋ฐ์ดํ„ฐ ํ™œ์šฉ ๋ฐฉ์•ˆ์ด ์ ˆ์‹คํžˆ ์š”๊ตฌ๋œ๋‹ค. | | ํ•ต์‹ฌ ์•„์ด๋””์–ด | 1. ํ€˜ํ„ฐ๋‹ˆ์–ธ ์‹ ๊ฒฝ๋ง(QNN) ์„ ์ด์šฉํ•ด ํŒŒ๋ผ๋ฏธํ„ฐ 4๋ฐฐ ์ ˆ๊ฐ๊ณผ ๋™์‹œ์— ์ฑ„๋„ ๊ฐ„ ์ƒํ˜ธ์ž‘์šฉ์„ ์ž์—ฐ์Šค๋Ÿฝ๊ฒŒ ๋ชจ๋ธ๋งํ•œ๋‹ค.<br>2. Aโ€‘QCF ๋ธ”๋ก โ€“ ๋™์  ๊ต์ฐจ

Data Network
Assessment Of Selected Trace Elements Concentration in Eleyele Lake-Water, Ibadan South Western Nigeria

Assessment Of Selected Trace Elements Concentration in Eleyele Lake-Water, Ibadan South Western Nigeria

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ชฉ์  ๊ฒฝ์ œยท์‚ฌํšŒ์  ์ค‘์š”์„ฑ : ์—˜๋ ˆ์˜๋ ˆ ํ˜ธ์ˆ˜๋Š” Ibadan ์„œ๋‚จ๋ถ€์˜ ์ฃผ์š” ์‹์ˆ˜ยท๋†์—…ยท์‚ฐ์—…์šฉ์ˆ˜์›์ด๋ฉฐ, ์–ด์—…ยท๊ด€๊ด‘ ๋“ฑ ๋‹ค์ค‘ ์ด์šฉ์ด ์ด๋ฃจ์–ด์ง„๋‹ค. ์˜ค์—ผ ์œ„ํ—˜์„ฑ : ์ฃผ๋ณ€ ๋งˆ์„์˜ ๋ฌด๋ถ„๋ณ„ํ•œ ์ƒํ™œยท์‚ฐ์—… ํ์ˆ˜ ๋ฐฐ์ถœ์ด ๋ฌผ๋ฆฌยทํ™”ํ•™ยท์ƒ๋ฌผํ•™์  ๋ณ€ํ™”๋ฅผ ์ดˆ๋ž˜ํ•œ๋‹ค๋Š” ์ ์„ ๊ฐ•์กฐ, โ€˜์ˆ˜์งˆ ๋ชจ๋‹ˆํ„ฐ๋งโ€™ ํ•„์š”์„ฑ์„ ์ œ์‹œํ•œ๋‹ค. 2. ์ƒ˜ํ”Œ๋ง ๋ฐ ์‹คํ—˜ ์„ค๊ณ„ | ํ•ญ๋ชฉ | ๋‚ด์šฉ | ํ‰๊ฐ€ | | | | | | ์‹œ๋ฃŒ ์ˆ˜ | 12๊ฐœ (์œ„์น˜๋ณ„ GPS ๊ธฐ๋ก) | ์ถฉ๋ถ„ํžˆ ๋Œ€ํ‘œ์ ์ด์ง€๋งŒ, ๊ณ„์ ˆยท์‹œ๊ฐ„ ๋ณ€๋™์„ ๋ฐ˜์˜ํ•˜๊ธฐ์—” ๋ถ€์กฑ | | ์‹œ์  | ์šฐ๊ธฐ(๋น„๊ฐ€ ๋งŽ์ด ์˜ค๋Š” ์‹œ๊ธฐ) | ์šฐ๊ธฐ

BashArena: A Control Setting for Highly Privileged AI Agents

BashArena: A Control Setting for Highly Privileged AI Agents

1. ์—ฐ๊ตฌ ๋™๊ธฐ์™€ ๋ฐฐ๊ฒฝ ๊ณ ๊ถŒํ•œ AI ์œ„ํ—˜ : ํด๋ผ์šฐ๋“œ ์ธํ”„๋ผ, CI/CD ํŒŒ์ดํ”„๋ผ์ธ, ์ž๋™ํ™”๋œ ๋ณด์•ˆ ๋„๊ตฌ ๋“ฑ์—์„œ AI๊ฐ€ ๋ฃจํŠธ ๊ถŒํ•œ์„ ๊ฐ–๊ณ  ๋™์ž‘ํ•  ๊ฒฝ์šฐ, ๋ฏธ์Šค์–ผ๋ผ์ธ๋“œ ๋ชจ๋ธ์ด ์‹œ์Šคํ…œ ์ „์ฒด๋ฅผ ์žฅ์•…ํ•˜๊ฑฐ๋‚˜ ๋ฐ์ดํ„ฐ ์œ ์ถœยท์•…์„ฑ์ฝ”๋“œ ๋ฐฐํฌ์™€ ๊ฐ™์€ ์‹ฌ๊ฐํ•œ ๊ณต๊ฒฉ์„ ์ˆ˜ํ–‰ํ•  ์ˆ˜ ์žˆ๋‹ค. AI ์ œ์–ด์˜ ํ•ต์‹ฌ ๊ณผ์ œ : โ€œ์•…์˜์  ํ–‰๋™์„ ์ฐจ๋‹จํ•˜๋ฉด์„œ ์ •์ƒ ์ž‘์—…์„ ๋ฐฉํ•ดํ•˜์ง€ ์•Š๊ฒŒ ํ•˜๋Š”โ€ ์ œ์–ด ํ”„๋กœํ† ์ฝœ์„ ์„ค๊ณ„ยท๊ฒ€์ฆํ•  ์‹คํ—˜ ํ™˜๊ฒฝ์ด ํ•„์š”ํ•˜๋‹ค. ๊ธฐ์กด ํ™˜๊ฒฝ(APPS backdooring, BashBench, SHADEโ€‘Arena)์€ ํ–‰๋™ ๋‹จ๊ณ„๊ฐ€ ์ œํ•œ์ ์ด๊ฑฐ๋‚˜ ๊ณต๊ฒฉ ํ‘œ๋ฉด์ด ์ข์•„ ํ˜„์‹ค

TF-MCL: Time-frequency Fusion and Multi-domain Cross-Loss for Self-supervised Depression Detection

TF-MCL: Time-frequency Fusion and Multi-domain Cross-Loss for Self-supervised Depression Detection

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ๋ผ๋ฒจ๋ง ๋น„์šฉ ๋ฌธ์ œ : MDD ์ง„๋‹จ์€ ์ •์‹ ๊ฑด๊ฐ• ์ „๋ฌธ๊ฐ€์˜ ์ˆ˜์ž‘์—…์ด ํ•„์ˆ˜์ด๋ฉฐ, ๋Œ€๊ทœ๋ชจ EEG ๋ฐ์ดํ„ฐ์— ๋ผ๋ฒจ์„ ๋ถ€์—ฌํ•˜๊ธฐ ์–ด๋ ต๋‹ค. ๋Œ€๋น„ ํ•™์Šต์˜ ํ•œ๊ณ„ : ๊ธฐ์กด ์‹œ๊ณ„์—ด ๋Œ€๋น„ ํ•™์Šต(TSTCC, TS2Vec ๋“ฑ)์€ ์‹œ๊ฐ„์ถ•์— ์ดˆ์ ์„ ๋งž์ถ”๊ณ  ์ฃผํŒŒ์ˆ˜ ์ •๋ณด๋ฅผ ์ถฉ๋ถ„ํžˆ ํ™œ์šฉํ•˜์ง€ ๋ชปํ•œ๋‹ค. EEG๋Š” ํŠนํžˆ ์ „๋ ฅ ์ŠคํŽ™ํŠธ๋Ÿผ(PSD) ๋“ฑ ์ฃผํŒŒ์ˆ˜ ํŠน์„ฑ์ด ์ค‘์š”ํ•œ๋ฐ, ์ด๋ฅผ ๋ฌด์‹œํ•˜๋ฉด ํ‘œํ˜„๋ ฅ์ด ์ €ํ•˜๋œ๋‹ค. 2. ํ•ต์‹ฌ ๊ธฐ์—ฌ | ๋ฒˆํ˜ธ | ๊ธฐ์—ฌ ๋‚ด์šฉ | ๊ธฐ์ˆ ์  ํ•ต์‹ฌ | | | | | | โ‘  | FMH ๋„์ž…์œผ๋กœ ์‹œ๊ฐ„โ€‘์ฃผํŒŒ์ˆ˜ ํŠน์ง•์„ ํ•˜๋‚˜์˜ ํ†ตํ•ฉ ๊ณต๊ฐ„์œผ๋กœ ๋งคํ•‘ |

Detection
Fractional Differential Equation Physics-Informed Neural Network and Its Application in Battery State Estimation

Fractional Differential Equation Physics-Informed Neural Network and Its Application in Battery State Estimation

1๏ธโƒฃ ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ๋ฐฐํ„ฐ๋ฆฌ ๊ด€๋ฆฌ ์‹œ์Šคํ…œ(BMS) ์—์„œ SOC ์ถ”์ •์€ ์ „๊ธฐ์ฐจยทESSยทํœด๋Œ€๊ธฐ๊ธฐ ๋“ฑ ๋ชจ๋“  ์‘์šฉ ๋ถ„์•ผ์˜ ํ•ต์‹ฌ. ๊ธฐ์กด ๋ชจ๋ธ ๊ธฐ๋ฐ˜ (Ampereโ€‘hour, OCV, ECM, P2D ๋“ฑ)๊ณผ ๋ฐ์ดํ„ฐ ๊ธฐ๋ฐ˜ (MLP, RNN, LSTM ๋“ฑ) ๋ฐฉ๋ฒ•์€ ๊ฐ๊ฐ ์ •ํ™•๋„ยท์‹ค์‹œ๊ฐ„์„ฑ ํ˜น์€ ํ•ด์„ ๊ฐ€๋Šฅ์„ฑยท์ผ๋ฐ˜ํ™” ๋ฌธ์ œ๋ฅผ ์•ˆ๊ณ  ์žˆ๋‹ค. ํŠนํžˆ ๋ฉ”๋ชจ๋ฆฌ ํšจ๊ณผ(์ „ํ•ด์งˆ ํ™•์‚ฐยท์ „๊ทน ์ด์ค‘์ธต) ๋ฅผ ์ •๋Ÿ‰ํ™”ํ•˜๊ธฐ ์–ด๋ ค์›Œ ์ €์˜จยท๊ณ ์ „๋ฅ˜ ๊ตฌ๊ฐ„์—์„œ ์˜ค์ฐจ๊ฐ€ ๊ธ‰์ฆํ•œ๋‹ค. 2๏ธโƒฃ ํ•ต์‹ฌ ์•„์ด๋””์–ด โ€“ FDIFFโ€‘PINN | ์š”์†Œ | ์„ค๋ช… | ๊ธฐ๋Œ€ ํšจ๊ณผ | | | | | | ๋ถ„์ˆ˜ ์ฐจ์ˆ˜

Network
Graph Distance as Surprise: Free Energy Minimization in Knowledge Graph Reasoning

Graph Distance as Surprise: Free Energy Minimization in Knowledge Graph Reasoning

1. ์ด๋ก ์  ๊ธฐ์—ฌ์™€ ๊ฐ•์  | ๊ตฌ๋ถ„ | ๋‚ด์šฉ | ์˜์˜ | | | | | | FEP์™€ KG์˜ ์—ฐ๊ฒฐ | KG๋ฅผ ์ƒ์„ฑ ๋ชจ๋ธ ๋กœ ํ•ด์„ํ•˜๊ณ , ( log P(e|C) propto d {mathcal G}(C,e)) ๋กœ ๋งคํ•‘ | ์‹ ๊ฒฝ๊ณผํ•™ยท์ธ๊ณต์ง€๋Šฅ ๊ฐ„ ์ด๋ก ์  ๋‹ค๋ฆฌ ๊ตฌ์ถ•, ๊ธฐ์กด FEP ์ ์šฉ ๋ฒ”์œ„๋ฅผ ์‹œ๋งจํ‹ฑ ์˜์—ญ์œผ๋กœ ํ™•์žฅ | | ๋†€๋ผ์›€์˜ ๊ทธ๋ž˜ํ”„ ๊ฑฐ๋ฆฌ ์ •์˜ | ์ตœ๋‹จ ๊ฒฝ๋กœ ๊ฑฐ๋ฆฌ + ์—ฐ๊ฒฐ ์—†์Œ์— ๋Œ€ํ•œ ํ•˜์ดํผํŒŒ๋ผ๋ฏธํ„ฐ (alpha) | ์ง๊ด€์ ์ด๋ฉฐ ๊ณ„์‚ฐ ๋น„์šฉ์ด ๋‚ฎ์Œ (BFS O(|V|+|E|)). ์‚ฌ์ดํดยท๋‹ค์ค‘ ๊ฒฝ๋กœ๋ฅผ ์ž์—ฐ์Šค๋Ÿฝ๊ฒŒ ์ฒ˜๋ฆฌ | | Kolmo

Hidden Leaks in Time Series Forecasting: How Data Leakage Affects LSTM Evaluation Across Configurations and Validation Strategies

Hidden Leaks in Time Series Forecasting: How Data Leakage Affects LSTM Evaluation Across Configurations and Validation Strategies

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ์‹œ๊ณ„์—ด ์˜ˆ์ธก์—์„œ LSTM ์€ ๋ณต์žกํ•œ ์‹œ๊ฐ„ ์˜์กด์„ฑ์„ ํ•™์Šตํ•˜๋Š” ๋ฐ ๊ฐ•์ ์ด ์žˆ์ง€๋งŒ, ํ‰๊ฐ€ ์‹ ๋ขฐ์„ฑ ์ด ๋ชจ๋ธ ๋ฐฐํฌ์— ์ง์ ‘์ ์ธ ์˜ํ–ฅ์„ ๋ฏธ์นœ๋‹ค. ๊ธฐ์กด ์—ฐ๊ตฌ๋“ค์€ ๋ชจ๋ธ ๊ตฌ์กฐยทํ•˜์ดํผํŒŒ๋ผ๋ฏธํ„ฐ ํŠœ๋‹์— ์ง‘์ค‘ํ–ˆ์œผ๋ฉฐ, ๊ฒ€์ฆ ์„ค๊ณ„์™€ ๋ฐ์ดํ„ฐ ๋ˆ„์ˆ˜ ๊ฐ„ ์ƒํ˜ธ์ž‘์šฉ์„ ์ถฉ๋ถ„ํžˆ ํƒ๊ตฌํ•˜์ง€ ์•Š์•˜๋‹ค. ๋ณธ ๋…ผ๋ฌธ์€ โ€œ ๊ฒ€์ฆ ์ž์ฒด๋ฅผ ํ…Œ์ŠคํŠธ ๊ฐ€๋Šฅํ•œ ์ปดํฌ๋„ŒํŠธ โ€๋กœ ๊ฐ„์ฃผํ•˜๊ณ , ๋ˆ„์ˆ˜๊ฐ€ ํ‰๊ฐ€์— ๋ฏธ์น˜๋Š” ์™œ๊ณก์„ ์ •๋Ÿ‰ํ™”ํ•จ์œผ๋กœ์จ ์ด ๊ฒฉ์ฐจ๋ฅผ ๋ฉ”์šด๋‹ค. 2. ์‹คํ—˜ ์„ค๊ณ„ | ์š”์†Œ | ์„ค๋ช… | | | | | ๋ฐ์ดํ„ฐ | Daily Climate dataset (1,462 ์ผ, 201

Data
MMCTOP: A Multimodal Textualization and Mixture-of-Experts Framework for Clinical Trial Outcome Prediction

MMCTOP: A Multimodal Textualization and Mixture-of-Experts Framework for Clinical Trial Outcome Prediction

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ์ž„์ƒ์‹œํ—˜ ์„ฑ๊ณต๋ฅ  ์ €์กฐ ์™€ ๋†’์€ ๋น„์šฉ ๋ฌธ์ œ๋Š” ์ œ์•ฝยท๋ฐ”์ด์˜ค์‚ฐ์—… ์ „๋ฐ˜์— ํฐ ๋ถ€๋‹ด์„ ์ฃผ๋ฉฐ, ์„ฑ๊ณต ์˜ˆ์ธก ๋ชจ๋ธ์˜ ํ•„์š”์„ฑ์ด ๊ธ‰์ฆํ•˜๊ณ  ์žˆ๋‹ค. ๊ธฐ์กด ์˜ˆ์ธก ๋ชจ๋ธ์€ ๋‹จ์ผ ๋ชจ๋‹ฌ๋ฆฌํ‹ฐ(๋ถ„์ž ๊ตฌ์กฐ, PK/PD, ์„ค๊ณ„ ๋ณ€์ˆ˜ ๋“ฑ) ์— ์˜์กดํ•˜๊ฑฐ๋‚˜, ๊ณ ์ •๋œ ์œตํ•ฉ ํŒŒ์ดํ”„๋ผ์ธ ์„ ์‚ฌ์šฉํ•ด ์ƒˆ๋กœ์šด ๋ฐ์ดํ„ฐ ์œ ํ˜•์„ ์ˆ˜์šฉํ•˜๊ธฐ ์–ด๋ ค์› ๋‹ค. ๋ฐ”์ด์˜ค๋ฉ”๋””์ปฌ ๋ฐ์ดํ„ฐ๋Š” ๊ทธ๋ž˜ํ”„(๋ถ„์ž), ๊ณ„์ธต(์˜จํ†จ๋กœ์ง€), ์žฅ๋ฌธ ํ…์ŠคํŠธ(ํ”„๋กœํ† ์ฝœ) ๋“ฑ ํ˜•ํƒœ๊ฐ€ ๋‹ค์–‘ํ•ด ํ‘œํ˜„ ๋ถˆ์ผ์น˜ ์™€ ํ”„๋ผ์ด๋ฒ„์‹œยท๋ณด์•ˆ ์ด์Šˆ๊ฐ€ ๋™๋ฐ˜๋œ๋‹ค. 2. ํ•ต์‹ฌ ๊ธฐ์—ฌ | ๋ฒˆํ˜ธ | ๋‚ด์šฉ | ์˜์˜ | | | | | |โ‘ | ์Šคํ‚ค๋งˆ

Framework
A Model of Causal Explanation on Neural Networks for Tabular Data

A Model of Causal Explanation on Neural Networks for Tabular Data

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ํ•ด์„ ๊ฐ€๋Šฅ์„ฑ vs. ์„ฑ๋Šฅ : NNยทGBM์€ ๋†’์€ ์„ฑ๋Šฅ์„ ๋ณด์ด์ง€๋งŒ, ํ•ด์„ ๊ฐ€๋Šฅ์„ฑ์€ ๋‚ฎ์•„ ์‹ค์ œ ๋น„์ฆˆ๋‹ˆ์Šคยท์˜๋ฃŒ ํ˜„์žฅ์—์„œ ํ™œ์šฉ์— ์ œ์•ฝ์ด ์žˆ๋‹ค. ์ธ๊ณผ์„ฑ ๊ฒฐ์—ฌ : ๊ธฐ์กด LIME, SHAP ๋“ฑ์€ ๋ณ€์ˆ˜์˜ ์ƒ๊ด€๊ด€๊ณ„ ๋งŒ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ์„ค๋ช…ํ•œ๋‹ค. ์ด๋Š” ์‚ฌ์šฉ์ž๊ฐ€ ๋ณ€์ˆ˜์™€ ๊ฒฐ๊ณผ ์‚ฌ์ด๋ฅผ ์ธ๊ณผ๊ด€๊ณ„๋กœ ์˜คํ•ดํ•˜๊ฒŒ ๋งŒ๋“ค๋ฉฐ, ํŠนํžˆ ๊ฐ€์งœ ์ƒ๊ด€ ์ด ์กด์žฌํ•  ๊ฒฝ์šฐ ์ž˜๋ชป๋œ ์˜์‚ฌ๊ฒฐ์ •์„ ์ดˆ๋ž˜ํ•œ๋‹ค. ๋น„๊ฐ€์‚ฐ์„ฑ(Nonโ€‘additivity) : ์‹ค์ œ ํ˜„์ƒ์€ ์—ฌ๋Ÿฌ ์š”์ธ์˜ ๋ณตํ•ฉ์  ์ƒํ˜ธ์ž‘์šฉ์— ์˜ํ•ด ๋ฐœ์ƒํ•œ๋‹ค. ๊ธฐ์กด ๋ฐฉ๋ฒ•์€ ๊ฐ€์‚ฐ์„ฑ์„ ์ „์ œ๋กœ ํ•˜์—ฌ ๋ณตํ•ฉ ์š”์ธ์„ ์ถฉ๋ถ„ํžˆ ํฌ์ฐฉํ•˜์ง€

Data Network Model
๊ธฐํ›„ยท์บ˜๋ฆฐ๋” ๋ณ€์ˆ˜์™€ ์ธ๊ณผ๊ด€๊ณ„๋ฅผ ๊ฒฐํ•ฉํ•œ ์—๋„ˆ์ง€ ์ˆ˜์š” ์˜ˆ์ธก ๋ชจ๋ธ

๊ธฐํ›„ยท์บ˜๋ฆฐ๋” ๋ณ€์ˆ˜์™€ ์ธ๊ณผ๊ด€๊ณ„๋ฅผ ๊ฒฐํ•ฉํ•œ ์—๋„ˆ์ง€ ์ˆ˜์š” ์˜ˆ์ธก ๋ชจ๋ธ

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ๊ธฐ์กด ๋จธ์‹ ๋Ÿฌ๋‹ ๊ธฐ๋ฐ˜ ์ „๋ ฅ ์ˆ˜์š” ์˜ˆ์ธก์€ ์ƒ๊ด€๊ด€๊ณ„ ์— ์˜์กดํ•ด ์ธ๊ณผ์  ๋ฉ”์ปค๋‹ˆ์ฆ˜์„ ๋ฌด์‹œํ•œ๋‹ค. ์ธ๊ณผ๊ด€๊ณ„๋ฅผ ๋ฌด์‹œํ•˜๋ฉด confounder bias (์˜ˆ: ์‹œ๊ฐ„ยท์›”์ด ์Šต๋„์™€ ์ˆ˜์š”๋ฅผ ๋™์‹œ์— ์˜ํ–ฅ์„ ๋ฏธ์นจ)์™€ ๋ถ„ํฌ ์ด๋™ ์ƒํ™ฉ์—์„œ์˜ ์ผ๋ฐ˜ํ™” ์‹คํŒจ๊ฐ€ ๋ฐœ์ƒํ•œ๋‹ค. ์ €์ž๋Š” Pearl์˜ ์ธ๊ณผ ํ”„๋ ˆ์ž„์›Œํฌ๋ฅผ ์ ์šฉํ•ด ์ด๋Ÿฌํ•œ ๋ฌธ์ œ๋ฅผ ๊ทผ๋ณธ์ ์œผ๋กœ ํ•ด๊ฒฐํ•˜๊ณ ์ž ํ•œ๋‹ค. 2. ๋ฐ์ดํ„ฐ ๋ฐ ๋ณ€์ˆ˜ ์ •์˜ | ๊ตฌ๋ถ„ | ๋ณ€์ˆ˜ | ์„ค๋ช… | | | | | | ์บ˜๋ฆฐ๋” | hour of day, month of year | ์ผ์ผยท์—ฐ๊ฐ„ ์ฃผ๊ธฐ์„ฑ์„ ํฌ์ฐฉ | | ๊ธฐํ›„ | temp

No Image

A Systematic Analysis of Biases in Large Language Models

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ์˜์˜ ํ•„์š”์„ฑ : LLM์ด ์ผ์ƒยท์ „๋ฌธ ์˜์‚ฌ๊ฒฐ์ •์— ๊นŠ์ˆ™์ด ์นจํˆฌํ•จ์— ๋”ฐ๋ผ ์‚ฌ์šฉ์ž๋Š” ๋ชจ๋ธ์ด โ€œ์ค‘๋ฆฝยท๊ณต์ •โ€ํ•˜๋‹ค๊ณ  ๊ฐ€์ •ํ•œ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ๊ธฐ์กด ์—ฐ๊ตฌ๋“ค์€ ์„ฑ๋ณ„ยท์ธ์ข…ยท๋ฌธํ™” ๋“ฑ ํŠน์ • ํŽธํ–ฅ๋งŒ์„ ์ง‘์ค‘์ ์œผ๋กœ ๋‹ค๋ฃจ์—ˆ์œผ๋ฉฐ, ์ •์น˜ยท์ด๋…ยท๋™๋งนยท์–ธ์–ดยท์„ฑ๋ณ„์„ ๋™์‹œ์— ํฌ๊ด„ํ•˜๋Š” ๋ถ„์„์€ ๋ถ€์กฑํ–ˆ๋‹ค. ๊ธฐ์—ฌ : ๋ณธ ๋…ผ๋ฌธ์€ ๋‹ค์„ฏ ์ถ•์„ ๋™์‹œ์— ํƒ์ƒ‰ํ•จ์œผ๋กœ์จ LLM ํŽธํ–ฅ ์—ฐ๊ตฌ์˜ ์ŠคํŽ™ํŠธ๋Ÿผ์„ ํ™•์žฅํ•˜๊ณ , ํ์‡„ํ˜•ยท๊ณ ๋น„์šฉ ๋ชจ๋ธ์— ๋Œ€ํ•œ ์ ‘๊ทผ์„ฑ์„ ๋†’์ด๋Š” ์‹คํ—˜ ํ”„๋ ˆ์ž„์›Œํฌ๋ฅผ ์ œ๊ณตํ•œ๋‹ค. 2. ์‹คํ—˜ ์„ค๊ณ„ ๋ฐ ๋ฐฉ๋ฒ•๋ก  | ์ฐจ์› | ์‹คํ—˜ ๊ณผ์ œ | ๋ฐ์ดํ„ฐยท์ƒ˜ํ”Œ ์ˆ˜ | ํ‰๊ฐ€ ์ง€ํ‘œ | | | |

System Analysis Model
AI ์ž๊ฐ€๋ชจ๋‹ˆํ„ฐ๋ง ๋ถ€์žฌ์™€ ์ธ์ง€ ์ž์œจ์„ฑ ํ™•๋ณด๋ฅผ ์œ„ํ•œ ์ผ๊ณฑ ๊ฐ€์ง€ ํ•ต์‹ฌ ๊ฒฐํ•จ

AI ์ž๊ฐ€๋ชจ๋‹ˆํ„ฐ๋ง ๋ถ€์žฌ์™€ ์ธ์ง€ ์ž์œจ์„ฑ ํ™•๋ณด๋ฅผ ์œ„ํ•œ ์ผ๊ณฑ ๊ฐ€์ง€ ํ•ต์‹ฌ ๊ฒฐํ•จ

1. ์—ฐ๊ตฌ์˜ ํ•ต์‹ฌ ๊ธฐ์—ฌ | ๋ฒˆํ˜ธ | ๊ฒฐํ•จ(ํ•ต์‹ฌ ์šฉ์–ด) | ๋…ผ๋ฌธ์—์„œ ์ œ์‹œ๋œ ์ฃผ์š” ์ฃผ์žฅ | ์™œ ์ค‘์š”ํ•œ๊ฐ€? | | | | | | | 1 | ์ž๊ธฐโ€‘๋ชจ๋‹ˆํ„ฐ๋ง(Selfโ€‘Monitoring) | ํ˜„์žฌ ๋ชจ๋ธ์€ ๋‚ด๋ถ€ ๋ถˆ์ผ์น˜ยท๋ถˆํ™•์‹ค์„ฑ์„ ์Šค์Šค๋กœ ๊ฐ์ง€ํ•˜์ง€ ๋ชปํ•œ๋‹ค. | ์˜ค๋ฅ˜ยทํ™˜๊ฐ์„ ์‚ฌ์ „์— ์ฐจ๋‹จํ•˜๊ณ , ์‹ ๋ขฐ์„ฑ ๋†’์€ ์„œ๋น„์Šค ์ œ๊ณต์— ํ•„์ˆ˜. | | 2 | ๋ฉ”ํƒ€โ€‘์ธ์ง€(Metaโ€‘Cognitive Awareness) | ๋ชจ๋ธ์ด โ€œ๋‚ด๊ฐ€ ๋ฌด์—‡์„ ๋ชจ๋ฅด๋Š”๊ฐ€โ€๋ฅผ ์ถ”๋ก ยทํ‘œํ˜„ํ•˜์ง€ ๋ชปํ•œ๋‹ค. | ํ•™์Šต ๋ชฉํ‘œ๋ฅผ ์Šค์Šค๋กœ ์„ค์ •ยท์กฐ์ •ํ•˜๋Š” โ€˜์ž๊ธฐโ€‘์ฃผ๋„ ํ•™์Šตโ€™์˜ ์ „์ œ. | | 3 | ์ ์‘ ํ•™์Šต ๊ทœ

The Agentic Leash: Extracting Causal Feedback Fuzzy Cognitive Maps with LLMs

The Agentic Leash: Extracting Causal Feedback Fuzzy Cognitive Maps with LLMs

1. ์—ฐ๊ตฌ์˜ ํ˜์‹ ์„ฑ | ์ธก๋ฉด | ๊ธฐ์กด ์ ‘๊ทผ | ๋ณธ ๋…ผ๋ฌธ์˜ ์ฐจ๋ณ„์  | | | | | | ๋ฐ์ดํ„ฐ ๋ผ๋ฒจ๋ง | ์ง€๋„ํ•™์Šต์„ ์œ„ํ•ด ์ธ๊ฐ„์ด ์ง์ ‘ (source, target, direction) ํŠœํ”Œ์„ ์ฃผ์„ | ์ฃผ์„ ์—†์ด ํ…์ŠคํŠธ ์ž์ฒด๋งŒ์œผ๋กœ ๋ช…์‚ฌยท๋ช…์‚ฌ๊ตฌ โ†’ ๋…ธ๋“œ โ†’ ์ธ๊ณผ ์—ฃ์ง€ ์ˆœ์ฐจ ์ถ”์ถœ | | ์—์ด์ „์‹œ | LLM์€ ๋‹จ์ˆœํžˆ ํ”„๋กฌํ”„ํŠธ์— ๋”ฐ๋ผ ๊ฒฐ๊ณผ๋ฅผ ์ƒ์„ฑ | ์–‘๋ฐฉํ–ฅ ํ”ผ๋“œ๋ฐฑ ๋ฃจํ”„ : FCM ํ‰ํ˜•์ด LLM์˜ ํƒ์ƒ‰ ๋ชฉํ‘œ๋ฅผ ์žฌ์กฐ์ •, โ€œ์—์ด์ „ํŠธ ๋ฆฌ์‰ฌโ€ ๊ฐœ๋… ๋„์ž… | | ๋‹ค์ค‘ ๋ชจ๋ธ ํ˜ผํ•ฉ | ๋Œ€๋ถ€๋ถ„ ๋‹จ์ผ ๋ชจ๋ธ ๊ธฐ๋ฐ˜ | Gemini๊ณผ ChatGPT๋ฅผ ๊ฐ€์ค‘ ํ‰๊ท  ํ•œ

Computer Science Artificial Intelligence
๋…ธ์ด์ฆˆ ๊ธฐ๋ฐ˜ ์•„๋ฐ”ํƒ€ ์ง€์˜ค๋ฉ”ํŠธ๋ฆฌ ์ƒ์„ฑ๊ณผ ๊ฐ€์šฐ์‹œ์•ˆ ์Šคํ”Œ๋ž˜ํŒ… ์‹œ๊ฐํ™”

๋…ธ์ด์ฆˆ ๊ธฐ๋ฐ˜ ์•„๋ฐ”ํƒ€ ์ง€์˜ค๋ฉ”ํŠธ๋ฆฌ ์ƒ์„ฑ๊ณผ ๊ฐ€์šฐ์‹œ์•ˆ ์Šคํ”Œ๋ž˜ํŒ… ์‹œ๊ฐํ™”

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ์„ธ๋ถ€ ๋””ํ…Œ์ผ ์š”๊ตฌ : ์˜ท ์ฃผ๋ฆ„ยท์ฃผ๋™ ๊ฐ™์€ ๊ณ ์ฃผํŒŒ ๊ตฌ์กฐ๋ฅผ ์žฌํ˜„ํ•˜๋ ค๋ฉด ์ ๋ฐ€ํ•œ ํ‘œ๋ฉด ํ‘œํ˜„์ด ํ•„์š”ํ•˜์ง€๋งŒ, ๊ธฐ์กด์˜ ํ…œํ”Œ๋ฆฟ ๊ธฐ๋ฐ˜ยทimplicit ๋ฐฉ์‹์€ ํ•ด์ƒ๋„ยท๋ฐ€๋„ ํ•œ๊ณ„๊ฐ€ ์žˆ๋‹ค. ๋ฐ์ดํ„ฐ ๋ถ€์กฑ : 3D ์ธ๊ฐ„ ์• ๋‹ˆ๋ฉ”์ด์…˜ ๋ฐ์ดํ„ฐ๋Š” ์ˆ˜์ง‘ ๋น„์šฉ์ด ๋†’์•„ ๊ณผ์ ํ•ฉ ์œ„ํ—˜์ด ํฌ๋‹ค. ๋”ฐ๋ผ์„œ ๋ฐ์ดํ„ฐ ํšจ์œจ์ ์ธ ๋ชจ๋ธ๋ง์ด ํ•ต์‹ฌ ๊ณผ์ œ๋‹ค. 2. ํ•ต์‹ฌ ๊ธฐ์—ฌ | ๋ฒˆํ˜ธ | ๋‚ด์šฉ | ์˜์˜ | | | | | |โ‘ | Human Geometry Distribution (HuGeoDis) ๊ธฐ๋ฐ˜ ์ž ์žฌ ๊ณต๊ฐ„ โ€“ SMPL โ†” ์•„๋ฐ”ํƒ€ ๋งคํ•‘์„ ๊ท ์ผํ™”ํ•˜๊ณ , ํฌ์ธํŠธ ์ˆ˜๋ฅผ ํฌ

CBA: Communication-Bound-Aware Cross-Domain Resource Assignment for Pipeline-Parallel Distributed LLM Training in Dynamic Multi-DC Optical Networks

CBA: Communication-Bound-Aware Cross-Domain Resource Assignment for Pipeline-Parallel Distributed LLM Training in Dynamic Multi-DC Optical Networks

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ฌธ์ œ ์ •์˜ LLM ํ•™์Šต์˜ GPU ์š”๊ตฌ๋Ÿ‰ ์ด ๊ธ‰์ฆํ•˜๋ฉด์„œ ๋‹จ์ผ ๋ฐ์ดํ„ฐ์„ผํ„ฐ์— GPU๋ฅผ ๋ชจ๋‘ ๋ฐฐ์น˜ํ•˜๊ธฐ ์–ด๋ ค์›Œ์กŒ๋‹ค. ๊ต์ฐจโ€‘DC GPU ์ง‘ํ•ฉ ์„ ํ†ตํ•œ ๋ถ„์‚ฐ ๋จธ์‹ ๋Ÿฌ๋‹(DML)์ด ํ˜„์‹ค์ ์ธ ํ•ด๊ฒฐ์ฑ…์œผ๋กœ ๋ถ€์ƒํ–ˆ์œผ๋ฉฐ, ํŠนํžˆ ํŒŒ์ดํ”„๋ผ์ธ ๋ณ‘๋ ฌ(PP) ์€ ๋ ˆ์ด์–ด๋ฅผ GPU์— ๋‚˜๋ˆ  ๋ฐฐ์น˜ํ•˜๊ณ  ์ธ์ ‘ ๋‹จ๊ณ„ ๊ฐ„ ์ ๋Œ€์  ํ†ต์‹ ์„ ํ™œ์šฉํ•ด ํšจ์œจ์„ฑ์„ ๋†’์ธ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ DC ๊ฐ„ ๊ด‘ํ†ต์‹  ์ง€์—ฐ ์ด ๊ธธ๊ณ  ๊ฐ€๋ณ€์ ์ด์–ด์„œ GPU๊ฐ€ ๋Œ€๊ธฐ(bubble) ์ƒํƒœ ๊ฐ€ ๋นˆ๋ฒˆํžˆ ๋ฐœ์ƒํ•˜๊ณ , ๊ธฐ์กด ์—ฐ๊ตฌ๋Š” ์ปดํ“จํŒ… ์ž‘์—… ์Šค์ผ€์ค„๋ง ์— ์ง‘์ค‘ํ–ˆ์ง€๋งŒ ํ†ต์‹  ์ž์› ํ• ๋‹น ์„ ๊ฐ„๊ณผํ–ˆ๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด

Network
์ตœ๋Œ€ ์†์‹ค ๋ชฉํ‘œ ๋น„์ค‘์‹ฌ ๊ตฐ์ง‘์—์„œ ํ•ต์‹ฌ ์•ˆ์ •์„ฑ์˜ ํ•œ๊ณ„์™€ ๋ถˆ๊ฐ€๋Šฅ์„ฑ

์ตœ๋Œ€ ์†์‹ค ๋ชฉํ‘œ ๋น„์ค‘์‹ฌ ๊ตฐ์ง‘์—์„œ ํ•ต์‹ฌ ์•ˆ์ •์„ฑ์˜ ํ•œ๊ณ„์™€ ๋ถˆ๊ฐ€๋Šฅ์„ฑ

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ฌธ์ œ ์ •์˜ ๊ตฐ์ง‘ํ™”(Clustering) ๋Š” ๋ฐ์ดํ„ฐ ๋ถ„์„ยท์ตœ์ ํ™”ยท๊ธฐ๊ณ„ํ•™์Šต ๋“ฑ์—์„œ ํ•ต์‹ฌ ๋ฌธ์ œ์ด๋ฉฐ, ์ „ํ†ต์ ์œผ๋กœ๋Š” ์ค‘์‹ฌ(centroid) ๋ชจ๋ธ ์ด ์ฃผ๋ฅผ ์ด๋ฃฌ๋‹ค. ๋น„์ค‘์‹ฌ(๋Œ€ํ‘œ์ž ์—†๋Š”) ๊ตฐ์ง‘ ์€ ํŒ€ ๊ตฌ์„ฑ, ์—ฐํ•ฉ ํ•™์Šต ๋“ฑ์—์„œ ์ž์—ฐ์Šค๋Ÿฝ๊ฒŒ ๋“ฑ์žฅํ•œ๋‹ค. ์—ฌ๊ธฐ์„œ๋Š” ๊ฐ ์—์ด์ „ํŠธ์˜ ์†์‹ค์ด ๊ตฐ์ง‘ ๋‚ด ๋‹ค๋ฅธ ๋ฉค๋ฒ„์™€์˜ ๊ฑฐ๋ฆฌ ์—๋งŒ ์˜์กดํ•œ๋‹ค. ํ•ต์‹ฌ(core) ๊ฐœ๋…์€ ํ˜‘๋™์  ์ฐจ๋‹จ(coalitional blocking) ์„ ๋ฐฉ์ง€ํ•˜๋Š” ์•ˆ์ •์„ฑ ๊ธฐ์ค€์ด๋‹ค. (alpha) core๋Š” ๋ชจ๋“  ์ตœ์†Œ ๊ทœ๋ชจ (frac{n}{k}) ์ด์ƒ์˜ ์—ฐํ•ฉ์ด ๊ฐ์ž์˜ ์†์‹ค์„

INFORM-CT: INtegrating LLMs and VLMs FOR Incidental Findings Management in Abdominal CT

INFORM-CT: INtegrating LLMs and VLMs FOR Incidental Findings Management in Abdominal CT

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ | ๋ฌธ์ œ์  | ๊ธฐ์กด ์ ‘๊ทผ๋ฒ• | ํ•œ๊ณ„ | | | | | | ์šฐ๋ฐœ ์†Œ๊ฒฌ์˜ ๋นˆ๋„ยท๋‹ค์–‘์„ฑ | ๋ฐฉ์‚ฌ์„ ๊ณผ ์ˆ˜์ž‘์—… ํŒ๋… | ์‹œ๊ฐ„ ์†Œ๋ชจยทํŒ๋…์ž ๊ฐ„ ๋ณ€๋™ | | ๋‹ค๊ธฐ๊ด€(๋ณต๋ถ€ ์ „์ฒด) ๊ฒ€์‚ฌ | ์žฅ๊ธฐ๋ณ„ ์ „์šฉ ๋”ฅ๋Ÿฌ๋‹ ๋ชจ๋ธ | ๋ฒ”์šฉ์„ฑยท๊ฐ€์ด๋“œ๋ผ์ธ ์ ์šฉ ์–ด๋ ค์›€ | | ๊ฐ€์ด๋“œ๋ผ์ธ(ACR ๋“ฑ) ํ…์ŠคํŠธยทํ‘œยท๊ทธ๋ฆผ ๋ณตํ•ฉ | ์ˆ˜๋™ ๊ตฌํ˜„ | ์˜ค๋ฅ˜ยท์—…๋ฐ์ดํŠธ ๋น„์šฉ ๋†’์Œ | ๋ณธ ๋…ผ๋ฌธ์€ โ€œ์ฝ”๋“œ ์ƒ์„ฑ + ์‹คํ–‰โ€ ์ด๋ผ๋Š” ์—์ด์ „ํŠธํ˜• ์ ‘๊ทผ์„ ํ†ตํ•ด ์œ„ ํ•œ๊ณ„๋ฅผ ๋™์‹œ์— ํ•ด์†Œํ•œ๋‹ค๋Š” ์ ์—์„œ ํ˜์‹ ์ ์ด๋‹ค. 2. ํ•ต์‹ฌ ๊ธฐ์ˆ  ๊ตฌ์„ฑ | ๊ตฌ์„ฑ ์š”์†Œ | ์—ญํ•  | ์ฃผ์š” ๊ตฌํ˜„ | |

No Image

A2P-Vis: an Analyzer-to-Presenter Agentic Pipeline for Visual Insights Generation and Reporting

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ์ตœ๊ทผ LLM ๊ธฐ๋ฐ˜ ์—์ด์ „ํŠธ๊ฐ€ ๋ฐ์ดํ„ฐ ํŒŒ์ดํ”„๋ผ์ธ(์˜ˆ: Google Data Science Agent)ยทํŠน์ • ๋„๋ฉ”์ธ(์œ ์ „์ฒด, ์˜๋ฃŒ ๋“ฑ)์—์„œ ํ™œ์šฉ๋˜๊ณ  ์žˆ์œผ๋‚˜, ์‹œ๊ฐํ™” ๋‹ค์–‘์„ฑยท๋ณด๊ณ ์„œ ํ’ˆ์งˆ ์€ ์—ฌ์ „ํžˆ ์ธ๊ฐ„ ์ „๋ฌธ๊ฐ€์˜ ๊ฐœ์ž…์ด ํ•„์š”ํ–ˆ๋‹ค. ๋…ผ๋ฌธ์€ ์ด ๋‘ ๊ฒฉ์ฐจ๋ฅผ ๋™์‹œ์— ํ•ด์†Œํ•˜๋ ค๋Š” ์ตœ์ดˆ ์‹œ๋„ ์ค‘ ํ•˜๋‚˜์ด๋ฉฐ, โ€œ๋ถ„์„โ€‘ํ”„๋ ˆ์  ํ„ฐโ€๋ผ๋Š” ๋ช…ํ™•ํ•œ ์—ญํ•  ๊ตฌ๋ถ„์„ ํ†ตํ•ด ๋ชจ๋“ˆํ™”์™€ ์ฑ…์ž„ ๋ถ„๋ฆฌ ๋ฅผ ๊ตฌํ˜„ํ•œ๋‹ค๋Š” ์ ์ด ํ˜์‹ ์ ์ด๋‹ค. 2. ์‹œ์Šคํ…œ ์„ค๊ณ„ ๋ฐ ํ•ต์‹ฌ ๋ชจ๋“ˆ | ๋ชจ๋“ˆ | ์ฃผ์š” ๊ธฐ๋Šฅ | ์„ค๊ณ„์ƒ์˜ ๊ฐ•์  | | | | | | Sniffer | ๋ฐ์ดํ„ฐ ์Šคํ‚ค

TriFlow: A Progressive Multi-Agent Framework for Intelligent Trip Planning

TriFlow: A Progressive Multi-Agent Framework for Intelligent Trip Planning

1. ํ•ต์‹ฌ ๊ธฐ์—ฌ | ๋ฒˆํ˜ธ | ๋‚ด์šฉ | ์˜์˜ | | | | | | โ‘  | ์ง„ํ–‰ํ˜• 3โ€‘stage ํŒŒ์ดํ”„๋ผ์ธ (๊ฒ€์ƒ‰โ€‘๊ณ„ํšโ€‘๊ฑฐ๋ฒ„๋„Œ์Šค) ๋„์ž… | ๋ฌธ์ œ ๊ณต๊ฐ„์„ ๋‹จ๊ณ„๋ณ„๋กœ ์ถ•์†Œํ•ด ํƒ์ƒ‰ ๋น„์šฉ์„ ํฌ๊ฒŒ ๊ฐ์†Œ์‹œํ‚ค๊ณ , ๊ฐ ๋‹จ๊ณ„์—์„œ ๋…๋ฆฝ์ ์ธ ๊ฒ€์ฆ์„ ์ˆ˜ํ–‰ํ•จ์œผ๋กœ์จ ์˜ค๋ฅ˜ ์ „ํŒŒ๋ฅผ ๋ฐฉ์ง€ | | โ‘ก | Ruleโ€‘LLM ํ˜‘์—… ๊ตฌ์กฐ | ๊ทœ์น™ ๊ธฐ๋ฐ˜ ์—”์ง„์ด ์ œ์•ฝ์„ ๊ฐ•์ œํ•˜๊ณ , LLM์ด ์ž์—ฐ์–ด ์œ ์—ฐ์„ฑ์„ ์ œ๊ณตํ•ด ๋‘ ์žฅ์ ์˜ ์‹œ๋„ˆ์ง€๋ฅผ ์ฐฝ์ถœ | | โ‘ข | ๋ชจ๋“ˆํ˜• ์ œ์•ฝ ๊ฑฐ๋ฒ„๋„Œ์Šค (์‹œ๊ฐ„ยท๊ณต๊ฐ„ยท์˜ˆ์‚ฐยท์„ ํ˜ธ) | ํ”Œ๋Ÿฌ๊ทธ์ธ ํ˜•ํƒœ๋กœ ์ œ์•ฝ์„ ์ถ”๊ฐ€ยท์ˆ˜์ • ๊ฐ€๋Šฅ, ์ƒˆ๋กœ์šด ๋„๋ฉ”์ธ(์˜ˆ: ์ฐจ๋Ÿ‰ ๋ Œํƒˆ)์—๋„ ์†์‰ฝ

Framework
A short technical comment on Bubs There is No Quantum World (arXiv:2512.18400v2) and a brief remark on related Grangiers reply (arXiv:2512.22965v1)

A short technical comment on Bubs There is No Quantum World (arXiv:2512.18400v2) and a brief remark on related Grangiers reply (arXiv:2512.22965v1)

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ชฉ์  ๋ฒ„๋ธŒ๋Š” ๋‘ ๊ฐ€์ง€ ์ฃผ์š” ํ๋ฆ„์„ ์ „๊ฐœํ•œ๋‹ค. 1) ๋„ค์˜ค ๋ณดํ๋ฆฌ์•ˆ ํ•ด์„ โ€“ ๋ช…์ œ ๊ตฌ์กฐ๊ฐ€ ๋น„๋ถ€์šธ์ด๋ฉฐ, ์ปจํ…์ŠคํŠธ์™€ ํ™•๋ฅ ์ด โ€˜๋ถˆํ™•์ •์„ฑโ€™์„ ๋‚˜ํƒ€๋‚ธ๋‹ค๊ณ  ์ฃผ์žฅํ•œ๋‹ค. 2) ๋ฌดํ•œ ํ…์„œ๊ณฑยท์„นํ„ฐํ™” ํ”„๋กœ๊ทธ๋žจ โ€“ ๋ฌดํ•œํžˆ ๋งŽ์€ ํ๋น„ํŠธ๋กœ ๊ตฌ์„ฑ๋œ ์ธก์ • ์žฅ์น˜๋ฅผ ํ†ตํ•ด ๊ณ ์ „์  โ€˜์ปทโ€™์„ ํšŒํ”ผํ•˜๋ ค๋Š” ์‹œ๋„๋‹ค. ๋ณธ ๋…ผ๋ฌธ์€ ์œ„ ๋‘ ํ๋ฆ„ ์ค‘ ๊ธฐ์ˆ ์ (์ˆ˜ํ•™์ ) ๋ถ€๋ถ„ ์— ์ดˆ์ ์„ ๋งž์ถ”์–ด, ๋ฒ„๋ธŒ์˜ ๋…ผ์ฆ์ด ์–ด๋””์—์„œ ์ •ํ™•ํžˆ ๊ฐ•์ œ๋˜๊ณ , ์–ด๋””์—์„œ ์„ ํƒ์  ๊ฐ€์ •์— ์˜์กดํ•˜๋Š”์ง€๋ฅผ ๊ตฌ๋ถ„ํ•œ๋‹ค. 2. ์ฃผ์š” ์ง€์  ์‚ฌํ•ญ | ๋ฒˆํ˜ธ | ์ง€์  ๋‚ด์šฉ | ์™œ ์ค‘์š”ํ•œ๊ฐ€ | ์ œ์•ˆ๋œ ์ˆ˜์ • | | | |

Quantum Physics
Darth Vecdor: An Open-Source System for Generating Knowledge Graphs Through Large Language Model Queries

Darth Vecdor: An Open-Source System for Generating Knowledge Graphs Through Large Language Model Queries

1. ์—ฐ๊ตฌ์˜ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ LLM ํ™œ์šฉ์˜ ๋น„์šฉยท์†๋„ ๋ฌธ์ œ : LLM API ํ˜ธ์ถœ์€ ํ† ํฐ๋‹น ๋น„์šฉ์ด ๋ฐœ์ƒํ•˜๊ณ , ๋Œ€๊ทœ๋ชจ ์‹ค์‹œ๊ฐ„ ์งˆ์˜ ์‹œ ์ง€์—ฐ์ด ๋ฐœ์ƒํ•œ๋‹ค. ๋ณด์•ˆยทํ”„๋ผ์ด๋ฒ„์‹œ ์œ„ํ—˜ : ์™ธ๋ถ€ LLM ์„œ๋น„์Šค์— ๋ฏผ๊ฐ ๋ฐ์ดํ„ฐ๋ฅผ ์ „์†กํ•˜๋ฉด ๊ฐœ์ธ์ •๋ณด ์œ ์ถœ ์œ„ํ—˜์ด ์กด์žฌํ•œ๋‹ค. ๊ฒฐ๊ณผ ์ผ๊ด€์„ฑยท์„ค๋ช… ๊ฐ€๋Šฅ์„ฑ ๋ถ€์กฑ : ๋™์ผ ํ”„๋กฌํ”„ํŠธ์—๋„ ๋ณ€๋™์„ฑ์ด ์žˆ์–ด ๊ฒ€์ฆยท๊ฐ์‚ฌ๊ฐ€ ์–ด๋ ค์›€. ์ด๋Ÿฌํ•œ ํ•œ๊ณ„๋ฅผ ๊ทน๋ณตํ•˜๊ธฐ ์œ„ํ•ด ์‚ฌ์ „ ์ถ”์ถœ โ†’ ์ •ํ˜•ํ™” โ†’ DB ์ €์žฅ ์ด๋ผ๋Š” ํŒŒ์ดํ”„๋ผ์ธ์„ ์ œ์‹œํ•œ ์ ์€ ์‹ค์šฉ์ ์ด๋ฉฐ, ํŠนํžˆ ์˜๋ฃŒ๊ธฐ๊ด€ ๊ฐ™์€ ๊ทœ์ œ ํ™˜๊ฒฝ์— ์ ํ•ฉํ•˜๋‹ค. 2. ์‹œ์Šคํ…œ ์„ค๊ณ„ ๋ฐ ๊ตฌํ˜„ | ์š”์†Œ | ๊ตฌํ˜„

System Model
Multi-Accent Mandarin Dry-Vocal Singing Dataset: Benchmark for Singing Accent Recognition

Multi-Accent Mandarin Dry-Vocal Singing Dataset: Benchmark for Singing Accent Recognition

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ๋…ธ๋ž˜ ์–ต์–‘ ์—ฐ๊ตฌ์˜ ๊ณต๋ฐฑ : ์Œ์„ฑ ์–ต์–‘ ์—ฐ๊ตฌ๋Š” ๋ฐฉ๋Œ€ํ•œ ๋ฐ์ดํ„ฐ์™€ ํ™œ๋ฐœํ•œ ๋ชจ๋ธ ๊ฐœ๋ฐœ ๋•๋ถ„์— ๋น ๋ฅด๊ฒŒ ์ง„์ „ํ–ˆ์ง€๋งŒ, ๋…ธ๋ž˜ ์˜์—ญ์€ โ€˜๋ณด์ปฌโ€‘์•…๊ธฐ ๋ถ„๋ฆฌโ€™ ๊ณผ์ •์—์„œ ๊ณ ์œ  ์Œํ–ฅ ํŠน์„ฑ์ด ์†์‹ค๋˜๊ณ , ์ง€์—ญ ์–ต์–‘ ๋ผ๋ฒจ์ด ๋ถ€์žฌํ•ด ํ•™์Šตยทํ‰๊ฐ€๊ฐ€ ์–ด๋ ค์› ๋‹ค. ๋ฐ์ดํ„ฐ ๋ถ€์กฑ์ด ์ดˆ๋ž˜ํ•˜๋Š” ํ•œ๊ณ„ : ๊ธฐ์กด์˜ ๋Œ€๊ทœ๋ชจ ๋…ธ๋ž˜ ๋ฐ์ดํ„ฐ์…‹(์˜ˆ: Opencpop, M4Singer ๋“ฑ)์€ ์ŠคํŠœ๋””์˜ค ๋…น์Œยท์Œ์•…์  ํšจ๊ณผ๊ฐ€ ํฌํ•จ๋œ โ€˜์›ป ๋ณด์ปฌโ€™์ด๋ฉฐ, ๋ฐฉ์–ธยท์–ต์–‘ ๋ผ๋ฒจ์ด ์—†์–ด์„œ ์–ต์–‘ ์ธ์‹ยท์ „์ด ์—ฐ๊ตฌ์— ํ™œ์šฉํ•˜๊ธฐ ํž˜๋“ค๋‹ค. 2. ๋ฐ์ดํ„ฐ์…‹ ์„ค๊ณ„ยท๊ตฌ์„ฑ | ํ•ญ๋ชฉ | ๋‚ด์šฉ | | | | |

Data
SPoRC-VIST: A Benchmark for Evaluating Generative Natural Narrative in Vision-Language Models

SPoRC-VIST: A Benchmark for Evaluating Generative Natural Narrative in Vision-Language Models

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ๋น„์ „โ€‘์–ธ์–ด ๋ชจ๋ธ์˜ ํ•œ๊ณ„ : ๊ธฐ์กด VLM์€ โ€œ๋ฌด์—‡์ด ๋ณด์ด๋Š”๊ฐ€โ€๋ฅผ ์ •ํ™•ํžˆ ์„ค๋ช…ํ•˜๋Š” ๋ฐ ๊ฐ•์ ์ด ์žˆ์ง€๋งŒ, ์ธ๊ฐ„์ด ์‚ฌ์ง„์„ ๋ณด๋ฉฐ ๋‚˜๋ˆ„๋Š” ์ž์—ฐ์Šค๋Ÿฌ์šด ๋Œ€ํ™”, ์œ ๋จธ, ๊ฐ์ • ํ‘œํ˜„ ๋“ฑ์„ ์žฌํ˜„ํ•˜๋Š” ๋ฐ๋Š” ๋ถ€์กฑํ•˜๋‹ค. ํ‰๊ฐ€ ์ง€ํ‘œ์˜ ๋ถ€์žฌ : BLEUยทROUGE ๋“ฑ ํ…์ŠคํŠธ ๊ฒน์นจ ๊ธฐ๋ฐ˜ ์ง€ํ‘œ๋Š” ์ฐฝ์˜์  ์„œ์‚ฌ์— ๋ถ€์ ํ•ฉํ•˜๋ฉฐ, ๋Œ€ํ™”ํ˜• AI์˜ โ€˜๊ฐœ์„ฑโ€™๊ณผ โ€˜ํ๋ฆ„โ€™์„ ์ •๋Ÿ‰ํ™”ํ•  ์ˆ˜ ์—†๋‹ค. 2. ์ฃผ์š” ๊ธฐ์—ฌ | ๋ฒˆํ˜ธ | ๊ธฐ์—ฌ ๋‚ด์šฉ | ์˜์˜ | | | | | | 1 | SPoRCโ€‘VIST ๋ฒค์น˜๋งˆํฌ : ํ•ฉ์„ฑ ์ด๋ฏธ์ง€์™€ ์‹ค์ œ ์‚ฌ์ง„์„ ์—ฐ๊ฒฐํ•œ ์ƒˆ๋กœ์šด ํ•™์Šตยทํ‰๊ฐ€ ํŒŒ์ดํ”„

Computer Science Machine Learning Model
์š”๋ฆฌ ๋‹จ๊ณ„๋ณ„ ์ด๋ฏธ์ง€ ์ƒ์„ฑ์˜ ์ƒˆ๋กœ์šด ํŒจ๋Ÿฌ๋‹ค์ž„

์š”๋ฆฌ ๋‹จ๊ณ„๋ณ„ ์ด๋ฏธ์ง€ ์ƒ์„ฑ์˜ ์ƒˆ๋กœ์šด ํŒจ๋Ÿฌ๋‹ค์ž„

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ์‹œ๊ฐ์  ์ˆœ์ฐจ์„ฑ : ์š”๋ฆฌ ๊ณผ์ •์€ โ€œ์ž๋ฅด๊ธฐ โ†’ ์„ž๊ธฐ โ†’ ๊ตฝ๊ธฐโ€์™€ ๊ฐ™์ด ๋‹จ๊ณ„๋งˆ๋‹ค ์‹œ๊ฐ์  ๋ณ€ํ™”๋ฅผ ๋™๋ฐ˜ํ•œ๋‹ค. ์ด๋ฅผ ํ…์ŠคํŠธ๋งŒ์œผ๋กœ ์ „๋‹ฌํ•˜๋ฉด ์ดํ•ด๋„๊ฐ€ ๋–จ์–ด์ง€๋ฏ€๋กœ, ๋‹จ๊ณ„๋ณ„ ์ด๋ฏธ์ง€๊ฐ€ ๊ต์œกยท๋ณด์กฐ ๊ธฐ์ˆ ์— ํฐ ๊ฐ€์น˜๋ฅผ ์ œ๊ณตํ•œ๋‹ค. ๊ธฐ์กด ํ•œ๊ณ„ : ๋‹จ์ผ ์ด๋ฏธ์ง€ ์ค‘์‹ฌ : ๋Œ€๋ถ€๋ถ„์˜ ์ตœ์‹  ํ™•์‚ฐ ๋ชจ๋ธ(์˜ˆ: Stable Diffusion, FLUX)์€ ํ•˜๋‚˜์˜ ํ”„๋กฌํ”„ํŠธ์— ๋Œ€ํ•ด ํ•˜๋‚˜์˜ ์ด๋ฏธ์ง€๋งŒ ์ƒ์„ฑํ•œ๋‹ค. ๊ณ ์ • ์ด๋ฏธ์ง€ ์ˆ˜ : StackedDiffusion ๋“ฑ์€ ๋ ˆ์‹œํ”ผ๋‹น ๊ณ ์ •๋œ ์ด๋ฏธ์ง€ ์ˆ˜๋ฅผ ์ถœ๋ ฅํ•ด, ๋ ˆ์‹œํ”ผ ๊ธธ์ด๊ฐ€ ์งง๊ฑฐ๋‚˜ ๊ธธ ๋•Œ ๊ณผยท๋ถ€์ƒ์„ฑ์„ ์ดˆ๋ž˜ํ•œ๋‹ค.

AI-Powered Annotation Pipelines for Stabilizing Large Language Models: A Human-AI Synergy Approach

AI-Powered Annotation Pipelines for Stabilizing Large Language Models: A Human-AI Synergy Approach

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ฌธ์ œ ์ •์˜ ๋ถˆ์•ˆ์ •์„ฑ์˜ ๊ตฌ์ฒด์  ์œ ํ˜• ์„ ํ‘œ๋กœ ์ •๋ฆฌ(์˜๋ฏธ ๋ฐœ์‚ฐ, ํ™˜๊ฐ, ๋…ผ๋ฆฌ ๋ถ•๊ดด, ์„ธ์…˜ ๋“œ๋ฆฌํ”„ํŠธ)ํ•˜๊ณ , ์ด๋Š” ํ•™์Šต ๋ฐ์ดํ„ฐ ๋…ธ์ด์ฆˆยท์‹ ๋ขฐ๋„ ๋ถ€์žฌยทํ”ผ๋“œ๋ฐฑ ๋ฃจํ”„ ๊ฒฐ์—ฌ ์—์„œ ๊ธฐ์ธํ•œ๋‹ค๋Š” ์ ์„ ๋ช…ํ™•ํžˆ ์ œ์‹œํ•œ๋‹ค. ๊ธฐ์กด ํ‰๊ฐ€ ๋ฒค์น˜๋งˆํฌ๊ฐ€ ์ •ํ™•๋„ยท์ •๋‹ต๋ฅ  ์—๋งŒ ์ดˆ์ ์„ ๋งž์ถ”๊ณ  ์ถœ๋ ฅ ๋ณ€๋™์„ฑ ์„ ์ธก์ •ํ•˜์ง€ ๋ชปํ•œ๋‹ค๋Š” ๋น„ํŒ์€ ๋งค์šฐ ์„ค๋“๋ ฅ ์žˆ๋‹ค. 2. ์ฃผ์š” ๊ธฐ์—ฌ (Contributions) | ๋ฒˆํ˜ธ | ๋‚ด์šฉ | ์˜์˜ | | | | | | 1 | ์ž๋™โ€‘์ธ๊ฐ„ ํ•˜์ด๋ธŒ๋ฆฌ๋“œ ์ฃผ์„ ํŒŒ์ดํ”„๋ผ์ธ ์„ค๊ณ„ (์•ฝํ•œ ๊ฐ๋… + ์‹ ๋ขฐ๋„ ๊ธฐ๋ฐ˜ ํ”Œ๋ž˜๊น… + ์ธ๊ฐ„ ๊ฒ€์ฆ) | ๋น„์šฉ

Model
Beyond Proximity: A Keypoint-Trajectory Framework for Classifying Affiliative and Agonistic Social Networks in Dairy Cattle

Beyond Proximity: A Keypoint-Trajectory Framework for Classifying Affiliative and Agonistic Social Networks in Dairy Cattle

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ์‚ฌํšŒ์  ํ–‰๋™์˜ ์ค‘์š”์„ฑ : ์นœํ™”ยท์ ๋Œ€ ์ƒํ˜ธ์ž‘์šฉ์€ ์ŠคํŠธ๋ ˆ์Šค, ์„œ์—ด, ์ž์› ์ ‘๊ทผ ๋“ฑ์— ์ง์ ‘์ ์ธ ์˜ํ–ฅ์„ ๋ฏธ์น˜๋ฉฐ, ๋ณต์ง€ยท์ƒ์‚ฐ์„ฑ ์ง€ํ‘œ์™€ ์—ฐ๊ด€๋œ๋‹ค. ๊ธฐ์กด ํ•œ๊ณ„ : ๊ฑฐ๋ฆฌ ์ž„๊ณ„๊ฐ’๋งŒ์œผ๋กœ๋Š” ํ–‰๋™์˜ โ€˜๊ฐ€์น˜(valence)โ€™๋ฅผ ํŒŒ์•…ํ•  ์ˆ˜ ์—†์–ด, ์‚ฌํšŒ ๋„คํŠธ์›Œํฌ ๋ถ„์„์˜ ํ•ด์„๋ ฅ์ด ํฌ๊ฒŒ ์ €ํ•˜๋œ๋‹ค. ํ‚คํฌ์ธํŠธ ๊ธฐ๋ฐ˜ ์ ‘๊ทผ : ๊ณจ๊ฒฉ(์Šค์ผˆ๋ ˆํ†ค) ์ •๋ณด๋ฅผ ํ™œ์šฉํ•˜๋ฉด ์˜์ƒ ์ „์ฒด๋ฅผ ์ฒ˜๋ฆฌํ•˜๋Š” ํ”ฝ์…€โ€‘๋ ˆ๋ฒจ ๋ชจ๋ธ๋ณด๋‹ค ์—ฐ์‚ฐ ํšจ์œจ์ด ๋†’๊ณ , ํ–‰๋™์˜ ๋™์  ํŠน์„ฑ์„ ์ง์ ‘ ํฌ์ฐฉํ•  ์ˆ˜ ์žˆ๋‹ค. 2. ๊ธฐ์ˆ ์  ๊ตฌ์„ฑ ๋ฐ ํ˜์‹ ์„ฑ | ๋‹จ๊ณ„ | ์‚ฌ์šฉ ๊ธฐ์ˆ  | ์ฃผ์š” ์„ฑ๋Šฅ | ํ˜์‹  ํฌ์ธํŠธ

Network Framework
Highly Efficient Test-Time Scaling for T2I Diffusion Models with Text Embedding Perturbation

Highly Efficient Test-Time Scaling for T2I Diffusion Models with Text Embedding Perturbation

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ํ…Œ์ŠคํŠธโ€‘ํƒ€์ž„ ์Šค์ผ€์ผ๋ง(TTS) ์€ LLM์—์„œ ์˜๊ฐ์„ ๋ฐ›์•„ ๋””ํ“จ์ „ ๋ชจ๋ธ์—์„œ๋„ โ€œ์—ฌ๋Ÿฌ ๋ฒˆ ์ƒ˜ํ”Œ๋ง โ†’ ๋ณด์ƒ ๊ธฐ๋ฐ˜ ์„ ํƒ โ†’ ์ตœ์ข… ์ถœ๋ ฅโ€ ํ๋ฆ„์œผ๋กœ ์ ์šฉ๋˜๊ณ  ์žˆ๋‹ค. ๊ธฐ์กด TTS๋Š” ๋ฌด์ž‘์œ„์„ฑ ์„ ์ดˆ๊ธฐ ๋…ธ์ด์ฆˆ ํ˜น์€ SDEโ€‘์ฃผ์ž… ๋…ธ์ด์ฆˆ์—๋งŒ ์˜์กดํ–ˆ์œผ๋ฉฐ, ์ด๋Š” ๊ฒ€์ƒ‰ ๊ณต๊ฐ„(search space) ์„ ์ œํ•œํ•œ๋‹ค. ํ…์ŠคํŠธ ์ž„๋ฒ ๋”ฉ ์ž์ฒด๊ฐ€ ๋ชจ๋ธ์˜ ์กฐ๊ฑด๋ถ€ ์ž…๋ ฅ์ด๋ผ๋Š” ์ ์„ ํ™œ์šฉํ•˜๋ฉด, ์กฐ๊ฑด๋ถ€ ์ฐจ์›์—์„œ๋„ ๋ฌด์ž‘์œ„์„ฑ์„ ์ฃผ์ž… ํ•  ์ˆ˜ ์žˆ๋‹ค๋Š” ์ƒˆ๋กœ์šด ๊ฐ€๋Šฅ์„ฑ์„ ์ œ์‹œํ•œ๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด | ๋ฌด์ž‘์œ„์„ฑ ์ข…๋ฅ˜ | ์ž‘์šฉ ์‹œ์  | ์ฃผํŒŒ์ˆ˜ ํŠน์„ฑ | ์ฃผ์š” ํšจ๊ณผ |

Model
PointRAFT: 3D deep learning for high-throughput prediction of potato tuber weight from partial point clouds

PointRAFT: 3D deep learning for high-throughput prediction of potato tuber weight from partial point clouds

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ | ๋ฌธ์ œ์  | ๊ธฐ์กด ์ ‘๊ทผ๋ฒ• | ํ•œ๊ณ„ | | | | | | ์ž์ฒด ๊ฐ€๋ฆผ ์œผ๋กœ ์ธํ•œ ๋ถ€๋ถ„ ๊ด€์ธก | RGBโ€‘D ๊ธฐ๋ฐ˜ ๋ถ€ํ”ผยท๋ฌด๊ฒŒ ์ถ”์ • (ํ”ฝ์…€โ€‘ํˆฌโ€‘์›”๋“œ ๋ณ€ํ™˜) | ๊ฐ€๋ฆผ์ด ์‹ฌํ•œ ๊ฒฝ์šฐ ์ฒด์ ยท๋ฌด๊ฒŒ ๊ณผ์†Œ์ถ”์ • | | ๋ฌด๊ฒŒ ์ธก์ •์— ํƒ‘์žฌ๋œ ๋กœ๋“œ์…€ | ์ง์ ‘ ์งˆ๋Ÿ‰ ์ธก์ • | ํ† ์–‘ยท๋Œยท์ž”์—ฌ๋ฌผ ํฌํ•จ โ†’ ๊ณผ๋Œ€์ถ”์ • | | ๋‹ค์ค‘ ์นด๋ฉ”๋ผ | ์‹œ์  ๋ณด์™„ | ์„ค์น˜ยท๋ณด์ • ๋ณต์žก, ์‹ค์‹œ๊ฐ„ ์ฒ˜๋ฆฌ ์–ด๋ ค์›€ | | 3D Shape Completion | Voxel/Implicit/Pointโ€‘Net ๊ธฐ๋ฐ˜ ์™„์ „ ํ˜•ํƒœ ๋ณต์› | ๋Œ€๊ทœ๋ชจ ๋ผ๋ฒจ๋งยท์™„์ „ ๋ฐ์ดํ„ฐ ํ•„์š”, ์ฒ˜๋ฆฌ

Computer Science Learning Computer Vision
RisConFix: LLM-based Automated Repair of Risk-Prone Drone Configurations

RisConFix: LLM-based Automated Repair of Risk-Prone Drone Configurations

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ๋“œ๋ก  ์ œ์–ด ํŒŒ๋ผ๋ฏธํ„ฐ์˜ ๋ณต์žก์„ฑ : PX4, ArduPilot ๋“ฑ ์ตœ์‹  ๋น„ํ–‰ ์ œ์–ด ์Šคํƒ์€ ์ˆ˜๋ฐฑ ๊ฐœ์— ๋‹ฌํ•˜๋Š” ํŒŒ๋ผ๋ฏธํ„ฐ๋ฅผ ์ œ๊ณตํ•ด ์ž„๋ฌด ๋งž์ถคํ˜• ํŠœ๋‹์„ ๊ฐ€๋Šฅํ•˜๊ฒŒ ํ•˜์ง€๋งŒ, ํŒŒ๋ผ๋ฏธํ„ฐ ๊ฐ„ ์ƒํ˜ธ์ž‘์šฉ์„ ๋ชจ๋‘ ์‚ฌ์ „์— ๊ฒ€์ฆํ•˜๊ธฐ๋Š” ํ˜„์‹ค์ ์œผ๋กœ ๋ถˆ๊ฐ€๋Šฅํ•˜๋‹ค. ๊ธฐ์กด ์—ฐ๊ตฌ์˜ ํ•œ๊ณ„ : LGDFUZZER, ICSEARCHER ๋“ฑ์€ ์˜คํ”„๋ผ์ธ ์—์„œ ์œ„ํ—˜ ์„ค์ •์„ ํƒ์ง€ํ•˜์ง€๋งŒ, ์‹ค์‹œ๊ฐ„ ์ƒํ™ฉ ๋ณ€ํ™”(ํ’๋™, ์„ผ์„œ ๋…ธ์ด์ฆˆ ๋“ฑ)๊นŒ์ง€ ํฌ๊ด„ํ•˜์ง€ ๋ชปํ•œ๋‹ค. ๋”ฐ๋ผ์„œ ์˜จ๋ผ์ธยท์ ์‘ํ˜• ๋ณต๊ตฌ ๋ฉ”์ปค๋‹ˆ์ฆ˜ ์ด ์ ˆ์‹คํžˆ ์š”๊ตฌ๋œ๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด ๋ฐ ๊ธฐ์—ฌ | ๊ธฐ์—ฌ | ๋‚ด์šฉ | ์˜์˜

๋ฃจ๋น…ํ๋ธŒ๋ฅผ ํ†ตํ•œ ํ‰์ƒ ์ „๋ฌธ๊ฐ€ ํ•™์Šต์˜ ๋ณดํŽธ์„ฑ

๋ฃจ๋น…ํ๋ธŒ๋ฅผ ํ†ตํ•œ ํ‰์ƒ ์ „๋ฌธ๊ฐ€ ํ•™์Šต์˜ ๋ณดํŽธ์„ฑ

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ชฉ์  ๋ฌธ์ œ ์ธ์‹ : ์ „๋ฌธ๊ฐ€ ์„ฑ๊ณผ์— ๋Œ€ํ•œ ์žฅ๊ธฐ์ ์ธ ์ •๋Ÿ‰ ๋ฐ์ดํ„ฐ๊ฐ€ ๋ถ€์กฑํ•ด, ํ•™์Šต ๋ฉ”์ปค๋‹ˆ์ฆ˜๊ณผ ์ง€์‹ ์ „ํŒŒ ๊ณผ์ •์„ ์ฒด๊ณ„์ ์œผ๋กœ ๋ชจ๋ธ๋งํ•˜๊ธฐ ์–ด๋ ต๋‹ค. ๋ฃจ๋น…ํ๋ธŒ ์„ ํƒ ์ด์œ  : ๋ฌผ๋ฆฌ์  ํผ์ฆ์ด๋ฉด์„œ๋„ 3โ€‘์ฐจ์› ํผ๋ฎคํ…Œ์ด์…˜ ๊ตฐ์˜ Cayley ๊ทธ๋ž˜ํ”„๋ผ๋Š” ์—„๊ฒฉํ•œ ์ˆ˜ํ•™์  ๊ตฌ์กฐ๋ฅผ ๊ฐ–๋Š”๋‹ค. ๋”ฐ๋ผ์„œ ์ธ์ง€ ๊ณต์˜ˆ(Cognitive Artifact) ๋กœ์„œ โ€˜์—”ํŠธ๋กœํ”ผยท๋‚œ์ด๋„ยท๊ฒ€์ƒ‰ยท์ตœ์ ์„ฑโ€™ ๋“ฑ ์ถ”์ƒ์  ๊ฐœ๋…์„ ๊ตฌ์ฒดํ™”ํ•  ์ˆ˜ ์žˆ๋‹ค. 2. ๋ฐ์ดํ„ฐ ๋ฐ ๋ฐฉ๋ฒ•๋ก  | ํ•ญ๋ชฉ | ๋‚ด์šฉ | | | | | ๋ฐ์ดํ„ฐ ์ถœ์ฒ˜ | World Cube Association (WCA) ๊ณต์‹

GAIR: GUI Automation via Information-Joint Reasoning and Group Reflection

GAIR: GUI Automation via Information-Joint Reasoning and Group Reflection

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ๋‹ค์–‘ํ•œ GUI ์‹œ๋‚˜๋ฆฌ์˜ค : ๋ฌธ์„œ ํŽธ์ง‘๋ถ€ํ„ฐ CAD, ์˜์ƒ ํŽธ์ง‘๊นŒ์ง€ ์ž‘์—… ๋ฒ”์œ„๊ฐ€ ๋„“์–ด ๋‹จ์ผ MLLM์œผ๋กœ๋Š” ๋ชจ๋“  ์ƒํ™ฉ์„ ์ปค๋ฒ„ํ•˜๊ธฐ ์–ด๋ ต๋‹ค. ์ด์งˆ์ ์ธ ๋ฐ์ดํ„ฐ ๋ถ„ํฌ : GUIโ€‘์ „๋ฌธ ๋ชจ๋ธ๋“ค์€ ๊ฐ๊ฐ ํŠนํ™”๋œ ๋ฐ์ดํ„ฐ์…‹์— ๊ธฐ๋ฐ˜ํ•ด ํ•™์Šต๋˜๋ฏ€๋กœ ๊ฐ•์ ยท์•ฝ์ ์ด ์ƒ์ดํ•˜๋‹ค. ๊ธฐ์กด ์ ‘๊ทผ๋ฒ•์˜ ํ•œ๊ณ„ : ๋‹จ์ผ ๋ชจ๋ธ ๊ธฐ๋ฐ˜ ํ˜น์€ ๋ณด์กฐ ํˆด์— ์˜์กดํ•˜๋Š” ๋ฐฉ๋ฒ•์€ ์ •๋ณด ์ •ํ•ฉ์„ฑ ๋ถ€์กฑ, ๊ฒฐ์ • ์ถฉ๋Œ, ์˜ค๋ฅ˜ ์ „ํŒŒ ๋“ฑ์˜ ๋ฌธ์ œ๋ฅผ ์•ผ๊ธฐํ•œ๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด ์ด์ข… ๋ชจ๋ธ ์•™์ƒ๋ธ” : ์—ฌ๋Ÿฌ GUIโ€‘์ „๋ฌธ MLLM์„ ๋™์‹œ์— ํ™œ์šฉํ•ด ๊ฐ ๋ชจ๋ธ์ด ๊ฐ€์žฅ ๊ฐ•์ ์ด ์žˆ๋Š” ์„œ๋ธŒํƒœ์Šคํฌ(์˜ˆ: ์š”

SWEnergy: An Empirical Study on Energy Efficiency in Agentic Issue Resolution Frameworks with SLMs

SWEnergy: An Empirical Study on Energy Efficiency in Agentic Issue Resolution Frameworks with SLMs

1. ์—ฐ๊ตฌ ๋™๊ธฐ์™€ ๋ฐฐ๊ฒฝ ํ™˜๊ฒฝยท๊ฒฝ์ œ์  ์••๋ฐ• : AI ๋ชจ๋ธ์˜ ์—ฐ์‚ฐ๋Ÿ‰ยท์ „๋ ฅ ์†Œ๋ชจ๊ฐ€ ์ง€์† ๊ฐ€๋Šฅํ•œ ์†Œํ”„ํŠธ์›จ์–ด ์—”์ง€๋‹ˆ์–ด๋ง์— ํฐ ์žฅ์• ๋ฌผ๋กœ ๋ถ€๊ฐ๋˜๊ณ  ์žˆ๋‹ค. ํŠนํžˆ ํด๋ผ์šฐ๋“œ ๊ธฐ๋ฐ˜ ๋Œ€ํ˜• LLM์€ ๋น„์šฉยทํ”„๋ผ์ด๋ฒ„์‹œยท์ง€์—ฐ ์‹œ๊ฐ„ ์ธก๋ฉด์—์„œ๋„ ํ•œ๊ณ„๊ฐ€ ์žˆ๋‹ค. SLM์˜ ๊ธฐ๋Œ€ : ํŒŒ๋ผ๋ฏธํ„ฐ ์ˆ˜๊ฐ€ ์ˆ˜์–ต~์ˆ˜์‹ญ์–ต ์ˆ˜์ค€์ธ ์˜คํ”ˆโ€‘์›จ์ดํŠธ ๋ชจ๋ธ์€ ์†Œ๋น„์ž ๋“ฑ๊ธ‰ ํ•˜๋“œ์›จ์–ด ์—์„œ๋„ ์‹คํ–‰ ๊ฐ€๋Šฅํ•˜๋‹ค๋Š” ์ ์—์„œ ๋งค๋ ฅ์ ์ด๋‹ค. ํ•˜์ง€๋งŒ ์ถ”๋ก  ๋Šฅ๋ ฅยทํ”„๋กฌํ”„ํŠธ ์ดํ•ด๋„ ๊ฐ€ ์ œํ•œ์ ์ด์–ด์„œ, ๋ณต์žกํ•œ ๋‹ค๋‹จ๊ณ„ ์—์ด์ „ํŠธ ์›Œํฌํ”Œ๋กœ์— ์ ์šฉํ–ˆ์„ ๋•Œ ์‹ค์ œ ์„ฑ๋Šฅ์ด ์–ด๋–ป๊ฒŒ ๋ณ€ํ• ์ง€๋Š” ๋ฏธ์ง€์ˆ˜์˜€๋‹ค. 2. ์‹คํ—˜ ์„ค๊ณ„ | ์š”์†Œ | ๋‚ด์šฉ

Framework

< Category Statistics (Total: 5055) >

Electrical Engineering and Systems Science
102
General
4153
General Relativity
3
HEP-EX
3
HEP-PH
1
HEP-TH
4
MATH-PH
7
Quantum Physics
18

Start searching

Enter keywords to search articles

โ†‘โ†“
โ†ต
ESC
โŒ˜K Shortcut