KOINEU Logo
No Image

DEEPAMBIGQA: Ambiguous Multi-hop Questions for Benchmarking LLM Answer Completeness

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ๋ฉ€ํ‹ฐํ™‰ ์ถ”๋ก  ๊ณผ ๋ช…์นญยท์ œ๋ชฉ ๋ชจํ˜ธ์„ฑ ์€ ์‹ค์ œ ๊ฒ€์ƒ‰ยทQA ํ™˜๊ฒฝ์—์„œ ๋นˆ๋ฒˆํžˆ ๋งˆ์ฃผ์น˜๋Š” ๋ฌธ์ œ์ด๋‹ค. ๊ธฐ์กด SQuAD, HotpotQA, AmbigQA ๋“ฑ์€ ๊ฐ๊ฐ ๋ฉ€ํ‹ฐํ™‰ ํ˜น์€ ๋ชจํ˜ธ์„ฑ์— ์ดˆ์ ์„ ๋งž์ถ”์ง€๋งŒ, ๋‘ ์š”์†Œ๋ฅผ ๋™์‹œ์— ์š”๊ตฌํ•˜๋Š” ์งˆ๋ฌธ์€ ๊ฑฐ์˜ ์—†์—ˆ๋‹ค. LLM+๊ฒ€์ƒ‰ ํŒŒ์ดํ”„๋ผ์ธ(์˜ˆ: ReAct, Selfโ€‘RAG)์€ โ€œ๊ฒ€์ƒ‰ โ†’ ์ถ”๋ก  โ†’ ๋‹ต๋ณ€โ€ ์ˆœํ™˜์„ ํ†ตํ•ด ์„ฑ๋Šฅ์„ ๋Œ์–ด์˜ฌ๋ฆฌ์ง€๋งŒ, ๋‹ต๋ณ€ ์ง‘ํ•ฉ์˜ ์™„์ „์„ฑ(completeness) ์„ ํ‰๊ฐ€ํ•  ๋ฉ”ํŠธ๋ฆญ์ด ๋ถ€์กฑํ–ˆ๋‹ค. 2. ๋ฐ์ดํ„ฐ ์ƒ์„ฑ ํŒŒ์ดํ”„๋ผ์ธ โ€“ DeepAmbigQAGen | ๋‹จ๊ณ„ | ํ•ต์‹ฌ

No Image

Deformation and orientation of a capsule with viscosity contrast in linear flows: a theoretical study

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ์˜์˜ ์บก์Аยท๋งˆ์ดํฌ๋กœ์บก์А์€ ๋ฐ”์ด์˜คยทํ™”ํ•™ยท์‹ํ’ˆยทํ™”์žฅํ’ˆ ๋“ฑ ๋‹ค์–‘ํ•œ ๋ถ„์•ผ์—์„œ โ€œํ™œ์„ฑ ๋ฌผ์งˆ์˜ ์ „๋‹ฌยท๋ณดํ˜ธยท๋ฐ˜์‘ ์šฉ๊ธฐโ€ ์—ญํ• ์„ ์ˆ˜ํ–‰ํ•œ๋‹ค. ๊ธฐ์กด ์ด๋ก ์€ ์ฃผ๋กœ ์ ์„ฑ ๋Œ€๋น„๊ฐ€ 1 ์ธ ๊ฒฝ์šฐ์™€ ํ‘œ๋ฉด ์žฅ๋ ฅยท๊ตฝํž˜ ๊ฐ•์„ฑ์ด ์—†๋Š” ์ด์ƒ์ ์ธ ์บก์А์„ ๋‹ค๋ฃจ์—ˆ์œผ๋ฉฐ, ๋ณ€ํ˜•โ€‘๋ฐฐํ–ฅ ๊ด€๊ณ„๋Š” 1์ฐจ Ca ์—๋งŒ ์˜์กดํ•œ๋‹ค๋Š” ๊ฒฐ๋ก ์„ ์–ป์—ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์‹ค์ œ ์บก์А์€ ๋‚ด๋ถ€ยท์™ธ๋ถ€ ์ ์„ฑ ์ฐจ์ด, ์–‡์€ ํƒ„์„ฑ ๋ง‰์˜ ํ‘œ๋ฉด ์žฅ๋ ฅ, ๊ทธ๋ฆฌ๊ณ  ์–‡์€ ๊ป์งˆ์˜ ๊ตฝํž˜ ์ €ํ•ญ์„ ๋™์‹œ์— ๊ฐ–๋Š”๋‹ค. ์ด ๋…ผ๋ฌธ์€ ์ด๋Ÿฌํ•œ ๋„ค ๊ฐ€์ง€ ๋ฌผ๋ฆฌ์  ํšจ๊ณผ(ฮป, ฯƒ, ฮบ, Gโ‚›) ๋ฅผ ๋ชจ๋‘ ํฌํ•จํ•œ ์ผ๋ฐ˜ํ™”๋œ ๋ชจ๋ธ์„ ๊ตฌ์ถ•ํ•˜๊ณ , 2์ฐจ

Condensed Matter
Distributional Deep Learning for Super-Resolution of 4D Flow MRI under Domain Shift

Distributional Deep Learning for Super-Resolution of 4D Flow MRI under Domain Shift

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ 4D Flow MRI์˜ ์ž„์ƒ ๊ฐ€์น˜ : ๋‡Œ๋™๋งฅ๋ฅ˜์™€ ๊ฐ™์€ ํ˜ˆ๊ด€ ์งˆํ™˜์—์„œ ํ˜ˆ๋ฅ˜ ์†๋„, ๋ฒฝ ์ „๋‹จ์‘๋ ฅ(WSS) ๋“ฑ์€ ํŒŒ์—ด ์œ„ํ—˜์„ ์˜ˆ์ธกํ•˜๋Š” ํ•ต์‹ฌ ์ง€ํ‘œ์ด๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ 4DF๋Š” ์ธก์ • ๋…ธ์ด์ฆˆ์™€ ๋‚ฎ์€ ๊ณต๊ฐ„ ํ•ด์ƒ๋„ ๋•Œ๋ฌธ์— ๋ฏธ์„ธ ํ˜ˆ๋ฅ˜ ํŠน์„ฑ์„ ํฌ์ฐฉํ•˜๊ธฐ ์–ด๋ ต๋‹ค. CFD์™€ 4DF์˜ ์ƒ๋ณด์„ฑ : CFD๋Š” ๊ณ ํ•ด์ƒ๋„ยท๋…ธ์ด์ฆˆโ€‘ํ”„๋ฆฌ ํ๋ฆ„์žฅ์„ ์ œ๊ณตํ•˜์ง€๋งŒ, ๋ชจ๋ธ๋ง ๊ฐ€์ •๊ณผ ์ดˆ๊ธฐ ์กฐ๊ฑด์— ๋ฏผ๊ฐํ•ด ์ž„์ƒ ์ ์šฉ์ด ์ œํ•œ์ ์ด๋‹ค. ๋”ฐ๋ผ์„œ CFD ๋ฐ์ดํ„ฐ๋ฅผ ํ™œ์šฉํ•ด 4DF ํ•ด์ƒ๋„๋ฅผ ๋ณด๊ฐ•ํ•˜๋Š” ์ „๋žต์ด ํ•„์š”ํ•˜๋‹ค. ๋„๋ฉ”์ธ ์ด๋™ ๋ฌธ์ œ : ๊ธฐ์กด SR ๋ชจ๋ธ์€ โ€œ๋‹ค์šด์ƒ˜ํ”Œ๋œ CFD ์‹ค

Computer Science Learning Computer Vision
No Image

EdgeRunner 20B: Military Task Parity with GPT-5 while Running on the Edge

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ๋ฐ์ดํ„ฐ ๋ฏผ๊ฐ์„ฑ : ๊ตฐ์‚ฌ ์ •๋ณด๋Š” ๊ตญ๊ฐ€ ์•ˆ๋ณด์™€ ์ง๊ฒฐ๋ผ ์™ธ๋ถ€ ํด๋ผ์šฐ๋“œ์— ์˜์กดํ•˜๊ธฐ ์–ด๋ ต๋‹ค. ๋”ฐ๋ผ์„œ ์˜คํ”„๋ผ์ธยท์—ฃ์ง€ ํ™˜๊ฒฝ ์—์„œ ๋™์ž‘ ๊ฐ€๋Šฅํ•œ ๋ชจ๋ธ์ด ํ•„์š”ํ•˜๋‹ค. ๋ชจ๋ธ ๊ทœ๋ชจ์™€ ๋น„์šฉ : GPTโ€‘5์™€ ๊ฐ™์€ ์ดˆ๋Œ€ํ˜• ๋ชจ๋ธ์€ ์—ฐ์‚ฐยท์ „๋ ฅยท๋ณด์•ˆ ์ธก๋ฉด์—์„œ ๊ตฐ์‚ฌ ํ˜„์žฅ์— ๋ถ€์ ํ•ฉํ•˜๋‹ค. 20 B ํŒŒ๋ผ๋ฏธํ„ฐ ์ˆ˜์ค€์€ GPU 1~2๋Œ€ ํ˜น์€ ๊ณ ์„ฑ๋Šฅ ์—ฃ์ง€ ASIC ์—์„œ๋„ ์‹ค์‹œ๊ฐ„ ์ถ”๋ก ์ด ๊ฐ€๋Šฅํ•˜๋„๋ก ์„ค๊ณ„๋˜์—ˆ๋‹ค. 2. ๋ฐ์ดํ„ฐ์…‹ ๊ตฌ์ถ• | ํ•ญ๋ชฉ | ๊ทœ๋ชจ | ์ถœ์ฒ˜ | ํŠน์ง• | | | | | | | ๊ตฐ์‚ฌ ๋ฌธ์„œยท์›น ๋ฐ์ดํ„ฐ | 1.6 M ๋ ˆ์ฝ”๋“œ | ๊ตฐ์‚ฌ ๋งค๋‰ด์–ผ, ์ž‘์ „ ๋ณด

No Image

Efficient LLM Safety Evaluation through Multi-Agent Debate

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ LLMโ€‘asโ€‘aโ€‘Judge ์ ‘๊ทผ๋ฒ•์€ ์ธ๊ฐ„ ํ‰๊ฐ€ ๋น„์šฉ์„ ํฌ๊ฒŒ ๋‚ฎ์ถ”์ง€๋งŒ, GPTโ€‘4oยทClaudeโ€‘2 ๋“ฑ ์ตœ์‹  ๋ชจ๋ธ์„ ์ง€์†์ ์œผ๋กœ ํ˜ธ์ถœํ•˜๋ฉด GPUยทAPI ๋น„์šฉ ์ด ๊ธ‰์ฆํ•œ๋‹ค. ์•ˆ์ „์„ฑ ํ‰๊ฐ€๋ฅผ ๋Œ€๊ทœ๋ชจยท๋ฐ˜๋ณต ์ ์œผ๋กœ ์ˆ˜ํ–‰ํ•˜๋ ค๋ฉด ๋น„์šฉ ํšจ์œจ์ ์ธ ๋Œ€์•ˆ ์ด ํ•„์ˆ˜์ ์ด๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด ๋ฉ€ํ‹ฐโ€‘์—์ด์ „ํŠธ ํ† ๋ก  : Critic โ€“ ์ž…๋ ฅ๋œ ํƒˆ์˜ฅ ์‹œ๋‚˜๋ฆฌ์˜ค์˜ ์œ„ํ—˜์„ฑ์„ ํƒ์ƒ‰ยท์ง€์ . Defender โ€“ ๋ชจ๋ธ์ด ํ•ด๋‹น ์‹œ๋‚˜๋ฆฌ์˜ค์— ๋Œ€ํ•ด ์–ด๋–ป๊ฒŒ ์ •์ƒ์ ์ธ ๋‹ต๋ณ€์„ ์œ ์ง€ํ•  ์ˆ˜ ์žˆ๋Š”์ง€ ๋…ผ์ฆ. Judge โ€“ ๋‘ ์˜๊ฒฌ์„ ์ข…ํ•ฉํ•ด ์ตœ์ข… ์•ˆ์ „์„ฑ ํŒ๋‹จ์„ ๋‚ด๋ฆผ. SL

No Image

Estimation of Conformal Metrics

1. ์ฃผ์š” ๊ธฐ์—ฌ | ๊ตฌ๋ถ„ | ๋‚ด์šฉ | ์˜์˜ | | | | | | ์ •๊ทœ์„ฑ ์ด๋ก  | ์–‘์˜ ๋ฆฌ์น˜์™€ ฮบโ€‘๋ฆฌํ”„์‹œ์ธ  ์ปจํฌ๋ฉ€ ํŒฉํ„ฐ๋งŒ์œผ๋กœ ์ง€์˜ค๋ฐ์‹œ์˜ Cยน,ยน ์ •๊ทœ์„ฑ ํ™•๋ณด | ๊ธฐ์กด ์—ฐ๊ตฌ๋Š” Cยฒ ์„œ๋ธŒ๋งค๋‹ˆํด๋“œ ๊ฐ€์ •์— ์˜์กดํ–ˆ์œผ๋‚˜, ์—ฌ๊ธฐ์„œ๋Š” ๋ฆฌ์น˜๋งŒ ์žˆ์œผ๋ฉด ์ถฉ๋ถ„ํ•จ์„ ์ฆ๋ช…ํ•ด ์ ์šฉ ๋ฒ”์œ„๋ฅผ ํฌ๊ฒŒ ํ™•๋Œ€ | | ๊ทธ๋ž˜ํ”„ ๊ธฐ๋ฐ˜ ๊ทผ์‚ฌ | ์งง์€ ์—ฃ์ง€ ์ œํ•œ๊ณผ ๊ฐ€์ค‘์น˜ ์„ค๊ณ„๋กœ ํด๋ฆฌ๊ณค ๊ฒฝ๋กœ๊ฐ€ ์‹ค์ œ ์ง€์˜ค๋ฐ์‹œ์™€ ๋™์ผํ•œ ์˜ค์ฐจ ํ•œ๊ณ„ | IsomapยทFermat ๊ฑฐ๋ฆฌ์™€ ์—ฐ๊ฒฐ, ์ปจํฌ๋ฉ€ ํŒฉํ„ฐ ๋ฅผ ์ผ๋ฐ˜ํ™”ํ•จ์œผ๋กœ์จ ๋‹ค์–‘ํ•œ ์‘์šฉ(๋ฐ€๋„ ๊ธฐ๋ฐ˜ ๊ฑฐ๋ฆฌ, ์ด๋ฏธ์ง€ ๋งค๋‹ˆํด๋“œ ๋“ฑ)์— ์ง์ ‘ ์ ์šฉ ๊ฐ€๋Šฅ | | ์ˆ˜๋ ด

Mathematics
No Image

EVLP:Learning Unified Embodied Vision-Language Planner with Reinforced Supervised Fine-Tuning

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ฌธ์ œ ์ •์˜ Embodied AI ๋ถ„์•ผ์—์„œ ์žฅ๊ธฐ ์กฐ์ž‘(task decomposition & execution) ์€ โ€œ์–ธ์–ด โ†’ ๋…ผ๋ฆฌ โ†’ ์‹œ๊ฐโ€‘๊ณต๊ฐ„ ์ƒ์ƒ โ†’ ํ–‰๋™โ€ ์˜ ๋ณตํ•ฉ ํ๋ฆ„์„ ์š”๊ตฌํ•œ๋‹ค. ๊ธฐ์กด ํ”Œ๋ž˜๋„ˆ๋“ค์€ ํ…์ŠคํŠธโ€‘์ค‘์‹ฌ ํ˜น์€ ์‹œ๊ฐโ€‘์ค‘์‹ฌ ์ ‘๊ทผ์— ๋จธ๋ฌผ๋Ÿฌ, ๋‘ ๋ชจ๋‹ฌ ๊ฐ„ ์ •ํ•ฉ์„ฑ์„ ํ™•๋ณดํ•˜์ง€ ๋ชปํ•˜๊ณ , ํŠนํžˆ ๋ฉ€ํ‹ฐ์Šคํ… ์ƒํ™ฉ์—์„œ ํ”Œ๋ž˜๋‹ ์ผ๊ด€์„ฑ์ด ํฌ๊ฒŒ ์ €ํ•˜๋œ๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด์™€ ์„ค๊ณ„ | ๊ตฌ์„ฑ ์š”์†Œ | ํ•ต์‹ฌ ๋ฉ”์ปค๋‹ˆ์ฆ˜ | ๊ธฐ๋Œ€ ํšจ๊ณผ | | | | | | ํ†ตํ•ฉ ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ ์ƒ์„ฑ ํ”„๋ ˆ์ž„์›Œํฌ | ์–ธ์–ด ํ† ํฐ๊ณผ ์‹œ๊ฐ ํ”ผ์ฒ˜๋ฅผ ํ•˜๋‚˜์˜ Transfor

Learning
No Image

EvtSlowTV -- A Large and Diverse Dataset for Event-Based Depth Estimation

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ์ด๋ฒคํŠธ ์นด๋ฉ”๋ผ์˜ ๊ฐ•์  : ๊ธฐ์กด ํ”„๋ ˆ์ž„ ๊ธฐ๋ฐ˜ ์นด๋ฉ”๋ผ๋Š” ๊ณ ์ •๋œ ํ”„๋ ˆ์ž„ ๋ ˆ์ดํŠธ์™€ ์ œํ•œ๋œ ๋‹ค์ด๋‚ด๋ฏน ๋ ˆ์ธ์ง€ ๋•Œ๋ฌธ์— ๊ธ‰๊ฒฉํ•œ ์กฐ๋ช… ๋ณ€ํ™”๋‚˜ ๋น ๋ฅธ ์›€์ง์ž„์—์„œ ๋ธ”๋Ÿฌยท๋…ธ์ด์ฆˆ๊ฐ€ ๋ฐœ์ƒํ•œ๋‹ค. ์ด๋ฒคํŠธ ์นด๋ฉ”๋ผ๋Š” ํ”ฝ์…€ ๋‹จ์œ„์˜ ๋น„๋™๊ธฐ ๋ณ€ํ™”๋งŒ์„ ๊ธฐ๋กํ•ด ์ด๋Ÿฌํ•œ ๋ฌธ์ œ๋ฅผ ๊ทผ๋ณธ์ ์œผ๋กœ ํšŒํ”ผํ•œ๋‹ค. ๋ฐ์ดํ„ฐ ๋ถ€์กฑ ๋ฌธ์ œ : ํ˜„์žฌ ๊ณต๊ฐœ๋œ ์ด๋ฒคํŠธ ๋ฐ์ดํ„ฐ์…‹(์˜ˆ: DVSโ€‘128, MVSEC ๋“ฑ)์€ ์ˆ˜์ฒœ๋งŒ ์ด๋ฒคํŠธ ์ˆ˜์ค€์— ๋จธ๋ฌผ๋Ÿฌ, ๋”ฅ๋Ÿฌ๋‹ ๊ธฐ๋ฐ˜ ๋ชจ๋ธ์ด ๋ณต์žกํ•œ ์‹ค์ œ ํ™˜๊ฒฝ์— ์ ์‘ํ•˜๊ธฐ์—” ํ•œ๊ณ„๊ฐ€ ์žˆ๋‹ค. ํŠนํžˆ ๋ผ๋ฒจ๋ง(๊นŠ์ด ์ง€๋„) ๋น„์šฉ์ด ๋†’์•„ ์ž๊ธฐ์ง€๋„ ํ•™์Šต์ด ์š”๊ตฌ๋œ๋‹ค. 2. E

Data
No Image

Exploring Federated Learning for Thermal Urban Feature Segmentation -- A Comparison of Centralized and Decentralized Approaches

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ๋ฐ์ดํ„ฐ ํ”„๋ผ์ด๋ฒ„์‹œยท๋ณด์•ˆ : ์—ด์˜์ƒ์€ ๊ฑด๋ฌผยท์ธํ”„๋ผ์˜ ์—ดํŠน์„ฑ์„ ํฌํ•จํ•ด ๋ฏผ๊ฐํ•œ ์ •๋ณด๋ฅผ ๋‹ด๊ณ  ์žˆ์–ด, ์›๋ณธ ๋ฐ์ดํ„ฐ๋ฅผ ์ค‘์•™ ์„œ๋ฒ„์— ์ „์†กํ•˜๋Š” ๊ฒƒ์ด ๋ฒ•ยท๊ทœ์ œ์ƒ ์–ด๋ ค์›€. ๋ถ„์‚ฐ ๋ฐ์ดํ„ฐ ํŠน์„ฑ : ๋‘ ๋„์‹œ์—์„œ ์ˆ˜์ง‘๋œ UAV ์—ด์˜์ƒ์€ ์ดฌ์˜ ํ™˜๊ฒฝยท๋‚ ์”จยท์„ผ์„œ ์„ค์ •์ด ๋‹ฌ๋ผ ๋ฐ์ดํ„ฐ ๋ถ„ํฌ๊ฐ€ ํฌ๊ฒŒ ๋‹ค๋ฆ„(Nonโ€‘IID). ์ด๋Š” ์ „ํ†ต์ ์ธ CL ๋ชจ๋ธ์ด ๊ณผ์ ํ•ฉํ•˜๊ฑฐ๋‚˜ ์ผ๋ฐ˜ํ™”์— ์‹คํŒจํ•  ์œ„ํ—˜์„ ๋†’์ž„. 2. ๋ฐ์ดํ„ฐ์…‹ ๋ฐ ์ „์ฒ˜๋ฆฌ | ํ•ญ๋ชฉ | ๋‚ด์šฉ | | | | | ์ดฌ์˜ ํ”Œ๋žซํผ | UAV(๋ฉ€ํ‹ฐ๋กœํ„ฐ) ํƒ‘์žฌ ์—ด ์ ์™ธ์„  ์นด๋ฉ”๋ผ | | ์ดฌ์˜ ์ง€์—ญ | ๋…์ผ ๋ฒ ๋ฅผ๋ฆฐ,

Learning
No Image

Fast algorithms enabling optimization and deep learning for photoacoustic tomography in a circular detection geometry

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ๊ด‘์Œํ–ฅ ๋‹จ์ธต์ดฌ์˜์€ ์ดˆ์ŒํŒŒ์™€ ๊ด‘ํ•™์„ ๊ฒฐํ•ฉํ•œ ๊ณ ํ•ด์ƒ๋„ ์˜์ƒ ๊ธฐ์ˆ ๋กœ, ์˜๋ฃŒยท์ƒ๋ฌผํ•™ ๋ถ„์•ผ์—์„œ ์ค‘์š”ํ•œ ์—ญํ• ์„ ํ•œ๋‹ค. ์—ญ๋ฌธ์ œ๋Š” ์ผ๋ฐ˜์ ์œผ๋กœ ์„ ํ˜• ์—ฐ์‚ฐ์ž A (ํฌ์›Œ๋“œ)์™€ Aแต€ (์–ด๋“œ์กฐ์ธํŠธ)๋ฅผ ๋ฐ˜๋ณต ํ˜ธ์ถœํ•˜๋Š” ์ตœ์ ํ™” ๋ฃจํ”„ ์•ˆ์—์„œ ํ•ด๊ฒฐ๋˜๋ฉฐ, ํŠนํžˆ ๋”ฅ๋Ÿฌ๋‹ ๊ธฐ๋ฐ˜ โ€œlearned primalโ€‘dualโ€ ๊ฐ™์€ ๊ตฌ์กฐ๋Š” ์ˆ˜๋ฐฑ ๋ฒˆ์˜ ์—ฐ์‚ฐ์„ ์š”๊ตฌํ•œ๋‹ค. ๊ธฐ์กด ๊ตฌํ˜„์€ FFT ๊ธฐ๋ฐ˜ ๊ฐ€์†์„ ํ™œ์šฉํ•˜๋”๋ผ๋„ O(nยณ) ํ˜น์€ O(nยฒ โˆšn) ์ˆ˜์ค€์œผ๋กœ, ๊ณ ํ•ด์ƒ๋„(์˜ˆ: 1024 ร— 1024) ์ด๋ฏธ์ง€์—์„œ๋Š” ์‹ค์‹œ๊ฐ„ ์ ์šฉ์ด ์–ด๋ ค์› ๋‹ค. 2. ํ•ต์‹ฌ ๊ธฐ์—ฌ ์•Œ๊ณ ๋ฆฌ์ฆ˜ ์„ค

Learning Detection
No Image

Fast Ewald Summation using Prolate Spheroidal Wave Functions

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ Ewald ๋ถ„ํ•  ์€ ์žฅ๊ฑฐ๋ฆฌ ์ƒํ˜ธ์ž‘์šฉ(์ „๊ธฐยท์ ์„ฑ ๋“ฑ)์„ ์‹ค๊ณต๊ฐ„ยทํ‘ธ๋ฆฌ์— ๊ณต๊ฐ„์œผ๋กœ ๋‚˜๋ˆ„์–ด ๊ณ„์‚ฐ๋Ÿ‰์„ O(nยฒ) โ†’ O(n log n) ์œผ๋กœ ๊ฐ์†Œ์‹œํ‚ค๋Š” ํ•ต์‹ฌ ๊ธฐ๋ฒ•์ด๋‹ค. ๊ธฐ์กด ๊ณ ์† Ewald ๋ณ€ํ˜•(PME, SPME, PPPM, SE, PยฒNFFT ๋“ฑ)์€ ๋‘ ๊ฐ€์ง€ ์ž์œ ๋„ ์— ํฌ๊ฒŒ ์˜์กดํ•œ๋‹ค. 1. ๋ชฐ๋ฆฌํ”ผ์ผ€์ดํ„ฐ (Ewald split์—์„œ ์งง์€ยท๊ธด ํŒŒํŠธ ๊ตฌ๋ถ„) 2. ์œˆ๋„์šฐ ํ•จ์ˆ˜ (์ž…์ž โ†’ ๊ฒฉ์ž ์ „ํŒŒ, ์—ญ์ „ํŒŒ) ๊ฐ€์šฐ์‹œ์•ˆ(๋ชฐ๋ฆฌํ”ผ์ผ€์ดํ„ฐ)ยทBโ€‘์Šคํ”Œ๋ผ์ธ(์œˆ๋„์šฐ) ์กฐํ•ฉ์€ ๊ตฌํ˜„์ด ๊ฐ„๋‹จํ•˜์ง€๋งŒ ํ‘ธ๋ฆฌ์— ์ŠคํŽ™ํŠธ๋Ÿผ ๊ฐ์†Œ๊ฐ€ ๋А๋ ค ๋งŽ์€ ํ‘ธ๋ฆฌ์— ๋ชจ๋“œ

Mathematics
Fault Detection in Electrical Distribution System using Autoencoders

Fault Detection in Electrical Distribution System using Autoencoders

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ์ „๋ ฅ๋ง์˜ ์ค‘์š”์„ฑ : ์ „๋ ฅ ๊ณต๊ธ‰์ด ํ˜„๋Œ€ ์‚ฌํšŒ์˜ ๊ธฐ๋ฐ˜์ด๋ฏ€๋กœ ๋ฐฐ์ „๋ง(Electric Distribution System, EDS)์˜ ์•ˆ์ •์„ฑ ํ™•๋ณด๊ฐ€ ํ•„์ˆ˜์ ์ด๋‹ค. ๊ธฐ์กด ๊ณ ์žฅ ํƒ์ง€ ํ•œ๊ณ„ : ์ „ํ†ต์ ์ธ ๋ชจ๋ธ ๊ธฐ๋ฐ˜ยท์‹ ํ˜ธ ์ฒ˜๋ฆฌ ๊ธฐ๋ฐ˜ ๋ฐฉ๋ฒ•์€ ํŒŒ๋ผ๋ฏธํ„ฐ ๊ฐ€์ •์— ์˜์กดํ•ด ์ •ํ™•๋„๊ฐ€ ๋–จ์–ด์ง€๊ณ , ๋‹ค์ˆ˜์˜ PMUยท๋™๊ธฐ ์œ„์ƒ ์ธก์ • ์žฅ์น˜(PMU) ์„ค์น˜๊ฐ€ ํ•„์š”ํ•ด ๋น„์šฉยท๋ณต์žก๋„๊ฐ€ ๋†’๋‹ค. ๋ฐ์ดํ„ฐ ๋ถ€์กฑ ๋ฌธ์ œ : ๊ณ ์žฅ ์ƒํ™ฉ์€ ๋“œ๋ฌผ์–ด ๋ผ๋ฒจ๋ง๋œ ๋ฐ์ดํ„ฐ ํ™•๋ณด๊ฐ€ ์–ด๋ ต๊ณ , ์‹ค์‹œ๊ฐ„ ์ ์šฉ์„ ์œ„ํ•œ ๋น ๋ฅธ ํ•™์Šตยท์ถ”๋ก ์ด ์š”๊ตฌ๋œ๋‹ค. 2. ์ œ์•ˆ ๋ฐฉ๋ฒ• ๊ฐœ์š” | ๋‹จ๊ณ„ | ์„ค๋ช… |

System Electrical Engineering and Systems Science Detection
No Image

FedPoP: Federated Learning Meets Proof of Participation

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ๋””์ง€ํ„ธ ๋ชจ๋ธ์˜ ์ž์‚ฐํ™” : ์ตœ๊ทผ AI ๋ชจ๋ธ์ด ์„œ๋น„์Šคยท๋ผ์ด์„ ์ŠคยทNFT ๋“ฑ์œผ๋กœ ๊ฑฐ๋ž˜๋˜๋Š” ์‚ฌ๋ก€๊ฐ€ ๋Š˜์–ด๋‚˜๋ฉด์„œ, ๋ชจ๋ธ์— ๋Œ€ํ•œ ์†Œ์œ ๊ถŒยท๊ธฐ์—ฌ๋„ ๋ฅผ ์ž…์ฆํ•  ๋ฒ•์ ยท๊ฒฝ์ œ์  ์š”๊ตฌ๊ฐ€ ๊ธ‰์ฆํ•˜๊ณ  ์žˆ๋‹ค. ๊ธฐ์กด ์ฆ๋ช… ๋ฐฉ์‹์˜ ํ•œ๊ณ„ : ๋ธ”๋ก์ฒด์ธ ๊ธฐ๋ฐ˜ ์˜์ง€์‹์ฆ๋ช…(ZKP) ๋“ฑ์€ ๋†’์€ ์—ฐ์‚ฐ ๋น„์šฉ๊ณผ ๋ณต์žกํ•œ ์ธํ”„๋ผ๋ฅผ ์š”๊ตฌํ•œ๋‹ค. ๋˜ํ•œ, ๋Œ€๋ถ€๋ถ„์˜ FL ์‹œ์Šคํ…œ์€ Secure Aggregation ์„ ์‚ฌ์šฉํ•ด ๊ฐœ๋ณ„ ์—…๋ฐ์ดํŠธ๋ฅผ ์ˆจ๊ธฐ๋ฏ€๋กœ, ์ฐธ์—ฌ ์ฆ๋ช…์„ ๋ณ„๋„๋กœ ์„ค๊ณ„ํ•˜๊ธฐ ์–ด๋ ต๋‹ค. 2. FedPoP์˜ ํ•ต์‹ฌ ์•„์ด๋””์–ด | ์š”์†Œ | ์„ค๋ช… | ๊ธฐ์กด ๋ฐฉ์‹ ๋Œ€๋น„ ์ฐจ๋ณ„์  | |

Learning
No Image

FiCABU: A Fisher-Based, Context-Adaptive Machine Unlearning Processor for Edge AI

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ํ”„๋ผ์ด๋ฒ„์‹œ ๊ทœ์ œ์™€ ์—ฃ์ง€ AI : GDPRยทCCPA ๋“ฑ์—์„œ โ€œ์žŠํž ๊ถŒ๋ฆฌโ€๋ฅผ ๋ช…์‹œํ•จ์— ๋”ฐ๋ผ, ํด๋ผ์šฐ๋“œ๊ฐ€ ์•„๋‹Œ ๋””๋ฐ”์ด์Šค ์ž์ฒด์—์„œ ๋ฐ์ดํ„ฐ๋ฅผ ์‚ญ์ œํ•˜๊ณ  ๋ชจ๋ธ์„ ์ฆ‰์‹œ ์—…๋ฐ์ดํŠธํ•ด์•ผ ํ•˜๋Š” ์š”๊ตฌ๊ฐ€ ์ฆ๊ฐ€ํ•˜๊ณ  ์žˆ๋‹ค. ๊ธฐ์กด ๋ฐฉ๋ฒ•์˜ ํ•œ๊ณ„ : ์„œ๋ฒ„โ€‘์ค‘์‹ฌ ์žฌํ•™์Šต, ์™„์ „ ํŒŒ๋ผ๋ฏธํ„ฐ ์žฌ์ดˆ๊ธฐํ™” ๋“ฑ์€ ์—ฐ์‚ฐยท๋ฉ”๋ชจ๋ฆฌยท์ „๋ ฅ ์†Œ๋ชจ๊ฐ€ ํฌ๊ณ , ๋„คํŠธ์›Œํฌ ์ง€์—ฐ์„ ์ดˆ๋ž˜ํ•œ๋‹ค. ์—ฃ์ง€ ๋””๋ฐ”์ด์Šค๋Š” ์ด๋Ÿฌํ•œ ๋น„์šฉ์„ ๊ฐ๋‹นํ•˜๊ธฐ ์–ด๋ ต๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด | ์š”์†Œ | ์„ค๋ช… | ์ฐจ๋ณ„์  | | | | | | Contextโ€‘Adaptive Unlearning | ์‚ญ์ œ ๋Œ€์ƒ ์ƒ˜ํ”Œ

Learning
No Image

Finding Molecules with Specific Properties: Simulated Annealing vs. Evolution

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ชฉ์  ๋น„์„ ํ˜• ๊ด‘ํ•™(NLO) ๋ฌผ์งˆ์˜ ํ•ต์‹ฌ ์„ฑ๋Šฅ ์ง€ํ‘œ์ธ ํ‰๊ท  ์ดˆ๊ทน์„ฑ(ฮฒ)์€ ์‹คํ—˜ยท๊ณ„์‚ฐ ๋ชจ๋‘์—์„œ ๋†’์€ ๊ฐ’์„ ์š”๊ตฌํ•œ๋‹ค. ์ „ํ†ต์ ์ธ ์ „์‚ฐ ํ™”ํ•™์€ ์ฃผ์–ด์ง„ ๊ตฌ์กฐ์—์„œ ๋ฌผ์„ฑ์„ ๊ณ„์‚ฐํ•˜๋Š” ๋ฐ ๊ฐ•์ ์ด ์žˆ์ง€๋งŒ, โ€œ์–ด๋–ค ๊ตฌ์กฐ๊ฐ€ ์›ํ•˜๋Š” ๋ฌผ์„ฑ์„ ๊ฐ€์งˆ๊นŒ?โ€๋ผ๋Š” ์—ญ๋ฌธ์ œ๋Š” ์กฐํ•ฉ ํญ์ด ๋ฐฉ๋Œ€ํ•˜๊ณ  ๋ณ€์ˆ˜๋“ค์ด ์ด์‚ฐ์ ์ด์–ด์„œ ํƒ์ƒ‰์ด ์–ด๋ ค์šด ๋ฌธ์ œ๋‹ค. ๋ณธ ๋…ผ๋ฌธ์€ ์ด๋Ÿฌํ•œ ์—ญ๋ฌธ์ œ์— ๋ฉ”ํƒ€ํœด๋ฆฌ์Šคํ‹ฑ ์ตœ์ ํ™” ๊ธฐ๋ฒ•์„ ์ ์šฉํ•ด, SMILES ๋ฌธ์ž์—ด์ด๋ผ๋Š” ์ด์‚ฐ์  ํ‘œํ˜„์„ ์ง์ ‘ ๋‹ค๋ฃจ๋Š” ๋‘ ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ๋น„๊ตํ•œ๋‹ค. 2. ์•Œ๊ณ ๋ฆฌ์ฆ˜ ์„ค๊ณ„ ์ง„ํ™” ์•Œ๊ณ ๋ฆฌ์ฆ˜ (EA) ์ดˆ๊ธฐ ๊ฐœ์ฒด 10๊ฐœ, ๋งค ์„ธ๋Œ€ 20

Physics
Finite elements for the space approximation of a differential model for salts crystallization

Finite elements for the space approximation of a differential model for salts crystallization

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ์„์žฌยท๋ฒฝ๋Œ ๋“ฑ ๋ฌธํ™”์œ ์‚ฐ ์žฌ๋ฃŒ๋Š” ๋‹ค๊ณต์„ฑ ๋งค์ฒด์ด๋ฉฐ, ์Šต๊ธฐยท์—ผ๋ถ„์— ์˜ํ•ด ๋‚ด๋ถ€์— ์†Œ๊ธˆ ๊ฒฐ์ •์ด ํ˜•์„ฑ๋ผ ๊ธฐ๊ณต์„ ๋ง‰๊ณ  ๋ฌผ ์ด๋™์„ ์ €ํ•ดํ•œ๋‹ค. ๊ธฐ์กด ๋ชจ๋ธ์€ 1์ฐจ์›์œผ๋กœ๋งŒ ์ œ์‹œ๋ผ ์‹ค์ œ ๊ตฌ์กฐ๋ฌผ์˜ ๋ณตํ•ฉ์ ์ธ ๋ฌผ๋ฆฌยทํ™”ํ•™ ํ˜„์ƒ์„ ํฌ์ฐฉํ•˜๊ธฐ์— ํ•œ๊ณ„๊ฐ€ ์žˆ์—ˆ๋‹ค. ๋‹ค์ฐจ์› ๋ชจ๋ธ๋ง์€ ๋ชจ์„ธ๊ด€ ์ƒ์Šน, ์ธก๋ฉดยท์ƒ๋ถ€ยทํ•˜๋ถ€ ๊ฒฝ๊ณ„ ์กฐ๊ฑด ๋“ฑ ์‹ค์ œ ์‹คํ—˜ยทํ˜„์žฅ ์ƒํ™ฉ์„ ๋ณด๋‹ค ์ •๋ฐ€ํ•˜๊ฒŒ ์žฌํ˜„ํ•  ์ˆ˜ ์žˆ๋‹ค. 2. ์ˆ˜ํ•™์  ๋ชจ๋ธ ๋ฏธ์ง€ ๋ณ€์ˆ˜ : ์•ก์ฒด ์ฒด์ ๋ถ„์œจ ฮธโ‚—, ์šฉํ•ด๋œ ์—ผ๋ถ„ ๋†๋„ cแตข, ๊ณ ์ฒด(๊ฒฐ์ •) ๋†๋„ cโ‚›, ๋‹ค๊ณต์„ฑ n(ํฌ์–ด์‹œํ‹ฐ). ๊ฒฐํ•ฉ ๊ตฌ์กฐ : ฮธโ‚— ๋ฐฉ์ •์‹ โ†’ ๋น„์„ ํ˜• ํ™•

Model Mathematics
No Image

Focused Relative Risk Information Criterion for Variable Selection in Linear Regression

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ์ „ํ†ต์ ์ธ ๋ณ€์ˆ˜ ์„ ํƒ(์˜ˆ: AIC, BIC, Mallows (C p))์€ ์ „์ฒด ๋ชจ๋ธ ์ ํ•ฉ๋„์™€ ๋ณต์žก๋„๋งŒ์„ ๊ณ ๋ คํ•œ๋‹ค. Focused Information Criterion (FIC) ๊ณ„์—ด์€ ํŠน์ • โ€œํฌ์ปค์Šค ํŒŒ๋ผ๋ฏธํ„ฐโ€(์˜ˆ: ํŠน์ • ํ™˜์ž์˜ ๊ธฐ๋Œ€๊ฐ’) ์— ๋Œ€ํ•œ ์ถ”์ • ์ •ํ™•๋„๋ฅผ ์ตœ์šฐ์„ ์œผ๋กœ ํ•œ๋‹ค๋Š” ์ ์—์„œ ์ฐจ๋ณ„ํ™”๋œ๋‹ค. ๊ธฐ์กด FIC๋Š” ๋Œ€์ฒด๋กœ ๋Œ€๊ทœ๋ชจ ๊ทผ์‚ฌ(largeโ€‘sample approximation) ์— ์˜์กดํ•ด ์ •ํ™•๋„๊ฐ€ ์ œํ•œ์ ์ด์—ˆ๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด โ€“ FRIC | ์š”์†Œ | ์„ค๋ช… | | | | | Relative Risk

Statistics
No Image

FP8-Flow-MoE: A Casting-Free FP8 Recipe without Double Quantization Error

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ฌธ์ œ ์ •์˜ MoE ๋ชจ๋ธ์˜ ๊ทœ๋ชจ : ํ˜„์žฌ ์ตœ์ฒจ๋‹จ MoE ๋ชจ๋ธ์€ ์ˆ˜๋ฐฑ ์–ต์—์„œ ์ˆ˜์ฒœ ์–ต ํŒŒ๋ผ๋ฏธํ„ฐ์— ๋‹ฌํ•˜๋ฉฐ, GPU ๋ฉ”๋ชจ๋ฆฌ์™€ ์—ฐ์‚ฐ๋Ÿ‰์ด ๋ณ‘๋ชฉ์ด ๋ฉ๋‹ˆ๋‹ค. ์ €์ •๋ฐ€๋„ ํ•™์Šต์˜ ํ•œ๊ณ„ : BF16์€ ๋ฉ”๋ชจ๋ฆฌ ์ ˆ๊ฐ ํšจ๊ณผ๊ฐ€ ์žˆ์ง€๋งŒ, ์‹ค์ œ ์—ฐ์‚ฐ์—์„œ๋Š” ์—ฌ์ „ํžˆ FP32โ†”BF16 ์‚ฌ์ด์˜ ์บ์ŠคํŒ…์ด ๋นˆ๋ฒˆํžˆ ๋ฐœ์ƒํ•ฉ๋‹ˆ๋‹ค. ์ด๋Š” ๋ฉ”๋ชจ๋ฆฌ ๋Œ€์—ญํญ๊ณผ ์—ฐ์‚ฐ ํŒŒ์ดํ”„๋ผ์ธ์„ ๋น„ํšจ์œจ์ ์œผ๋กœ ๋งŒ๋“ค๊ณ , FP8์˜ ์ž ์žฌ์  ์ด๋“์„ ๊ฐ์†Œ์‹œํ‚ต๋‹ˆ๋‹ค. ์ด์ค‘ ์–‘์žํ™” ์˜ค๋ฅ˜ : ์„œ๋กœ ๋‹ค๋ฅธ ์ฐจ์›(์˜ˆ: ํ–‰/์—ด)์—์„œ ๋…๋ฆฝ์ ์œผ๋กœ ์Šค์ผ€์ผ์„ ์ ์šฉํ•˜๋ฉด, ๊ฐ™์€ ํ…์„œ๊ฐ€ ์—ฌ๋Ÿฌ ๋ฒˆ ์–‘์žํ™”๋  ๋•Œ ์Šค์ผ€์ผ์ด ๋ถˆ์ผ์น˜

No Image

From Model Training to Model Raising

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ฌธ์ œ ์ œ๊ธฐ ํ˜„ํ–‰ ํ›ˆ๋ จ ๋ฐฉ์‹์˜ ํ•œ๊ณ„ : ๋Œ€๋ถ€๋ถ„์˜ ๋Œ€ํ˜• ์–ธ์–ด ๋ชจ๋ธ(LLM)์€ ๋Œ€๊ทœ๋ชจ ๋น„์ง€๋„ ํ•™์Šต ํ›„์— ๋ณ„๋„์˜ ์ •๋ ฌ ๋‹จ๊ณ„(์˜ˆ: RLHF)๋ฅผ ๊ฑฐ์นœ๋‹ค. ์ด ๊ณผ์ •์€ โ€˜๋Šฅ๋ ฅ โ†’ ์ •๋ ฌโ€™ ์ˆœ์„œ๋กœ, ์ด๋ฏธ ํ˜•์„ฑ๋œ ๋Šฅ๋ ฅ์— ๊ฐ€์น˜๊ฐ€ ๋’ค๋Šฆ๊ฒŒ ์ž…ํ˜€์ง€๋Š” ๊ตฌ์กฐ๋‹ค. ์˜ค์ •๋ ฌ ์œ„ํ—˜ : ๋Šฅ๋ ฅ์ด ๋จผ์ € ๊ณ ์ •๋˜๋ฉด, ์ดํ›„ ์ •๋ ฌ ๊ณผ์ •์—์„œ ๊ฐ€์น˜์™€ ๋Šฅ๋ ฅ์„ ๋ถ„๋ฆฌํ•˜๋ ค๋Š” ์‹œ๋„๊ฐ€ ํ•„์š”ํ•ด, ๋ชจ๋ธ์ด ์˜๋„์น˜ ์•Š์€ ํ–‰๋™์„ ๋ณด์ผ ํ™•๋ฅ ์ด ๋†’์•„์ง„๋‹ค. 2. โ€œ๋ชจ๋ธ ์–‘์œก(Model Raising)โ€ ๊ฐœ๋… ์ •์˜ : ๋ชจ๋ธ์„ โ€˜์•„์ดโ€™๋ฅผ ํ‚ค์šฐ๋“ฏ์ด, ์ดˆ๊ธฐ ํ•™์Šต ๋‹จ๊ณ„๋ถ€ํ„ฐ ๊ฐ€์น˜์™€ ํ–‰๋™ ๊ทœ๋ฒ”์„ ํ•จ๊ป˜

Model
From Theory to Throughput: CUDA-Optimized APML for Large-Batch 3D Learning

From Theory to Throughput: CUDA-Optimized APML for Large-Batch 3D Learning

๋ณธ ์—ฐ๊ตฌ๋Š” 3์ฐจ์› ํฌ์ธํŠธ ํด๋ผ์šฐ๋“œ ์ฒ˜๋ฆฌ์—์„œ ์†์‹ค ํ•จ์ˆ˜ ์„ ํƒ์ด ๋ชจ๋ธ ์„ฑ๋Šฅ๊ณผ ํ•™์Šต ํšจ์œจ์— ๋ฏธ์น˜๋Š” ์˜ํ–ฅ์„ ์‹ฌ๋„ ์žˆ๊ฒŒ ํƒ๊ตฌํ•œ๋‹ค. ๊ธฐ์กด์— ๋„๋ฆฌ ์‚ฌ์šฉ๋˜๋Š” Chamfer Distance๋Š” ๊ฐ ํฌ์ธํŠธ๋ฅผ ๊ฐ€์žฅ ๊ฐ€๊นŒ์šด ์ƒ๋Œ€ ํฌ์ธํŠธ์™€ ๋งคํ•‘ํ•˜๋Š” ๋ฐฉ์‹์œผ๋กœ, ๊ณ„์‚ฐ๋Ÿ‰์ด O(N) ์ˆ˜์ค€์— ๋จธ๋ฌผ๋Ÿฌ ์‹ค์‹œ๊ฐ„ ์‘์šฉ์— ์ ํ•ฉํ•˜์ง€๋งŒ, ๋‹ค๋Œ€์ผ ๋งคํ•‘์ด ํ—ˆ์šฉ๋˜๊ธฐ ๋•Œ๋ฌธ์— ์‹ค์ œ ๊ธฐํ•˜ํ•™์  ์ฐจ์ด๋ฅผ ๊ณผ์†Œํ‰๊ฐ€ํ•˜๋Š” ๊ฒฝํ–ฅ์ด ์žˆ๋‹ค. ๋ฐ˜๋ฉด Earth Mover Distance(EMD)๋Š” ์ตœ์  ์ˆ˜์†ก ๋ฌธ์ œ๋ฅผ ํ’€์–ด ์ผ๋Œ€์ผ ๋งคํ•‘์„ ๋ณด์žฅํ•˜๋ฏ€๋กœ ๊ธฐํ•˜ํ•™์  ์ •ํ™•๋„๊ฐ€ ๋›ฐ์–ด๋‚˜์ง€๋งŒ, ์ตœ์ ํ™” ๊ณผ์ •์ด O(Nยณ) ์ •๋„์˜

Learning
Functional Decomposition and Shapley Interactions for Interpreting Survival Models

Functional Decomposition and Shapley Interactions for Interpreting Survival Models

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ์ž„์ƒยท๊ณต์ค‘๋ณด๊ฑด ์—์„œ ํŠน์„ฑ ๊ฐ„ ์ƒํ˜ธ์ž‘์šฉ (์˜ˆ: ์œ ์ „ยทํ™˜๊ฒฝ, ๋น„๋งŒยท์น˜๋ฃŒ ํšจ๊ณผ ๋“ฑ)์€ ์˜์‚ฌ๊ฒฐ์ •์— ํ•ต์‹ฌ์ ์ธ ์ •๋ณด๋ฅผ ์ œ๊ณตํ•œ๋‹ค. ๊ธฐ์กด ํ†ต๊ณ„ ๋ชจ๋ธ(CoxPH, Aalen ๋“ฑ)์€ ๋ช…์‹œ์  ๊ต์ฐจํ•ญ ์„ ํ†ตํ•ด ์ƒํ˜ธ์ž‘์šฉ์„ ํƒ์ƒ‰ํ•˜์ง€๋งŒ, ๋น„์„ ํ˜•ยท์‹œ๊ฐ„โ€‘๋ณ€ํ™” ํšจ๊ณผ๋ฅผ ์ž๋™์œผ๋กœ ํฌ์ฐฉํ•˜์ง€ ๋ชปํ•œ๋‹ค. ์ตœ์‹  ์ƒ์กด ๋จธ์‹ ๋Ÿฌ๋‹(DeepSurv, Random Survival Forest ๋“ฑ)์€ ๋ณต์žกํ•œ ํŒจํ„ด์„ ํ•™์Šตํ•˜์ง€๋งŒ โ€˜๋ธ”๋ž™๋ฐ•์Šคโ€™ ํŠน์„ฑ์œผ๋กœ ํ•ด์„์ด ์ œํ•œ๋œ๋‹ค. ๊ธฐ๋Šฅ ๋ถ„ํ•ด(FD) ๋Š” ์˜ˆ์ธก ํ•จ์ˆ˜๋ฅผ ์ฃผ ํšจ๊ณผ์™€ ์ƒํ˜ธ์ž‘์šฉ ํšจ๊ณผ๋กœ ๊ฐ€๋ฒ•์ ์œผ๋กœ ๋ถ„๋ฆฌํ•˜๋Š” ์ด๋ก ์  ๋„๊ตฌ์ด์ง€๋งŒ

Machine Learning Statistics Model
No Image

GAIA: Geothermal Analytics and Intelligent Agent

1. ์—ฐ๊ตฌ์˜ ์ฃผ์š” ๊ธฐ์—ฌ | ๊ตฌ๋ถ„ | ๋‚ด์šฉ | ์˜์˜ | | | | | | ํ†ตํ•ฉ AI ํ”Œ๋žซํผ | ์ง€์งˆยท์ง€๊ตฌ๋ฌผ๋ฆฌยท์ €๋ฅ˜๊ณตํ•™ ๋ฐ์ดํ„ฐ๋ฅผ ํ•˜๋‚˜์˜ ์—์ด์ „ํŠธ ๊ธฐ๋ฐ˜ ์›Œํฌํ”Œ๋กœ์šฐ์— ํ†ตํ•ฉ | ๋‹คํ•™์ œ ํ˜‘์—… ๋น„์šฉ ์ ˆ๊ฐ ๋ฐ ์˜์‚ฌ๊ฒฐ์ • ์†๋„ ํ–ฅ์ƒ | | RAG ๊ธฐ๋ฐ˜ ์—์ด์ „ํŠธ | LLM์ด ์ง€์‹๋ฒ ์ด์Šค๋ฅผ ์ž๋™ ์งˆ์˜ํ•˜๊ณ , ๋‹ค๋‹จ๊ณ„ ๋ถ„์„ ํŒŒ์ดํ”„๋ผ์ธ์„ ์ž๋™ ์ƒ์„ฑ | ์ธ๊ฐ„์ด ์„ค๊ณ„ํ•˜๋˜ ๋ณต์žกํ•œ ํŒŒ์ดํ”„๋ผ์ธ์„ ์ž๋™ํ™”, ์žฌํ˜„์„ฑ ํ™•๋ณด | | ๋””์ง€ํ„ธ ํŠธ์œˆ | ๊ณ ์ „ ๋ฌผ๋ฆฌ ๋ชจ๋ธ + ์„œ๋ผ์šด๋“œ ๋ชจ๋ธ(๋Œ€๋ฆฌ ๋ชจ๋ธ) + ๋„๋ฉ”์ธ ์„œ๋ธŒ๋ฃจํ‹ด | ๋น ๋ฅธ ์‹œ๋ฎฌ๋ ˆ์ด์…˜๊ณผ ์‹ค์‹œ๊ฐ„ ์˜ˆ์ธก์ด ๊ฐ€๋Šฅ, ํ˜„์žฅ ์šด์˜์— ๋ฐ”๋กœ ์ ์šฉ

Generalized bilinear Koopman realization from input-output data for multi-step prediction with metaheuristic optimization of lifting function and its application to real-world industrial system

Generalized bilinear Koopman realization from input-output data for multi-step prediction with metaheuristic optimization of lifting function and its application to real-world industrial system

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ์‚ฐ์—… ํ˜„์žฅ์˜ ์„ผ์„œ ์ œ์•ฝ : ์ „์ฒด ์ƒํƒœ๋ฅผ ์ธก์ •ํ•˜๊ธฐ ์–ด๋ ค์šด ์‹ค์ œ ๊ณต์ •์—์„  ์ž…๋ ฅโ€‘์ถœ๋ ฅ ๋ฐ์ดํ„ฐ๋งŒ์œผ๋กœ ๋ชจ๋ธ์„ ๊ตฌ์ถ•ํ•ด์•ผ ํ•จ. ๊ธฐ์กด Koopman ์—ฐ๊ตฌ๋Š” ์ „์ œ์กฐ๊ฑด์œผ๋กœ ์ „ ์ƒํƒœ ๊ฐ€์šฉ์„ฑ์„ ๊ฐ€์ •ํ•˜๋Š” ๊ฒฝ์šฐ๊ฐ€ ๋งŽ์•„ ์‚ฐ์—… ์ ์šฉ์— ํ•œ๊ณ„๊ฐ€ ์žˆ์—ˆ๋‹ค. ๋ฆฌํ”„ํŒ… ํ•จ์ˆ˜ ์„ค๊ณ„์˜ ๋‚œ์  : RBF, ๋‹คํ•ญ์‹, ๊ธฐ๊ณ„ ์˜๊ฐ ๊ธฐ๋ฐ˜ ๋“ฑ ๋‹ค์–‘ํ•œ ๋ฐฉ์‹์ด ์ œ์•ˆ๋์ง€๋งŒ, ์ตœ์  ํŒŒ๋ผ๋ฏธํ„ฐ ์„ ํƒ์ด ์ˆ˜๋™์ ์ด๋ฉฐ ์žฌํ˜„์„ฑ์ด ๋–จ์–ด์ง„๋‹ค. ์ตœ๊ทผ NN ๊ธฐ๋ฐ˜ ์ ‘๊ทผ๋„ ํ•˜์ดํผํŒŒ๋ผ๋ฏธํ„ฐ ํŠœ๋‹ยท๊ณผ์ ํ•ฉยทํ•™์Šต ๋น„์šฉ ๋“ฑ ์‹ค์šฉ์  ๋ฌธ์ œ๋ฅผ ์•ˆ๊ณ  ์žˆ๋‹ค. LTI Koopman์˜ ํ•œ๊ณ„ : ๋น„์„ ํ˜• ์ œ์–ดโ€‘affi

System Electrical Engineering and Systems Science Data
No Image

Generalized Leverage Score for Scalable Assessment of Privacy Vulnerability

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ํ˜„๋Œ€ ๋”ฅ๋Ÿฌ๋‹ ๋ชจ๋ธ์€ ํ•™์Šต ๋ฐ์ดํ„ฐ์˜ ์ผ๋ถ€๋ฅผ ๊ธฐ์–ต(memorizโ€‹e) ํ•˜๋ฉฐ, ์ด๋Š” ๋ฉค๋ฒ„์‹ญ ์ถ”๋ก  ๊ณต๊ฒฉ์„ ํ†ตํ•ด ๊ฐœ์ธ ์ •๋ณด๊ฐ€ ๋…ธ์ถœ๋  ์œ„ํ—˜์„ ๋งŒ๋“ ๋‹ค. ๊ธฐ์กด ๊ฐœ๋ณ„ ์œ„ํ—˜ ํ‰๊ฐ€ ๋ฐฉ๋ฒ•์€ ๊ทธ๋ฆผ์ž ๋ชจ๋ธ ์„ ์—ฌ๋Ÿฌ ๋ฒˆ ์žฌํ•™์Šตํ•ด์•ผ ํ•˜๋Š” ๋†’์€ ๊ณ„์‚ฐ ๋น„์šฉ์ด ํฐ ์žฅ์• ๋ฌผ์ด๋‹ค. ๋”ฐ๋ผ์„œ โ€œ ์žฌํ•™์Šต ์—†์ด ๊ฐœ๋ณ„ ๋ฐ์ดํ„ฐ ํฌ์ธํŠธ์˜ ํ”„๋ผ์ด๋ฒ„์‹œ ์ทจ์•ฝ์„ฑ์„ ์ธก์ •ํ•  ์ˆ˜ ์žˆ๋Š”๊ฐ€?โ€๋ผ๋Š” ์งˆ๋ฌธ์ด ํ•ต์‹ฌ ์—ฐ๊ตฌ ๋ชฉํ‘œ๊ฐ€ ๋œ๋‹ค. 2. ํ•ต์‹ฌ ์ด๋ก ์  ๊ธฐ์—ฌ | ๊ตฌ๋ถ„ | ๋‚ด์šฉ | ์˜๋ฏธ | | | | | | ๋ ˆ๋ฒ„๋ฆฌ์ง€ ์ ์ˆ˜์™€ MIA | ๊ณ ์ • ์„ค๊ณ„ ๊ฐ€์šฐ์‹œ์•ˆ ์„ ํ˜• ํšŒ๊ท€์—์„œ ๋ ˆ๋ฒ„๋ฆฌ์ง€ ์ ์ˆ˜

Machine Learning Statistics
No Image

Generating quantum entanglement from sunlight

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ์—๋„ˆ์ง€ ๋ฌธ์ œ : ํ˜„์žฌ ์–‘์ž ์ปดํ“จํŒ…ยทํ†ต์‹ ยท์„ผ์„œ๋Š” ๋ƒ‰๊ฐยท๊ณ ์ „์••ยท๋ ˆ์ด์ € ๊ตฌ๋™ ๋“ฑ์œผ๋กœ ์ˆ˜ kW ์ˆ˜์ค€์˜ ์ „๋ ฅ์„ ์†Œ๋ชจํ•œ๋‹ค. ์ „ ์„ธ๊ณ„ ICT ๋ถ€๋ฌธ์˜ ์˜จ์‹ค๊ฐ€์Šค ๋ฐฐ์ถœ๋Ÿ‰์ด 1.8~3.9 %์— ๋‹ฌํ•˜๋Š” ์ƒํ™ฉ์—์„œ, ์–‘์ž ๊ธฐ์ˆ ์˜ ๋Œ€๊ทœ๋ชจ ์ƒ์šฉํ™”๋Š” ์—๋„ˆ์ง€ยทํ™˜๊ฒฝ ์ธก๋ฉด์—์„œ ํฐ ๋„์ „์ด๋‹ค. ๊ด‘์ž ์–ฝํž˜ ์ƒ์„ฑ์˜ ์ „ํ†ต์  ๋ฐฉ์‹ : ๋Œ€๋ถ€๋ถ„์˜ ๊ด‘์ž ๊ธฐ๋ฐ˜ ์–‘์ž ์‹œ์Šคํ…œ์€ ๊ณ ์ฝ”ํžˆ๋Ÿฐ์Šค ๋ ˆ์ด์ €(์ˆ˜ mW ์ˆ˜์ค€)๋กœ SPDC๋ฅผ ๊ตฌ๋™ํ•œ๋‹ค. ๋ ˆ์ด์ €๋Š” ์ „๊ธฐโ€‘๊ด‘ ๋ณ€ํ™˜ ํšจ์œจ์ด ๋‚ฎ๊ณ , ์˜จ๋„ยท์ „๋ฅ˜ ์•ˆ์ •ํ™”์— ๋งŽ์€ ์ „๋ ฅ์ด ํ•„์š”ํ•˜๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด ๋น„์ฝ”ํžˆ๋Ÿฐ์Šค ๊ด‘์› ํ™œ์šฉ ๊ฐ€๋Šฅ์„ฑ :

Quantum Physics
No Image

Grouping Nodes With Known Value Differences: A Lossless UCT-based Abstraction Algorithm

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ MCTS์™€ ์ƒ˜ํ”Œ ํšจ์œจ์„ฑ : MCTS๋Š” ํƒ์ƒ‰ ๋น„์šฉ์ด ํฐ ๊ฒŒ์ž„ยท์‹œ๋ฎฌ๋ ˆ์ด์…˜์—์„œ ๋„๋ฆฌ ์“ฐ์ด์ง€๋งŒ, ํƒ์ƒ‰ ๊นŠ์ด๊ฐ€ ์–•์•„์ง€๋ฉด ์„ฑ๋Šฅ์ด ๊ธ‰๊ฒฉํžˆ ๋–จ์–ด์ง„๋‹ค. ์ถ”์ƒํ™”์˜ ํ•„์š”์„ฑ : ๋™์ผํ•˜๊ฑฐ๋‚˜ ์œ ์‚ฌํ•œ ๊ฐ€์น˜(๊ฐ€์น˜ ๋“ฑ๊ฐ€)์ธ ์ƒํƒœยท์•ก์…˜์„ ํ•˜๋‚˜๋กœ ๋ฌถ์–ด ํ†ต๊ณ„๋Ÿ‰์„ ๊ณต์œ ํ•˜๋ฉด, ๋™์ผํ•œ ์ •๋ณด๋ฅผ ์—ฌ๋Ÿฌ ๋ฒˆ ์ˆ˜์ง‘ํ•  ํ•„์š”๊ฐ€ ์—†์–ด ์ƒ˜ํ”Œ ํšจ์œจ์ด ํฌ๊ฒŒ ํ–ฅ์ƒ๋œ๋‹ค. ASAP์˜ ํ•œ๊ณ„ : ASAP์€ โ€œ์ฆ‰์‹œ ๋ณด์ƒ์ด ๋™์ผโ€์ด๋ผ๋Š” ๊ฐ•์ œ ์กฐ๊ฑด์„ ๋‘์–ด, ์‹ค์ œ๋กœ๋Š” ๊ฐ€์น˜ ์ฐจ์ด๊ฐ€ ์กด์žฌํ•˜์ง€๋งŒ ๋ณด์ƒ ์ฐจ์ด๊ฐ€ ์—†์„ ๊ฒฝ์šฐ์—๋งŒ ์ถ”์ƒ์ด ๊ฐ€๋Šฅํ•˜๋‹ค. ์ด๋Š” ํŠนํžˆ ๋ณด์ƒ์ด ํฌ์†Œํ•˜๊ฑฐ๋‚˜ ๋น„๋Œ€์นญ์ ์ธ deter

No Image

Higher-Order Hit-&-Run Samplers for Linearly Constrained Densities

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ์ œ์•ฝ๋œ ๋„๋ฉ”์ธ์—์„œ์˜ MCMC : ๋ฒ ์ด์ฆˆ ์—ญ๋ฌธ์ œ, ๋Œ€๊ทœ๋ชจ ์ตœ์ ํ™”, ์‹œ์Šคํ…œ ์ƒ๋ฌผํ•™ ๋“ฑ์—์„œ ๋ณ€์ˆ˜๋Š” ์ข…์ข… `Ax โ‰ค b` ํ˜•ํƒœ์˜ ์„ ํ˜• ์ œ์•ฝ์„ ๋งŒ์กฑํ•ด์•ผ ํ•œ๋‹ค. ๊ธฐ์กด ๋ฐฉ๋ฒ•์˜ ํ•œ๊ณ„ : HR ๋Š” ์ œ์•ฝ์„ ์™„๋ฒฝํžˆ ๋งŒ์กฑํ•˜์ง€๋งŒ 0์ฐจ(๊ท ๋“ฑ) ์ œ์•ˆ์— ๋จธ๋ฌผ๋Ÿฌ ๋น„๊ท ๋“ฑ ๋ฐ€๋„์—์„œ๋Š” ํšจ์œจ์ด ๋‚ฎ๋‹ค. MALA/HMC ๋“ฑ ๊ณ ์ฐจ ์ •๋ณด ํ™œ์šฉ ์ƒ˜ํ”Œ๋Ÿฌ๋Š” ์ œ์•ฝ์„ ๋ฌด์‹œํ•˜๋ฉด ์ œ์•ˆ์ด ๋Œ€๋ถ€๋ถ„ ๋ถˆ๊ฐ€๋Šฅํ•ด์ ธ ์ˆ˜์šฉ๋ฅ ์ด ๊ธ‰๊ฒฉํžˆ ๊ฐ์†Œํ•œ๋‹ค. MAPLA ๋Š” ์ œ์•ฝ ์ •๋ณด๋ฅผ ๋ฉ”ํŠธ๋ฆญ์œผ๋กœ ๋ฐ˜์˜ํ•˜์ง€๋งŒ ์—ฌ์ „ํžˆ ๊ฐ€์šฐ์‹œ์•ˆ ์ œ์•ˆ์ด๋ฏ€๋กœ ๋ถˆ๊ฐ€๋Šฅํ•œ ์ œ์•ˆ์ด ์กด์žฌํ•œ๋‹ค. ๋”ฐ๋ผ์„œ โ€œ์ œ์•ˆ์ด ๋ฐ˜๋“œ์‹œ ์ œ์•ฝ ๋‚ด

Statistics
No Image

History-Aware Reasoning for GUI Agents

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ LLM ์ด GUI ์ž๋™ํ™”์— ์ ์šฉ๋˜๋ฉด์„œ ๋ณต์žกํ•œ ์‹œ๊ฐยท์–ธ์–ด ์ •๋ณด๋ฅผ ๋™์‹œ์— ์ฒ˜๋ฆฌํ•  ์ˆ˜ ์žˆ๊ฒŒ ๋˜์—ˆ์ง€๋งŒ, ์—ํ”ผ์†Œ๋“œ ์ˆ˜์ค€์˜ ์—ฐ์†์„ฑ ์„ ์œ ์ง€ํ•˜๋Š” ๊ฒƒ์ด ์—ฌ์ „ํžˆ ์–ด๋ ค์›€. ๊ธฐ์กด RL + CoT ์ ‘๊ทผ๋ฒ•์€ ๋‹จ์ผ ํ™”๋ฉด ์— ๋Œ€ํ•œ ์ถ”๋ก ์€ ๊ฐ•ํ™”ํ–ˆ์ง€๋งŒ, ์‹œ๊ฐ„์  ๋งฅ๋ฝ(ํžˆ์Šคํ† ๋ฆฌ) ์„ ์ถฉ๋ถ„ํžˆ ํ™œ์šฉํ•˜์ง€ ๋ชปํ•จ. ์ด๋Š” ํŠนํžˆ ์žฅ๊ธฐ(๋‹ค๋‹จ๊ณ„) ์ž‘์—… ์—์„œ ๋ˆ„์  ์˜ค๋ฅ˜๋ฅผ ์ดˆ๋ž˜. 2. ํ•ต์‹ฌ ๊ธฐ์—ฌ | ๋ฒˆํ˜ธ | ๊ธฐ์—ฌ ๋‚ด์šฉ | ์˜์˜ | | | | | | 1 | Historyโ€‘Aware Reasoning (HAR) ํ”„๋ ˆ์ž„์›Œํฌ ์„ค๊ณ„ | ์—์ด์ „ํŠธ๊ฐ€ ๊ณผ๊ฑฐ ์ƒํ˜ธ์ž‘์šฉ

How Well Do Large-Scale Chemical Language Models Transfer to Downstream Tasks?

How Well Do Large-Scale Chemical Language Models Transfer to Downstream Tasks?

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ํ™”ํ•™ ์–ธ์–ด ๋ชจ๋ธ(ํด๋ž˜์Šค) : SMILES ๋ฌธ์ž์—ด์„ ํ† ํฐ ์‹œํ€€์Šค๋กœ ์ทจ๊ธ‰ํ•ด NLP ๊ธฐ๋ฒ•(ํŠนํžˆ Transformer ๊ธฐ๋ฐ˜) ์ ์šฉ. ์Šค์ผ€์ผ๋ง ๊ธฐ๋Œ€ : NLP์—์„œ โ€œํŒŒ๋ผ๋ฏธํ„ฐยท๋ฐ์ดํ„ฐยท์ปดํ“จํŒ… ๊ทœ๋ชจ โ†‘ โ†’ ์†์‹ค โ†“ โ†’ ๋‹ค์šด์ŠคํŠธ๋ฆผ ์„ฑ๋Šฅ โ†‘โ€๋ผ๋Š” ๊ฒฝํ—˜๋ฒ•์น™์ด ํ™”ํ•™ ๋ถ„์•ผ์—๋„ ๊ทธ๋Œ€๋กœ ์ ์šฉ๋  ๊ฒƒ์ด๋ผ๋Š” ๊ฐ€์ •์ด ๋„๋ฆฌ ํผ์ ธ ์žˆ์—ˆ๋‹ค. ๋ฌธ์ œ ์ œ๊ธฐ : ์ตœ๊ทผ LLM ์—ฐ๊ตฌ์—์„œ โ€œ์†์‹ค ๊ฐ์†Œ โ‰  ์ „์ด ์„ฑ๋Šฅ ํ–ฅ์ƒโ€์ด๋ผ๋Š” ๋ถ€์ •์  ์ „์ด๊ฐ€ ๋ณด๊ณ ๋˜์—ˆ์œผ๋ฉฐ, ํ™”ํ•™ ๋ถ„์•ผ์—์„œ๋„ ์ด๋ฅผ ๊ฒ€์ฆํ•  ์ฒด๊ณ„์  ์‹คํ—˜์ด ๋ถ€์กฑํ–ˆ๋‹ค. 2. ์‹คํ—˜ ์„ค๊ณ„ | ๋ณ€์ˆ˜ | ๋ฒ”์œ„ | ์„ค๋ช… | |

Machine Learning Computer Science Model
Humanlike AI Design Increases Anthropomorphism but Yields Divergent Outcomes on Engagement and Trust Globally

Humanlike AI Design Increases Anthropomorphism but Yields Divergent Outcomes on Engagement and Trust Globally

์ด ๋…ผ๋ฌธ์€ ์ธ๊ฐ„ํ˜• AI์— ๋Œ€ํ•œ ์ธ๊ฐ„์˜ ์ธ์ง€์  ๋ฐ˜์‘์„ ์‹ค์ฆ์ ์œผ๋กœ ๊ฒ€์ฆํ•œ ์ตœ์ดˆ์˜ ๋Œ€๊ทœ๋ชจ ๊ตญ์ œ ์—ฐ๊ตฌ ์ค‘ ํ•˜๋‚˜์ด๋‹ค. ๋จผ์ € ์—ฐ๊ตฌ์ง„์€ 10๊ฐœ ๊ตญ๊ฐ€(๋ถ๋ฏธยท๋‚จ๋ฏธยท์•„์‹œ์•„ยท์œ ๋Ÿฝยท์•„ํ”„๋ฆฌ์นด ๋“ฑ)์—์„œ ์ด 3,500๋ช…์˜ ์ฐธ๊ฐ€์ž๋ฅผ ๋ชจ์ง‘ํ•˜์—ฌ, ์ตœ์‹  ์ž์—ฐ์–ด ์ฒ˜๋ฆฌ ๋ชจ๋ธ์„ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•œ ์ฑ—๋ด‡๊ณผ ์‹ค์‹œ๊ฐ„ ๋Œ€ํ™”๋ฅผ ์ง„ํ–‰ํ•˜๋„๋ก ์„ค๊ณ„ํ–ˆ๋‹ค. ์‹คํ—˜์€ ๋‘ ๊ฐ€์ง€ ์กฐ๊ฑด์œผ๋กœ ๋‚˜๋‰˜์—ˆ๋Š”๋ฐ, ํ•˜๋‚˜๋Š” โ€˜์ธ๊ฐ„ํ˜• ๋””์ž์ธ ๋ ˆ๋ฒ„โ€™๋ฅผ ์ ์šฉํ•ด ๋Œ€ํ™” ํ๋ฆ„์„ ๋ถ€๋“œ๋Ÿฝ๊ฒŒ ํ•˜๊ณ , ์‚ฌ์šฉ์ž์˜ ๊ฐ์ •์„ ๋ฐ˜์˜ํ•˜๋Š” ์–ธ์–ด์  ํ‘œํ˜„์„ ์‚ฝ์ž…ํ•œ ๋ฒ„์ „์ด๋ฉฐ, ๋‹ค๋ฅธ ํ•˜๋‚˜๋Š” โ€˜๋น„์ธ๊ฐ„ํ˜•โ€™ ์ฆ‰, ๊ธฐ๊ณ„์ ์ด๊ณ  ๊ธฐ๋Šฅ ์ค‘์‹ฌ์ ์ธ ์‘๋‹ต๋งŒ์„ ์ œ๊ณตํ•˜๋Š” ๋ฒ„์ „์ด

Improved Accuracy of Robot Localization Using 3-D LiDAR in a Hippocampus-Inspired Model

Improved Accuracy of Robot Localization Using 3-D LiDAR in a Hippocampus-Inspired Model

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ 3โ€‘D LiDAR๋Š” ๊ณ ๋ฐ€๋„ ๊ฑฐ๋ฆฌ ์ •๋ณด๋ฅผ ์ œ๊ณตํ•˜์ง€๋งŒ, ์žก์Œ, ๋ฐ˜์‚ฌ์œจ ์ฐจ์ด, ๋™์  ๋ฌผ์ฒด์— ์ทจ์•ฝํ•˜๋‹ค. ๊ธฐ์กด SLAM ๊ธฐ๋ฒ•์€ ์ฃผ๋กœ ๊ธฐํ•˜ํ•™์  ๋งค์นญ์ด๋‚˜ ๊ทธ๋ž˜ํ”„ ์ตœ์ ํ™”์— ์˜์กดํ•ด ํ™˜๊ฒฝ ๋ณ€ํ™”์— ๋ฏผ๊ฐํ•˜๋‹ค. ์ธ๊ฐ„์˜ ํ•ด๋งˆ๋Š” ๋ถˆ์™„์ „ํ•˜๊ณ  ์žก์Œ์ด ์„ž์ธ ๊ฐ๊ฐ ์ž…๋ ฅ์—์„œ๋„ ์•ˆ์ •์ ์ธ ๊ณต๊ฐ„ ์ธ์ฝ”๋”ฉ์„ ์ˆ˜ํ–‰ํ•œ๋‹ค๋Š” ์ ์—์„œ ์˜๊ฐ์˜ ์›์ฒœ์ด ๋œ๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด ํ•ด๋งˆโ€‘์˜๊ฐ ๋ชจ๋ธ : ์žฅ์†Œ ์„ธํฌ์™€ ๊ฒฝ๋กœ ์„ธํฌ๋ฅผ ์ˆ˜ํ•™์ ์œผ๋กœ ๋ชจ๋ธ๋งํ•˜์—ฌ, LiDAR ํฌ์ธํŠธ ํด๋ผ์šฐ๋“œ์˜ ํŠน์ง•์„ โ€˜์œ„์น˜ ์ฝ”๋“œโ€™์™€ โ€˜๊ฒฝ๋กœ ์ฝ”๋“œโ€™๋กœ ๋ณ€ํ™˜ํ•œ๋‹ค. ๋‹ค์ค‘ ์Šค์ผ€์ผ ๋งคํ•‘ : 3โ€‘D ํฌ์ธํŠธ๋ฅผ ์—ฌ

Model
No Image

Is Your Prompt Poisoning Code? Defect Induction Rates and Security Mitigation Strategies

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ์‹ค์ œ ๊ฐœ๋ฐœ ํ˜„์žฅ ์—์„œ๋Š” ์•…์˜์ ์ธ ๊ณต๊ฒฉ๋ณด๋‹ค ํ”„๋กฌํ”„ํŠธ ์ž‘์„ฑ ์‹ค์ˆ˜ ๊ฐ€ ๋” ๋นˆ๋ฒˆํ•˜๊ฒŒ ๋ฐœ์ƒํ•œ๋‹ค. ๊ธฐ์กด ๋ณด์•ˆ ์—ฐ๊ตฌ๊ฐ€ โ€œ๋ชจ๋ธ ์ž์ฒด์˜ ์ทจ์•ฝ์ โ€์— ์ง‘์ค‘ํ•œ ๋ฐ˜๋ฉด, ํ”„๋กฌํ”„ํŠธโ€‘์ฝ”๋“œ ๋ณด์•ˆ ์—ฐ๊ณ„ ๋Š” ์•„์ง ์ฒด๊ณ„์ ์ธ ํ‰๊ฐ€๊ฐ€ ๋ถ€์กฑํ–ˆ๋‹ค. 2. ํ”„๋กฌํ”„ํŠธ ํ’ˆ์งˆ ํ‰๊ฐ€ ํ”„๋ ˆ์ž„์›Œํฌ | ์ฐจ์› | ์ •์˜ | ํ‰๊ฐ€ ๊ธฐ์ค€ ์˜ˆ์‹œ | | | | | | ๋ชฉํ‘œ ๋ช…ํ™•์„ฑ (Goal Clarity) | ๋ชจ๋ธ์ด ์ˆ˜ํ–‰ํ•ด์•ผ ํ•  ์ž‘์—…์ด ๋ช…ํ™•ํžˆ ์ œ์‹œ๋˜๋Š”๊ฐ€ | โ€œ์ฃผ์–ด์ง„ ํ•จ์ˆ˜์— ๋Œ€ํ•œ ํ…Œ์ŠคํŠธ ์ผ€์ด์Šค๋ฅผ ์ž‘์„ฑํ•ด ์ฃผ์„ธ์š”.โ€ vs โ€œ๋ญ ์ข€ ํ•ด์ค˜.โ€ | | ์ •๋ณด ์™„์ „์„ฑ (Informatio

No Image

KFCPO: Kronecker-Factored Approximated Constrained Policy Optimization

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ Safe RL ์€ ์‹ค์ œ ์‹œ์Šคํ…œ์— ์ ์šฉ๋  ๋•Œ ๋ณด์•ˆยท์•ˆ์ „ ์ œ์•ฝ์„ ๋ฐ˜๋“œ์‹œ ๋งŒ์กฑํ•ด์•ผ ํ•˜๋Š”๋ฐ, ๊ธฐ์กด ์ •์ฑ… ์ตœ์ ํ™” ๊ธฐ๋ฒ•์€ ๋ณด์ƒ์—๋งŒ ์ง‘์ค‘ํ•ด ์ œ์•ฝ ์œ„๋ฐ˜ ์œ„ํ—˜์ด ํฌ๋‹ค. 2์ฐจ ์ตœ์ ํ™” (์ž์—ฐ ๊ทธ๋ž˜๋””์–ธํŠธ)๋Š” ์ˆ˜๋ ด ์†๋„์™€ ์•ˆ์ •์„ฑ์„ ํฌ๊ฒŒ ๊ฐœ์„ ํ•˜์ง€๋งŒ, ํ”ผ์…” ํ–‰๋ ฌ์„ ์ •ํ™•ํžˆ ๊ณ„์‚ฐํ•˜๊ฑฐ๋‚˜ ๊ทผ์‚ฌํ•˜๋Š” ๋น„์šฉ์ด prohibitive(๋น„์šฉ์ด ๋งŽ์ด ๋“ฆ)ํ•˜๋‹ค. 2. ํ•ต์‹ฌ ๊ธฐ๋ฒ• | ๊ตฌ์„ฑ ์š”์†Œ | ์„ค๋ช… | ์žฅ์  | | | | | | Kโ€‘FAC ๊ธฐ๋ฐ˜ ์ž์—ฐ ๊ทธ๋ž˜๋””์–ธํŠธ | ๊ฐ ๋ ˆ์ด์–ด์˜ ํŒŒ๋ผ๋ฏธํ„ฐ๋ฅผ Kronecker product ํ˜•ํƒœ๋กœ ๊ทผ์‚ฌํ•ด FIM์„ ๋น ๋ฅด๊ฒŒ ์—ญ์ „

No Image

Laboratory observation of collective beam-plasma instabilities in a relativistic pair jet

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ์˜์˜ ์ฒœ์ฒด ๋ฌผ๋ฆฌํ•™์  ๋™๊ธฐ : ๋ธ”๋ž™ํ™€ ์ฃผ๋ณ€, ์ค‘์„ฑ์ž๋ณ„, ๋ธ”๋ ˆ์ด์ € ์ œํŠธ ๋“ฑ์—์„œ ์ „์žโ€‘์–‘์„ฑ์ž ์Œ ํ”Œ๋ผ์ฆˆ๋งˆ๊ฐ€ ์กด์žฌํ•œ๋‹ค๋Š” ์ด๋ก ์  ์˜ˆ์ธก์ด ๋งŽ๋‹ค. ์ด๋Ÿฌํ•œ ํ”Œ๋ผ์ฆˆ๋งˆ๋Š” ์ž…์ž ๊ฐ€์†, ์ž๊ธฐ์žฅ ์ƒ์„ฑ, ๋น„์—ด๋ณต์‚ฌ(nonโ€‘thermal radiation) ๋“ฑ ๋‹ค์–‘ํ•œ ๊ณ ์—๋„ˆ์ง€ ํ˜„์ƒ์˜ ํ•ต์‹ฌ ๋ฉ”์ปค๋‹ˆ์ฆ˜์œผ๋กœ ์ œ์‹œ๋œ๋‹ค. ์‹คํ—˜์  ๋‚œ์ œ : ์ง€๊ตฌ ์‹คํ—˜์‹ค์—์„œ ์ถฉ๋ถ„ํžˆ ๋†’์€ ์Œ ๋ฐ€๋„์™€ ์—๋„ˆ์ง€๋ฅผ ํ™•๋ณดํ•ด ์ง‘๋‹จ ํšจ๊ณผ๋ฅผ ๊ด€์ธกํ•˜๊ธฐ๋Š” ๋งค์šฐ ์–ด๋ ค์› ๋‹ค. ๊ธฐ์กด ๋ ˆ์ด์ €โ€‘๋“œ๋ฆฌ๋ธ ๋ฐฉ์‹์€ ๋†’์€ ์ž…์ž ์ˆ˜์™€ ์—๋„ˆ์ง€ ์ŠคํŽ™ํŠธ๋Ÿผ์„ ๋™์‹œ์— ๋‹ฌ์„ฑํ•˜๊ธฐ ํž˜๋“ค์—ˆ์œผ๋ฉฐ, ํ•ต๋ฐ˜์‘๋กœ๋‚˜ ์ „ํ†ต์ ์ธ ๊ฐ€์†๊ธฐ ๊ธฐ๋ฐ˜

Physics
No Image

Learning General Policies with Policy Gradient Methods

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ์ผ๋ฐ˜ํ™” ๋ฌธ์ œ ๋Š” ํ˜„์žฌ ๋”ฅ RL ์—ฐ๊ตฌ์—์„œ ๊ฐ€์žฅ ํฐ ๊ฑธ๋ฆผ๋Œ ์ค‘ ํ•˜๋‚˜์ด๋ฉฐ, ํŠนํžˆ ํ”Œ๋ž˜๋‹ ๋„๋ฉ”์ธ์—์„œ๋Š” ์ธ์Šคํ„ด์Šค๋งˆ๋‹ค ์ƒํƒœยท์•ก์…˜ ๊ณต๊ฐ„์ด ํฌ๊ฒŒ ๋‹ฌ๋ผ ๊ธฐ์กด ์ •์ฑ…์ด ๋ฐ”๋กœ ์ ์šฉ๋˜๊ธฐ ์–ด๋ ต๋‹ค. ์ „ํ†ต์ ์ธ ์กฐํ•ฉ๋ก ์  ํ”Œ๋ž˜๋‹ ์€ ๋„๋ฉ”์ธ ๋ชจ๋ธ์„ ์ด์šฉํ•ด ์ฆ๋ช… ๊ฐ€๋Šฅํ•œ ์ผ๋ฐ˜ ์ •์ฑ…์„ ๋„์ถœํ•˜์ง€๋งŒ, ์กฐํ•ฉ ํญ๋ฐœ ๊ณผ ํŠน์ง• ํ’€ ์„ค๊ณ„์— ํฌ๊ฒŒ ์˜์กดํ•œ๋‹ค๋Š” ๋‹จ์ ์ด ์žˆ๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด | ์กฐํ•ฉ๋ก ์  ์ ‘๊ทผ | ๋”ฅ RL ์ ‘๊ทผ | ์œตํ•ฉ ์ „๋žต | | | | | | ์ •์ฑ… ์ƒํƒœ ์ „์ด ๋ถ„๋ฅ˜๊ธฐ (์•ก์…˜์ด ์•„๋‹Œ ์ƒํƒœ ๋ณ€ํ™”๋ฅผ ์˜ˆ์ธก) | ์ •์ฑ… ํŒŒ๋ผ๋ฏธํ„ฐํ™”๋œ ํ•จ์ˆ˜ (์‹ ๊ฒฝ๋ง) | ์ •

Learning
No Image

Learning Low Rank Neural Representations of Hyperbolic Wave Dynamics from Data

| ๊ตฌ๋ถ„ | ๋‚ด์šฉ | | | | | ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ | ํ•˜์ดํผ๋ณผ๋ฆญ ํŒŒ๋™ ๋ฐฉ์ •์‹(์˜ˆ: ์Œํ–ฅ, ์ „์ž๊ธฐ, ํƒ„์„ฑ ํŒŒ๋™)์€ ๊ณ ์ฐจ์› ์—ฐ์†์ฒด๋ฅผ ๋‹ค๋ฃจ๋ฉฐ, ์ „ํ†ต์ ์ธ ์ˆ˜์น˜ ํ•ด์„์€ ๋ฉ”๋ชจ๋ฆฌยท์—ฐ์‚ฐ ๋น„์šฉ์ด ํฌ๊ฒŒ ๋“ ๋‹ค. ์ €์ฐจ์› ๋ชจ๋ธ๋ง์€ ์‹ค์‹œ๊ฐ„ ์‹œ๋ฎฌ๋ ˆ์ด์…˜, ์ œ์–ด, ๋ฐ์ดํ„ฐ ์ „์†ก ๋“ฑ์— ํ•„์ˆ˜์ ์ด์ง€๋งŒ, ๊ธฐ์กด POD(Proper Orthogonal Decomposition)ยทAutoโ€‘Encoder ๊ธฐ๋ฐ˜ ๋ฐฉ๋ฒ•์€ ๋น„์„ ํ˜• ํŒŒ๋™ ๊ตฌ์กฐ๋ฅผ ์ถฉ๋ถ„ํžˆ ํฌ์ฐฉํ•˜์ง€ ๋ชปํ•œ๋‹ค. | | ํ•ต์‹ฌ ์•„์ด๋””์–ด | 1๏ธโƒฃ Lowโ€‘Rank Neural Representation (LRNR) : ํŒŒ๋™ ํ•ด

Learning Data
No Image

Leveraging Generic Time Series Foundation Models for EEG Classification

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ์‹œ๊ณ„์—ด ํŒŒ์šด๋ฐ์ด์…˜ ๋ชจ๋ธ : ๋Œ€๊ทœ๋ชจ ์‹œ๊ณ„์—ด ๋ฐ์ดํ„ฐ(์˜ˆ: ์„ผ์„œ, ๊ธˆ์œต, ๊ธฐ์ƒ)์—์„œ ์‚ฌ์ „ํ•™์Šต๋œ ๋ชจ๋ธ์ด ๋‹ค์–‘ํ•œ ๋‹ค์šด์ŠคํŠธ๋ฆผ ์ž‘์—…์— ๋›ฐ์–ด๋‚œ ์ „์ด ํ•™์Šต ์„ฑ๋Šฅ์„ ๋ณด์ด๊ณ  ์žˆ๋‹ค. EEG ๋ถ„์•ผ์˜ ๊ฒฉ์ฐจ : EEG๋Š” ๊ณ ์ฐจ์›ยท์ €์‹ ํ˜ธโ€‘๋Œ€โ€‘๋…ธ์ด์ฆˆ ํŠน์„ฑ์„ ๊ฐ€์ง€๋ฉฐ, ๊ธฐ์กด์—๋Š” ์ „์šฉ CNNยทRNN ๊ตฌ์กฐ(EEGNet ๋“ฑ) ํ˜น์€ ๋„๋ฉ”์ธ ํŠนํ™” ํŒŒ์šด๋ฐ์ด์…˜ ๋ชจ๋ธ(CBraMod)๋งŒ์ด ์ฃผ๋กœ ์‚ฌ์šฉ๋ผ ์™”๋‹ค. ์—ฐ๊ตฌ ์งˆ๋ฌธ : โ€œ๋น„์‹ ๊ฒฝ๊ณ„ ์‹œ๊ณ„์—ด ํ˜น์€ ํ•ฉ์„ฑ ์‹œ๊ณ„์—ด๋งŒ์œผ๋กœ ์‚ฌ์ „ํ•™์Šต๋œ ๋ฒ”์šฉ ๋ชจ๋ธ์ด EEG ์ž‘์—…์— ์ถฉ๋ถ„ํžˆ ์ ์šฉ๋  ์ˆ˜ ์žˆ๋Š”๊ฐ€?โ€ 2. ๋ฐฉ๋ฒ•๋ก  | ์š”์†Œ | ์„ค๋ช…

Model
No Image

Lifecycle-Aware code generation: Leveraging Software Engineering Phases in LLMs

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ๋‹จ์ผโ€‘step ์ฝ”๋“œ ์ƒ์„ฑ์˜ ํ•œ๊ณ„ : ๊ธฐ์กด LLM ๊ธฐ๋ฐ˜ ์ฝ”๋“œ ์ƒ์„ฑ์€ โ€œ๋ฌธ์ œ โ†’ ์ฝ”๋“œโ€์˜ ์ง๊ด€์  ๋งคํ•‘์— ์˜์กดํ•ด, ๋ณต์žกํ•œ ์š”๊ตฌ์‚ฌํ•ญ์ด๋‚˜ ์„ค๊ณ„ ๋‹จ๊ณ„์—์„œ ๋ฐœ์ƒํ•˜๋Š” ์˜ค๋ฅ˜๋ฅผ ํฌ์ฐฉํ•˜๊ธฐ ์–ด๋ ต๋‹ค. ์†Œํ”„ํŠธ์›จ์–ด ๊ณตํ•™ ํ”„๋กœ์„ธ์Šค์™€์˜ ๊ดด๋ฆฌ : ์š”๊ตฌ์‚ฌํ•ญ ๋ช…์„ธ, ์„ค๊ณ„, ๊ตฌํ˜„, ํ…Œ์ŠคํŠธ ๋“ฑ ๋‹จ๊ณ„์  ์ ‘๊ทผ์ด ๊ฒ€์ฆ๋œ ์‚ฐ์—… ํ˜„์žฅ๊ณผ ๋‹ฌ๋ฆฌ, ํ˜„์žฌ LLM์€ ์ด๋Ÿฌํ•œ ๊ตฌ์กฐ์  ๊ฒ€์ฆ ์ ˆ์ฐจ๋ฅผ ๋ฌด์‹œํ•œ๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด ์ƒ๋ช…์ฃผ๊ธฐ ์ธ์‹ ํ”„๋ ˆ์ž„์›Œํฌ : 1. ์š”๊ตฌ์‚ฌํ•ญ ๋ถ„์„ โ€“ ์ž์—ฐ์–ด๋กœ ๋œ ๋ฌธ์ œ ์„ค๋ช…์„ ๊ตฌ์กฐํ™”๋œ ์š”๊ตฌ์‚ฌํ•ญ ๋ฆฌ์ŠคํŠธ๋กœ ๋ณ€ํ™˜. 2. ์ƒํƒœ ๋จธ์‹  ๋ชจ๋ธ๋ง โ€“ ์‹œ์Šคํ…œ

No Image

Lightning Grasp: High Performance Procedural Grasp Synthesis with Contact Fields

1. ์ฃผ์š” ๊ธฐ์—ฌ | ๋ฒˆํ˜ธ | ๊ธฐ์—ฌ ๋‚ด์šฉ | ์˜์˜ | | | | | | 1 | Contact Field ๋ผ๋Š” ์ƒˆ๋กœ์šด ๋ฐ์ดํ„ฐ ๊ตฌ์กฐ ๋„์ž… | ๋ณต์žกํ•œ ์ถฉ๋Œยท์ ‘์ด‰ ๊ณ„์‚ฐ์„ ์‚ฌ์ „ ์ฒ˜๋ฆฌํ•ด ๊ฒ€์ƒ‰ ๋‹จ๊ณ„์—์„œ O(1) ์ˆ˜์ค€์˜ ์กฐํšŒ ๊ฐ€๋Šฅ. | | 2 | ์ ˆ์ฐจ์ (Procedural) ํƒ์ƒ‰ ์•Œ๊ณ ๋ฆฌ์ฆ˜ ์„ค๊ณ„ | ์ „ํ†ต์ ์ธ ์ตœ์ ํ™” ๊ธฐ๋ฐ˜ ์ ‘๊ทผ๋ฒ• ๋Œ€๋น„ 10โ€‘100ร— ๋น ๋ฅธ ๊ทธ๋ฆฝ ํ›„๋ณด ์ƒ์„ฑ. | | 3 | ๋น„์ง€๋„ ๊ทธ๋ฆฝ ์ƒ์„ฑ ์ง€์› | ์‚ฌ์ „ ๋ผ๋ฒจ๋งยท์—๋„ˆ์ง€ ํŒŒ๋ผ๋ฏธํ„ฐ ํŠœ๋‹ ์—†์ด๋„ ๋‹ค์–‘ํ•œ ํ˜•ํƒœ(๋ถˆ๊ทœ์น™, ๋„๊ตฌํ˜•) ๋ฌผ์ฒด์— ์ ์šฉ ๊ฐ€๋Šฅ. | | 4 | ์˜คํ”ˆ์†Œ์Šค ๊ตฌํ˜„ ์ œ๊ณต | ์—ฐ๊ตฌ ์žฌํ˜„์„ฑ

LLM-based Behaviour Driven Development for Hardware Design

LLM-based Behaviour Driven Development for Hardware Design

ํ•˜๋“œ์›จ์–ด ์„ค๊ณ„ ๊ณผ์ •์—์„œ ํ…Œ์ŠคํŠธ์™€ ๊ฒ€์ฆ์€ ์„ค๊ณ„ ์˜ค๋ฅ˜๋ฅผ ์ดˆ๊ธฐ์— ๋ฐœ๊ฒฌํ•˜๊ณ  ์‹œ์Šคํ…œ ์‹ ๋ขฐ์„ฑ์„ ํ™•๋ณดํ•˜๊ธฐ ์œ„ํ•œ ํ•ต์‹ฌ ๋‹จ๊ณ„์ด๋‹ค. ์„ค๊ณ„ ๊ทœ๋ชจ๊ฐ€ ์ˆ˜์‹ญ๋งŒ ๊ฐœ ์ด์ƒ์˜ ๋…ผ๋ฆฌ ์…€์ด๋‚˜ ๋ณต์žกํ•œ ์ธํ„ฐํŽ˜์ด์Šค๋ฅผ ํฌํ•จํ•˜๊ฒŒ ๋˜๋ฉด, ์ „ํ†ต์ ์ธ ๊ฒ€์ฆ ํ๋ฆ„์€ ์ˆ˜์ž‘์—…์œผ๋กœ ์ž‘์„ฑ๋œ ํ…Œ์ŠคํŠธ๋ฒค์น˜, ์‹œ๋ฎฌ๋ ˆ์ด์…˜ ์Šคํฌ๋ฆฝํŠธ, ๊ทธ๋ฆฌ๊ณ  ๊ฒ€์ฆ ๋ชฉํ‘œ๋ฅผ ์ •์˜ํ•˜๋Š” ์‹œ๋‚˜๋ฆฌ์˜ค์— ํฌ๊ฒŒ ์˜์กดํ•œ๋‹ค. ์ด๋Ÿฌํ•œ ์ ‘๊ทผ ๋ฐฉ์‹์€ ์ธ๋ ฅ ๋น„์šฉ์ด ๊ธ‰์ฆํ•˜๊ณ , ์ธ๊ฐ„์ด ์ž‘์„ฑํ•œ ์‹œ๋‚˜๋ฆฌ์˜ค๊ฐ€ ์‚ฌ์–‘์˜ ๋ชจ๋“  ๋ณ€ํ˜•์„ ํฌ๊ด„ํ•˜์ง€ ๋ชปํ•˜๋Š” ์œ„ํ—˜์„ ๋‚ดํฌํ•œ๋‹ค. ํ–‰๋™ ์ฃผ๋„ ๊ฐœ๋ฐœ(BDD)์€ โ€œGivenโ€‘Whenโ€‘Thenโ€ ํ˜•์‹์˜ ์ž์—ฐ์–ด ์‹œ๋‚˜๋ฆฌ์˜ค๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ

Computer Science Software Engineering
No Image

LouvreSAE: Sparse Autoencoders for Interpretable and Controllable Style Transfer

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ฌธ์ œ์  ์Šคํƒ€์ผ ์ „์ด์˜ ๋ณตํ•ฉ์„ฑ : ๊ธฐ์กด GANยทDiffusion ๊ธฐ๋ฐ˜ ์Šคํƒ€์ผ ์ „์ด๋Š” ์ฃผ๋กœ ์ „์ฒด ๋ชจ๋ธ์„ ์žฌํ•™์Šตํ•˜๊ฑฐ๋‚˜ LoRAยทAdapter์™€ ๊ฐ™์€ ๊ฒฝ๋Ÿ‰ ํŒŒ๋ผ๋ฏธํ„ฐ๋ฅผ ์ถ”๊ฐ€ํ•œ๋‹ค. ์ด๋Š” ์—ฐ์‚ฐ ๋น„์šฉ ๊ณผ ๋ฉ”๋ชจ๋ฆฌ ์‚ฌ์šฉ๋Ÿ‰ ์„ ํฌ๊ฒŒ ๋Š˜๋ฆฌ๋ฉฐ, ์Šคํƒ€์ผ๊ณผ ๋‚ด์šฉ์ด ์™„์ „ํžˆ ๋ถ„๋ฆฌ๋˜์ง€ ์•Š์•„ ์›ํ•˜๋Š” ์Šคํƒ€์ผ๋งŒ์„ ์ •๋ฐ€ํ•˜๊ฒŒ ์ œ์–ดํ•˜๊ธฐ ์–ด๋ ต๋‹ค. ํ•ด์„ ๊ฐ€๋Šฅ์„ฑ ๋ถ€์žฌ : ๋Œ€๋ถ€๋ถ„์˜ ๋ฐฉ๋ฒ•์€ โ€œ๋ธ”๋ž™๋ฐ•์Šคโ€ ํ˜•ํƒœ๋กœ ์ž‘๋™ํ•ด, ์–ด๋–ค ํŒŒ๋ผ๋ฏธํ„ฐ๊ฐ€ ์–ด๋–ค ์Šคํƒ€์ผ ์š”์†Œ์— ๋Œ€์‘ํ•˜๋Š”์ง€ ์•Œ๊ธฐ ํž˜๋“ค๋‹ค. ์ด๋Š” ์•„ํ‹ฐ์ŠคํŠธ์™€ ๋””์ž์ด๋„ˆ๊ฐ€ ์ง๊ด€์ ์œผ๋กœ ์Šคํƒ€์ผ์„ ์กฐ์ •ํ•˜๊ธฐ ์–ด๋ ต๊ฒŒ ๋งŒ๋“ ๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด

No Image

MAGNET: A Multi-Graph Attentional Network for Code Clone Detection

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋ฌธ์ œ ์ •์˜ ์ฝ”๋“œ ๋ณต์ œ ํƒ์ง€ ๋Š” ๋™์ผํ•˜๊ฑฐ๋‚˜ ์œ ์‚ฌํ•œ ์ฝ”๋“œ ์กฐ๊ฐ์„ ์ž๋™์œผ๋กœ ์‹๋ณ„ํ•˜๋Š” ์ž‘์—…์œผ๋กœ, ์œ ์ง€๋ณด์ˆ˜ ๋น„์šฉ ์ ˆ๊ฐ ๋ฐ ๋ณด์•ˆ ์ทจ์•ฝ์  ํƒ์ง€์— ํ•„์ˆ˜์ ์ด๋‹ค. ๊ธฐ์กด ์—ฐ๊ตฌ๋Š” ๋‹จ์ผ ๊ทธ๋ž˜ํ”„ (AST, CFG, DFG ๋“ฑ) ๊ธฐ๋ฐ˜ ๋ชจ๋ธ์— ์˜์กดํ•ด ๊ตฌ๋ฌธ ํ˜น์€ ์˜๋ฏธ ์ •๋ณด๋งŒ์„ ๋ถ€๋ถ„์ ์œผ๋กœ ํ™œ์šฉํ•œ๋‹ค. ์ด๋Š” ๋ณต์ œ ์ฝ”๋“œ๊ฐ€ ๊ตฌ์กฐ์ ์œผ๋กœ ๋ณ€ํ˜•๋˜๊ฑฐ๋‚˜ ๋ณ€์ˆ˜๋ช…ยท๋ฆฌํ„ฐ๋Ÿด์ด ๋ฐ”๋€Œ๋Š” ๊ฒฝ์šฐ ํƒ์ง€ ์„ฑ๋Šฅ์„ ์ €ํ•˜์‹œํ‚จ๋‹ค. ํ•˜์ด๋ธŒ๋ฆฌ๋“œ ์ ‘๊ทผ๋ฒ•์€ ์—ฌ๋Ÿฌ ๊ทธ๋ž˜ํ”„๋ฅผ ๊ฒฐํ•ฉํ•˜๋ ค ํ–ˆ์ง€๋งŒ, ์ˆ˜์ž‘์—… ์œตํ•ฉ ๊ทœ์น™ (์˜ˆ: ๋‹จ์ˆœ ์—ฐ๊ฒฐ, ๊ฐ€์ค‘ ํ‰๊ท )์œผ๋กœ๋Š” ์„œ๋กœ ๋‹ค๋ฅธ ๊ทธ๋ž˜ํ”„ ๊ฐ„ ๋ณต์žกํ•œ ์ƒํ˜ธ์ž‘์šฉ์„ ์ถฉ๋ถ„ํžˆ ๋ชจ

Network Detection
No Image

Memristive tabular variational autoencoder for compression of analog data in high energy physics

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ๋ฐ์ดํ„ฐ ํญ์ฆ : FCCโ€‘ee, ฮผโ€‘Collider ๋“ฑ ์ฐจ์„ธ๋Œ€ ๊ณ ์—๋„ˆ์ง€ ์‹คํ—˜์—์„œ๋Š” ์ดˆ๋‹น ์ˆ˜์‹ญ kHz, ์ˆ˜์‹ญ ์–ต ์ฑ„๋„ ๊ทœ๋ชจ์˜ ์ŠคํŠธ๋ฆฌ๋ฐ ๋ฐ์ดํ„ฐ๊ฐ€ ๋ฐœ์ƒํ•œ๋‹ค. ๊ธฐ์กด์˜ ์˜คํ”„โ€‘๋””ํ…ํ„ฐ ์ €์žฅ ๋ฐฉ์‹์€ ๋ฌผ๋ฆฌ์ ยท๊ฒฝ์ œ์  ํ•œ๊ณ„์— ์ง๋ฉดํ•œ๋‹ค. AIโ€‘๊ธฐ๋ฐ˜ ์••์ถ•์˜ ํ•„์š”์„ฑ : ๋ณ€๋ถ„ ์˜คํ† ์ธ์ฝ”๋”(VAE)๋Š” ๋น„์„ ํ˜• ์ฐจ์› ์ถ•์†Œ์™€ ๋ณต์› ์ •ํ™•๋„ ์‚ฌ์ด์˜ ํŠธ๋ ˆ์ด๋“œ์˜คํ”„๋ฅผ ์ตœ์ ํ™”ํ•  ์ˆ˜ ์žˆ์–ด, ๋ฌผ๋ฆฌ์  ํŠน์„ฑ์„ ๋ณด์กดํ•˜๋ฉด์„œ ์••์ถ•๋ฅ ์„ ๋†’์ด๋Š” ๋ฐ ์ ํ•ฉํ•˜๋‹ค. ๋ฉ”๋ชจ๋ฆฌโ€‘์ปดํ“จํŒ…(MC) ํ•œ๊ณ„ ๊ทน๋ณต : ์ „ํ†ต์ ์ธ vonโ€‘Neumann ๊ตฌ์กฐ๋Š” ๋ฉ”๋ชจ๋ฆฌโ€‘์›”(memory wall)๋กœ ์ธํ•ด

Physics Data
No Image

Microscopic Rydberg electron orbit manipulation with optical tweezers

1. ์—ฐ๊ตฌ ๋™๊ธฐ์™€ ๋ฐฐ๊ฒฝ ์ „ํ†ต์ ์ธ ์ „์ž ๊ถค๋„ ์ œ์–ด๋Š” STMยทAFM ๋“ฑ ์›์ž ์ˆ˜์ค€์˜ ํƒ์นจ์ด ํ•„์š”ํ•˜์ง€๋งŒ, Rydberg ์›์ž๋Š” ์ „์ž ํŒŒ๋™ํ•จ์ˆ˜๊ฐ€ ์ˆ˜ ๋งˆ์ดํฌ๋ก ๊นŒ์ง€ ํ™•์žฅ๋ผ ๊ด‘ํ•™ ํ˜„๋ฏธ๊ฒฝ(ํŠนํžˆ ๊ณ ์ง‘์† ๊ด‘ํ•™ ํŠธ์œ„์ €)์œผ๋กœ๋„ ์ ‘๊ทผ ๊ฐ€๋Šฅํ•˜๊ฒŒ ๋งŒ๋“ ๋‹ค. ์ด๋Š” โ€œ๊ด‘ํ•™์ ์œผ๋กœ ํฐ ์ „์ž ๊ตฌ๋ฆ„โ€์„ ์ง์ ‘ ์กฐ์ž‘ํ•  ์ˆ˜ ์žˆ๋Š” ์ „๋ก€ ์—†๋Š” ํ”Œ๋žซํผ์„ ์ œ๊ณตํ•œ๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด โ€“ ฮท ํŒŒ๋ผ๋ฏธํ„ฐ ฮท wโ‚€ / s ฮฝ (๋น” ๋ฐ˜๊ฒฝ / Rydberg ๊ถค๋„ ๋ฐ˜๊ฒฝ) ๋ฅผ ์ œ์–ด ๋ณ€์ˆ˜๋กœ ์‚ผ์•„ ๋‘ ๊ฐ€์ง€ ๊ทนํ•œ์„ ํƒ๊ตฌํ•œ๋‹ค. ฮท โ‰ช 1 (์ดˆ์ง‘์†): ํŽ€๋”๋จธํ‹ฐ๋ธŒ ํฌํ…์…œ์ด ์ „์ž ํŒŒ๋™ํ•จ์ˆ˜์˜ ์ผ๋ถ€๋ถ„์—๋งŒ ๊ฐ•ํ•˜๊ฒŒ

Physics
Mixture of Attention Schemes (MoAS): Learning to Route Between MHA, GQA, and MQA

Mixture of Attention Schemes (MoAS): Learning to Route Between MHA, GQA, and MQA

Transformer ๊ธฐ๋ฐ˜ ์–ธ์–ด ๋ชจ๋ธ์—์„œ ์ฃผ์˜ ๋ฉ”์ปค๋‹ˆ์ฆ˜์€ ์ž…๋ ฅ ํ† ํฐ ๊ฐ„์˜ ์ƒํ˜ธ์ž‘์šฉ์„ ์บก์ฒ˜ํ•˜๋Š” ํ•ต์‹ฌ ์š”์†Œ์ด๋ฉฐ, ๊ทธ ์„ค๊ณ„๋Š” ๋ชจ๋ธ์˜ ํ‘œํ˜„๋ ฅ๊ณผ ์ถ”๋ก  ์‹œ ์ž์› ์†Œ๋น„์— ์ง์ ‘์ ์ธ ์˜ํ–ฅ์„ ๋ฏธ์นœ๋‹ค. ์ „ํ†ต์ ์ธ Multiโ€‘Head Attention(MHA)์€ ์—ฌ๋Ÿฌ ๊ฐœ์˜ ๋…๋ฆฝ์ ์ธ ํ—ค๋“œ๋ฅผ ํ†ตํ•ด ๋‹ค์–‘ํ•œ ํ‘œํ˜„ ๊ณต๊ฐ„์„ ๋™์‹œ์— ํƒ์ƒ‰ํ•จ์œผ๋กœ์จ ๋›ฐ์–ด๋‚œ ํ•™์Šต ์„ฑ๋Šฅ์„ ์ œ๊ณตํ•œ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ถ”๋ก  ๋‹จ๊ณ„์—์„œ ๊ฐ ํ—ค๋“œ๋งˆ๋‹ค ๋ณ„๋„์˜ Keyโ€‘Value(KV) ์บ์‹œ๋ฅผ ์œ ์ง€ํ•ด์•ผ ํ•˜๋ฏ€๋กœ ๋ฉ”๋ชจ๋ฆฌ ์š”๊ตฌ๋Ÿ‰์ด ๊ธ‰๊ฒฉํžˆ ์ฆ๊ฐ€ํ•œ๋‹ค. ํŠนํžˆ ๊ธด ์‹œํ€€์Šค๋ฅผ ์ฒ˜๋ฆฌํ•˜๊ฑฐ๋‚˜ ๋ชจ๋ฐ”์ผยท์ž„๋ฒ ๋””๋“œ ํ™˜๊ฒฝ์—์„œ ์‹ค์‹œ๊ฐ„ ์‘๋‹ต์„ฑ์„ ์š”

Computer Science Learning Artificial Intelligence
No Image

Multi-Agent Reinforcement Learning for Market Making: Competition without Collusion

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ์˜์˜ ์•Œ๊ณ ๋ฆฌ์ฆ˜ ๋‹ดํ•ฉ ์€ ๊ทœ์ œ ๋‹น๊ตญ๊ณผ ํ•™๊ณ„ ๋ชจ๋‘๊ฐ€ ์šฐ๋ คํ•˜๋Š” ํ˜„์ƒ์ด๋ฉฐ, ํŠนํžˆ ๋งˆ์ผ“ ๋ฉ”์ดํ‚น ๊ณผ ๊ฐ™์ด ์œ ๋™์„ฑ์„ ์ œ๊ณตํ•˜๋Š” ์—ญํ• ์—์„œ ์ž๋™ํ™”๋œ ์ „๋žต์ด ์„œ๋กœ ํ•™์Šตํ•˜๋ฉด์„œ ๋น„์˜๋„์  ํ˜‘์กฐ(๋˜๋Š” ๊ฒฝ์Ÿ)๋ฅผ ํ˜•์„ฑํ•  ๊ฐ€๋Šฅ์„ฑ์ด ํฌ๋‹ค. ๊ธฐ์กด ์—ฐ๊ตฌ๋Š” ์ฃผ๋กœ ๋‘ ์—์ด์ „ํŠธ ํ˜น์€ ๋‹จ์ผ ์‹œ์žฅ ๋ฉ”์ด์ปค ์— ์ดˆ์ ์„ ๋งž์ท„์ง€๋งŒ, ๋ณธ ๋…ผ๋ฌธ์€ ๋‹ค์ธตยท๋‹ค์ค‘ ์—์ด์ „ํŠธ ๊ตฌ์กฐ๋ฅผ ๋„์ž…ํ•ด ๋ณด๋‹ค ํ˜„์‹ค์ ์ธ ์‹œ์žฅ ํ™˜๊ฒฝ์„ ๋ชจ์‚ฌํ•œ๋‹ค๋Š” ์ ์—์„œ ์ฐจ๋ณ„ํ™”๋œ๋‹ค. 2. ํ”„๋ ˆ์ž„์›Œํฌ ์„ค๊ณ„ | ๊ตฌ์„ฑ ์š”์†Œ | ์—ญํ•  | ํ•™์Šต ๋ชฉํ‘œ | | | | | | Agent A (์ƒ์œ„ ๋ ˆ์ด์–ด) | ๋ถˆํ™•์‹คํ•œ ํ™˜๊ฒฝ(์ ๋Œ€์ 

Learning
No Image

Multi-Modal Feature Fusion for Spatial Morphology Analysis of Traditional Villages via Hierarchical Graph Neural Networks

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ํ•„์š”์„ฑ ์ „ํ†ต ๋งˆ์„์˜ ์†Œ๋ฉธ ์œ„๊ธฐ : ๊ธ‰๊ฒฉํ•œ ๋„์‹œํ™”๋Š” ๋งˆ์„ ๊ณ ์œ ์˜ ๊ณต๊ฐ„ ๊ตฌ์กฐ์™€ ๋ฌธํ™”์  ํ’๊ฒฝ์„ ๊ธ‰์†ํžˆ ๋ณ€ํ˜•ยท์†Œ๋ฉธ์‹œํ‚จ๋‹ค. ๊ธฐ์กด ์—ฐ๊ตฌ์˜ ํ•œ๊ณ„ : ๋Œ€๋ถ€๋ถ„์ด ์ •์„ฑ์ , ํ˜น์€ ๋‹จ์ผ ๋ฐ์ดํ„ฐ(์˜ˆ: ์œ„์„ฑ ์ด๋ฏธ์ง€, ์„ค๋ฌธ) ๊ธฐ๋ฐ˜์œผ๋กœ ์ง„ํ–‰๋ผ, ๋ณตํ•ฉ์ ์ธ ๊ณต๊ฐ„โ€‘์‚ฌํšŒ์  ์š”์ธ์„ ๋™์‹œ์— ๊ณ ๋ คํ•˜๊ธฐ ์–ด๋ ต๋‹ค. ๋””์ง€ํ„ธ ์ธํ”„๋ผ ๋ถ€์กฑ : ์ „ํ†ต ๋งˆ์„์€ GIS, ๋ผ์ด๋‹ค ๋“ฑ ๊ณ ํ•ด์ƒ๋„ ๋ฐ์ดํ„ฐ๊ฐ€ ๋ถ€์กฑํ•ด ์ •๋Ÿ‰์  ๋ถ„์„์ด ์ œํ•œ์ ์ด์—ˆ๋‹ค. 2. ์ œ์•ˆ ๋ชจ๋ธ(HGNN)์˜ ํ•ต์‹ฌ ์•„์ด๋””์–ด | ์š”์†Œ | ์„ค๋ช… | ์—ญํ•  | | | | | | ์ž…๋ ฅ ๋…ธ๋“œ(Input Nodes) | ํ† ์ง€ ์ด์šฉ

Network Analysis
No Image

Multiscale Astrocyte Network Calcium Dynamics for Biologically Plausible Intelligence in Anomaly Detection

1. ์—ฐ๊ตฌ ๋™๊ธฐ์™€ ๋ฐฐ๊ฒฝ ๊ฐœ๋… ๋“œ๋ฆฌํ”„ํŠธ ๋ฌธ์ œ : ์ „ํ†ต์ ์ธ ์ •์  ๋ชจ๋ธ์€ ํ•™์Šต ์‹œ์  ์ดํ›„ ๋ฐ์ดํ„ฐ ๋ถ„ํฌ๊ฐ€ ๋ณ€ํ•˜๋ฉด ์„ฑ๋Šฅ์ด ๊ธ‰๊ฒฉํžˆ ์ €ํ•˜๋œ๋‹ค. ์ƒ๋ฌผํ•™์  ์˜๊ฐ : ๋‡Œ์˜ ๋ณ„์•„๊ต์„ธํฌ๋Š” Caยฒโบ ํŒŒ๋™์„ ํ†ตํ•ด ๋น ๋ฅด๊ณ  ๊ตญ์†Œ์ ์ธ ์‹ ๊ฒฝ ๊ฐ€์†Œ์„ฑ์„ ๊ตฌํ˜„ํ•œ๋‹ค. ์ด๋Ÿฌํ•œ ๋ฉ”์ปค๋‹ˆ์ฆ˜์„ ์ธ๊ณต ์‹œ์Šคํ…œ์— ์˜ฎ๊ธฐ๋ฉด ์‹ค์‹œ๊ฐ„ ์ ์‘์ด ๊ฐ€๋Šฅํ•˜๋‹ค๋Š” ๊ฐ€์„ค์ด ํ•ต์‹ฌ์ด๋‹ค. 2. ๋ชจ๋ธ ์„ค๊ณ„ ๋‹ค์ค‘ ์Šค์ผ€์ผ ์‹œ๋ฎฌ๋ ˆ์ดํ„ฐ : ์„ธํฌ ์ˆ˜์ค€(IPโ‚ƒโ€‘Caยฒโบ ๋ฐฉ์ถœยทSERCA ์žฌํก์ˆ˜)๊ณผ ๋„คํŠธ์›Œํฌ ์ˆ˜์ค€(๊ฐญ์ ‘ํ•ฉ์„ ํ†ตํ•œ ํ™•์‚ฐ)์„ ๋™์‹œ์— ๋ชจ๋ธ๋งํ•จ์œผ๋กœ์จ ๊ณต๊ฐ„ยท์‹œ๊ฐ„์  ์ƒํ˜ธ์ž‘์šฉ์„ ํฌ์ฐฉํ•œ๋‹ค. DNN๊ณผ์˜ ๊ฒฐํ•ฉ ๋ฐฉ์‹ : ์‹œ๋ฎฌ๋ ˆ์ดํ„ฐ๊ฐ€

Network Detection
No Image

Neural Green's Functions

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ๋™๊ธฐ ๊ทธ๋ฆฐ ํ•จ์ˆ˜์™€ ์„ ํ˜• PDE : ๊ทธ๋ฆฐ ํ•จ์ˆ˜๋Š” ์„ ํ˜• ์—ฐ์‚ฐ์ž์˜ ์—ญ์„ ํ‘œํ˜„ํ•˜๋Š” ํ•ต์‹ฌ ๋„๊ตฌ๋กœ, ๋„๋ฉ”์ธ ๊ธฐํ•˜์—๋งŒ ์˜์กดํ•œ๋‹ค๋Š” ์ ์—์„œ ๋‹ค์–‘ํ•œ ๋ฌผ๋ฆฌ ํ˜„์ƒ์— ๋ณดํŽธ์ ์œผ๋กœ ์ ์šฉ ๊ฐ€๋Šฅํ•˜๋‹ค. ๊ธฐ์กด ์‹ ๊ฒฝ ์—ฐ์‚ฐ์ž์˜ ํ•œ๊ณ„ : DeepONet, Fourier Neural Operator ๋“ฑ์€ ํ›ˆ๋ จ ๋ฐ์ดํ„ฐ์— ํฌํ•จ๋œ ์†Œ์Šคยท๊ฒฝ๊ณ„ ํ•จ์ˆ˜์— ๊ฐ•ํ•˜๊ฒŒ ์˜์กดํ•ด, ์ƒˆ๋กœ์šด ํ•จ์ˆ˜๊ฐ€ ๋“ค์–ด์˜ค๋ฉด ์„ฑ๋Šฅ์ด ๊ธ‰๊ฒฉํžˆ ์ €ํ•˜๋œ๋‹ค. ํŠนํžˆ ๋ณต์žกํ•˜๊ณ  ๋ถˆ๊ทœ์น™ํ•œ ํ˜•์ƒ์—์„œ๋Š” ๋ฉ”์‹ฑ ๋น„์šฉ์ด ํฌ๊ฒŒ ๋Š˜์–ด, ์‹ค์‹œ๊ฐ„ ์‹œ๋ฎฌ๋ ˆ์ด์…˜์— ๋ถ€์ ํ•ฉํ•˜๋‹ค. 2. ํ•ต์‹ฌ ์•„์ด๋””์–ด Neural Green's Functi

No Image

Nonlinear Frequency Shifts due to Phase Coherent Interactions in Incompressible Hall MHD Turbulence

1. ์—ฐ๊ตฌ ๋ฐฐ๊ฒฝ ๋ฐ ์˜์˜ Hall MHD์™€ ํŒŒ๋™ ์ƒ‰์‚ฐ : Hall ํ•ญ์„ ํฌํ•จํ•˜๋ฉด Alfvรฉn ํŒŒ๋™์ด ๋ถ„์‚ฐ์„ฑ์„ ๋ ์–ด whistlerยทcyclotron ๋‘ ๊ฐ€์ง€ ๋ถ„๊ธฐ(ฮฑโ‚–โบ, ฮฑโ‚–โป)๊ฐ€ ์ƒ๊ธด๋‹ค. ์ด๋Š” ์ „ํ†ต์ ์ธ MHD ๋‚œ๋ฅ˜์™€๋Š” ๋‹ค๋ฅธ ๋น„์„ ํ˜• ์ƒํ˜ธ์ž‘์šฉ์„ ๊ธฐ๋Œ€ํ•˜๊ฒŒ ํ•œ๋‹ค. ์œ„์ƒโ€‘์ผ๊ด€์„ฑ(Phaseโ€‘Coherent) ์ ‘๊ทผ : ๊ธฐ์กด์˜ ๋‹ค์ค‘ ์Šค์ผ€์ผ(multipleโ€‘scale) ํ˜น์€ ๋‘ ์‹œ๊ฐ„๋ฒ•(twoโ€‘timing)๊ณผ ๋‹ฌ๋ฆฌ, ์ €์ž๋Š” ํŠน์ • ํŒŒ๋™์— ๋Œ€ํ•ด ์œ„์ƒ์ด ์ผ์น˜ํ•˜๋Š” ๋น„์„ ํ˜• ํ•ญ๋งŒ์„ ์ถ”์ถœํ•œ๋‹ค. ์ด๋Š” โ€œ๋น„์„ ํ˜• ์—ฐ์‚ฐ์ž โ†’ ์„ ํ˜• ์—ฐ์‚ฐ์žโ€ ๋ณ€ํ™˜์„ ๊ฐ€๋Šฅํ•˜๊ฒŒ ํ•˜์—ฌ, ์ง

Physics

< Category Statistics (Total: 5051) >

Electrical Engineering and Systems Science
102
General
4147
General Relativity
4
HEP-EX
3
HEP-PH
3
HEP-TH
6
MATH-PH
12
Quantitative Finance
2
Quantum Physics
21

Start searching

Enter keywords to search articles

โ†‘โ†“
โ†ต
ESC
โŒ˜K Shortcut