AI Digest by Angtiger

2026-02-17 21:22 KST Β· by Angtiger Β· 맀일 10:00 KST μ—…λ°μ΄νŠΈ
5 sources
10 new posts
πŸ” 검색 Β· πŸ“’ TelegramAI DigestDEX Digest Β· 🌐 DigestAI DigestDEX Digest Β· πŸ“§ λ¬΄λ£Œκ΅¬λ…
πŸ“§ λ¬΄λ£Œκ΅¬λ…ν•˜κΈ°
← 전체 보기 πŸ“‚ Research & Papers
총 10건
26.02.16 HF Daily Papers EditCtrl: Disentangled Local and Global Control for Real-Time Generative Video Editing
β–² 0
EditCtrl: μ‹€μ‹œκ°„ 생성적 λΉ„λ””μ˜€ νŽΈμ§‘μ„ μœ„ν•œ 효율적 λΉ„λ””μ˜€ μΈνŽ˜μΈνŒ… μ œμ–΄ ν”„λ ˆμž„μ›Œν¬. ν•„μš”ν•œ κ³³μ—λ§Œ 계산을 μ§‘μ€‘ν•˜μ—¬ 둜컬과 κΈ€λ‘œλ²Œ νŽΈμ§‘μ„ 뢄리 μ œμ–΄ν•œλ‹€.
26.02.16 HF Daily Papers AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
β–² 0
AnchorWeave: 카메라 μ œμ–΄ κ°€λŠ₯ν•œ λΉ„λ””μ˜€ μƒμ„±μ—μ„œ μž₯κΈ°κ°„ 곡간 일관성을 μœ μ§€ν•˜κΈ° μœ„ν•΄ 둜컬 곡간 λ©”λͺ¨λ¦¬λ₯Ό ν™œμš©ν•˜λŠ” 세계 일관적 λΉ„λ””μ˜€ 생성 방법.
26.02.15 HF Daily Papers LM-Lexicon: Improving Definition Modeling via Harmonizing Semantic Experts
β–² 0
LM-Lexicon: 데이터 ν΄λŸ¬μŠ€ν„°λ§, μ‹œλ§¨ν‹± μ „λ¬Έκ°€ ν•™μŠ΅, 슀파슀 MoE μ•„ν‚€ν…μ²˜λ₯Ό κ²°ν•©ν•œ μ •μ˜ λͺ¨λΈλ§ 접근법. κΈ°μ‘΄ SOTA λŒ€λΉ„ BLEU 점수 7% ν–₯상 달성.
26.02.15 HF Daily Papers STATe-of-Thoughts: Structured Action Templates for Tree-of-Thoughts
β–² 0
STATe-of-Thoughts: κ³ μˆ˜μ€€ μΆ”λ‘  νŒ¨ν„΄μ„ νƒμƒ‰ν•˜λŠ” 해석 κ°€λŠ₯ν•œ Inference-Time-Compute 방법. κΈ°μ‘΄ Tree-of-Thoughts의 λ‹€μ–‘μ„± λΆ€μ‘± 문제λ₯Ό κ΅¬μ‘°ν™”λœ μ•‘μ…˜ ν…œν”Œλ¦ΏμœΌλ‘œ ν•΄κ²°ν•œλ‹€.
26.02.13 HF Daily Papers SPILLage: Agentic Oversharing on the Web
β–² 0
SPILLage: LLM 기반 μ›Ή μ—μ΄μ „νŠΈκ°€ μ‚¬μš©μž λ¦¬μ†ŒμŠ€(이메일, μΊ˜λ¦°λ” λ“±)λ₯Ό 제3μžμ—κ²Œ κ³Όλ„ν•˜κ²Œ κ³΅μœ ν•˜λŠ” μ—μ΄μ „νŠΈ μ˜€λ²„μ…°μ–΄λ§ 문제λ₯Ό κ³΅μ‹ν™”ν•˜κ³  λΆ„μ„ν•œ 연ꡬ.
26.02.13 HF Daily Papers Nanbeige4.1-3B: A Small General Model that Reasons, Aligns, and Acts
β–² 0
3B νŒŒλΌλ―Έν„°λ§ŒμœΌλ‘œ μ—μ΄μ „νŠΈ 행동, μ½”λ“œ 생성, 일반 좔둠을 λ™μ‹œμ— λ‹¬μ„±ν•˜λŠ” 톡합 λ²”μš© μ–Έμ–΄ λͺ¨λΈ Nanbeige4.1-3B λ°œν‘œ. 졜초의 μ˜€ν”ˆμ†ŒμŠ€ μ†Œν˜• μ–Έμ–΄ λͺ¨λΈ(SLM)λ‘œμ„œ μ΄λŸ¬ν•œ λ‹€μž¬λ‹€λŠ₯함을 μ‹€ν˜„ν–ˆλ‹€.
26.02.11 HF Daily Papers DeepImageSearch: Benchmarking Multimodal Agents for Context-Aware Image Retrieval in Visual Histories
β–² 0
DeepImageSearch: μ‹œκ°μ  νžˆμŠ€ν† λ¦¬μ—μ„œ μ»¨ν…μŠ€νŠΈ 인식 이미지 검색을 μœ„ν•œ λ©€ν‹°λͺ¨λ‹¬ μ—μ΄μ „νŠΈ 벀치마크. 이미지 검색을 자율 탐색 과제둜 μž¬μ •μ˜ν•˜λŠ” μƒˆλ‘œμš΄ μ—μ΄μ „νŠΈ νŒ¨λŸ¬λ‹€μž„μ„ μ œμ‹œν•œλ‹€.
26.02.10 HF Daily Papers Benchmarking Knowledge-Extraction Attack and Defense on Retrieval-Augmented Generation
β–² 0
RAG(Retrieval-Augmented Generation)에 λŒ€ν•œ 지식 μΆ”μΆœ 곡격과 λ°©μ–΄λ₯Ό λ²€μΉ˜λ§ˆν‚Ή. μ—”ν„°ν”„λΌμ΄μ¦ˆ 챗봇, 의료 μ–΄μ‹œμŠ€ν„΄νŠΈ λ“±μ—μ„œ 지적 μž¬μ‚° λ„μš©κ³Ό ν”„λΌμ΄λ²„μ‹œ 유좜 μœ„ν—˜μ„ μ²΄κ³„μ μœΌλ‘œ ν‰κ°€ν•œλ‹€.
26.02.08 HF Daily Papers Data Darwinism Part I: Unlocking the Value of Scientific Data for Pre-training
β–² 0
Data Darwinism: 데이터 ν’ˆμ§ˆμ΄ νŒŒμš΄λ°μ΄μ…˜ λͺ¨λΈ μ„±λŠ₯을 κ²°μ •ν•˜μ§€λ§Œ 체계적 처리 ν”„λ ˆμž„μ›Œν¬κ°€ λΆ€μ‘±ν•œ 문제λ₯Ό ν•΄κ²°. 10단계 λΆ„λ₯˜μ²΄κ³„(L0-L9)λ₯Ό ν†΅ν•œ 데이터-λͺ¨λΈ 곡진화 κ°œλ…κ³Ό 900B 토큰 규λͺ¨μ˜ κ³Όν•™ μ½”νΌμŠ€λ₯Ό κ΅¬μΆ•ν–ˆλ‹€.
26.02.07 HF Daily Papers Blind to the Human Touch: Overlap Bias in LLM-Based Summary Evaluation
β–² 0
LLM νŒμ •μžκ°€ μš”μ•½ ν‰κ°€μ—μ„œ λ³΄μ΄λŠ” μ˜€λ²„λž© 편ν–₯(overlap bias) 뢄석. LLM νŒμ •μžκ°€ 길이, μˆœμ„œ λ“±μ˜ 편ν–₯을 가지며 μ λŒ€μ  μž…λ ₯에 μ·¨μ•½ν•œ 문제λ₯Ό μ„Έλ°€ν•˜κ²Œ λΆ„μ„ν•œλ‹€.