Files

T

Antigravity Agent f8b21af4be Wiki cleanup: error-doc removal, dedup merge, link normalization

10_Wiki/Topics 대규모 정리:
- 오류 캡처/미완성 stub 문서 227개 제거
- 교차폴더 중복 43클러스터 병합 (63파일 → redirect)
- 링크명 정규화: 깨진 링크 수정·redirect 직결·개념 매핑 ~2,400건
- 카테고리 MOC 6개 신규 생성
- Graph 섹션 미해결 related-keyword 링크 10,058건 제거

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-05-20 23:52:15 +09:00

7.0 KiB

Raw Blame History

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

title

Scientific Communication

매 한 줄

"매 result 의 가치는 매 communication 의 quality 에 bound — 매 unread paper 는 0 impact". 매 origin 은 1665 Philosophical Transactions 의 letter format; 매 modern state 는 IMRaD structure, preprint culture (arXiv, bioRxiv), open peer review, 그리고 매 AI-aided drafting (Claude Opus 4.7 의 paper review + outline).

매 핵심

매 IMRaD 구조 (paper)

Introduction: 매 gap → 매 question → 매 contribution.
Methods: 매 reproducible (data, code, hyperparams).
Results: 매 figures + tables, 매 narrative.
Discussion: 매 implication, limitation, future work.

매 audience layer

Title: 매 1-line — 매 99% 의 reader 가 only 보는 것.
Abstract: 매 250 words — 매 hook + result + implication.
Figure 1: 매 reader-grabbing visual.
Body: 매 0.5% 의 deep reader.

매 modern delivery

Preprint: arXiv (cs/stat/ml), bioRxiv, OSF — 매 priority claim.
Conf talk: 15min + Q&A — 매 pyramid (conclusion first).
Twitter/X thread: 매 paper drop 시 1 thread = 매 5x download.
Blog post: 매 distill.pub-style — 매 interactive.
Video: 매 5-min explainer (CVPR/NeurIPS 의 supplementary).

매 응용

Research paper writing.
Grant proposal.
Conference talk.
Tech blog (engineering science).

💻 패턴

매 LaTeX paper skeleton (NeurIPS 2026 style)

\documentclass{article}
\usepackage[final]{neurips_2026}
\usepackage{graphicx, amsmath, hyperref, cleveref}
\usepackage[capitalize, noabbrev]{cleveref}

\title{<Catchy Title: Method on Task with Number\%>}
\author{Alice Smith \\ Acme Lab \\ \texttt{alice@acme.com}}

\begin{document}
\maketitle

\begin{abstract}
We address <gap>. We propose <method>. On <benchmark>, our approach
achieves <X\%> ($\Delta$+Y\% over prior best). Code: \url{...}.
\end{abstract}

\section{Introduction}
\input{sections/intro}

\section{Method}
\input{sections/method}

\section{Experiments}
\input{sections/experiments}

\section{Related Work}
\input{sections/related}

\section{Conclusion}
\input{sections/conclusion}

\bibliographystyle{plainnat}
\bibliography{refs}
\end{document}

매 abstract template (250 words, 매 6-sentence hook)

1. [Context]      <Field> has long pursued <goal>.
2. [Gap]          However, existing methods <limit>.
3. [Insight]      We observe that <key insight>.
4. [Method]       Building on this, we propose <method>:
                  <2-sentence description>.
5. [Result]       On <benchmark>, our method achieves <X%>,
                  improving over <baseline> by <Δ>.
6. [Impact]       This suggests <broader implication> and
                  enables <downstream application>.

매 figure 1 ("teaser" — 매 abstract 의 visual)

매 좋은 figure 1 의 5 rule:
1. 매 self-contained — caption 만 읽고 message 이해 가능.
2. 매 axes labeled, units 명시.
3. 매 baseline + ours comparison (color-blind safe).
4. 매 ≤ 3 message — 매 더 많으면 split.
5. 매 vector format (PDF) — 매 raster 의 X.

매 talk pyramid (15-min conf talk)

00:00 — Hook (1 slide, 매 result tease)
01:00 — Problem (2 slides, 매 why care?)
03:00 — Insight (1 slide, 매 핵심 idea)
04:00 — Method (3-4 slides, 매 just enough)
08:00 — Results (3-4 slides, 매 main + ablation)
12:00 — Limitation (1 slide, 매 honest)
13:00 — Take-aways (1 slide, 매 3 bullets)
14:00 — Q&A

매 Twitter/X thread 의 paper drop

1/ 매 [TL;DR] — 매 1 sentence + result number + figure.
2/ Why does this matter? (매 problem framing)
3/ Insight (매 1 핵심 idea, 매 figure)
4/ How (매 architecture, 매 다이어그램)
5/ Results (매 main number, 매 baseline 대비)
6/ Limitations (매 honest)
7/ Code + paper + colab links.

매 Claude Opus 4.7 paper-review prompt (1M ctx, 매 full paper)

import anthropic
client = anthropic.Anthropic()

paper_pdf_text = open("draft.txt").read()  # 매 full paper

msg = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=8192,
    system=(
        "You are a senior NeurIPS reviewer. Review this paper:\n"
        "1. Summary (3 sentences).\n"
        "2. Strengths (3 bullets).\n"
        "3. Weaknesses (3 bullets, specific section refs).\n"
        "4. Questions for authors (3).\n"
        "5. Score (1-10) with justification.\n"
        "Be specific, cite line numbers, no generic praise."
    ),
    messages=[{"role": "user", "content": paper_pdf_text}],
)
print(msg.content[0].text)

매 distill-style explainer (interactive, 매 React + MDX)

// 매 D3 + MDX 로 매 inline interactive figure
import { useState } from "react";
import { MathJax } from "better-react-mathjax";

export const TempScaling = () => {
  const [T, setT] = useState(1.0);
  return (
    <div>
      <p>Temperature <code>T={T.toFixed(2)}</code></p>
      <input type="range" min={0.1} max={5} step={0.1}
             value={T} onChange={e => setT(+e.target.value)} />
      <SoftmaxPlot T={T} />
      <MathJax>{`$$p_i = \\frac{e^{z_i/T}}{\\sum_j e^{z_j/T}}$$`}</MathJax>
    </div>
  );
};

매 결정 기준

상황	Approach
매 conference paper	IMRaD + LaTeX + arXiv preprint
매 industry blog	distill-style + interactive figures
매 talk (15 min)	pyramid: conclusion → method → results
매 social drop	thread 7-tweet + figure 1 + code link
매 grant	story arc: problem → impact → method → milestones

기본값: paper 면 IMRaD + arXiv + 1-thread X drop + Claude Opus 4.7 self-review.

🔗 Graph

부모: Research Methodology · Technical Writing

🤖 LLM 활용

언제: 매 abstract 의 6-sentence drafting. 매 paper self-review (Claude Opus 4.7 1M ctx). 매 X thread draft. 매 grammar/clarity pass. 언제 X: 매 result fabrication. 매 method 의 invention. 매 LLM 의 "novel contribution" claim.

❌ 안티패턴

Methods-section first sentence: 매 reader 가 not knowing 'why' 도착. 매 motivation lead.
Equation salad: 매 prose 없이 equation 만 — 매 narrative 필요.
Result-only abstract: 매 context 없이 number 만.
AI-generated filler: 매 reviewer 가 매 hallmark detect.
Buried lead: 매 main result 가 page 8 — 매 figure 1 으로.

🧪 검증 / 중복

Verified (Mensh & Kording "Ten simple rules for structuring papers", Pinker "Sense of Style", NeurIPS guidelines).
신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — IMRaD + LaTeX + Claude Opus 4.7 review + distill

7.0 KiB Raw Blame History