Skip to content

[DOCS#73] Boltzmann machine 계열 논문 5편 정리 (energy-based 신규 카테고리) - #74

Merged
juhy0987 merged 2 commits into
mainfrom
docs/#73/energy-based-papers
May 27, 2026
Merged

juhy0987 merged 2 commits into
mainfrom
docs/#73/energy-based-papers

Conversation

@juhy0987

Copy link
Copy Markdown
Member

연관 이슈


구현 내용

Phase 15 (PR #72) 완료 후 Phase 16+ 진입 전 historical foundation 정리. graph energy + Boltzmann machine 계열 의 핵심 5편 논문을 신규 energy-based/ 카테고리에 정리.

신규 디렉토리

docs/papers/energy-based/ — graph energy 표현 reference (Boltzmann machine / RBM / DBN / DBM)

5편 lineage

# 파일 의미
1 1985-boltzmann-machine-hinton.md Ackley+Hinton+Sejnowski, BM 원조 — energy + Gibbs sampling + contrastive update
2 1986-harmony-theory-smolensky.md Smolensky, RBM 원조 ("harmonium") — bipartite + conditional independence
3 2002-contrastive-divergence-hinton.md Hinton, CD-1 — RBM 실용 학습 breakthrough
4 2006-dbn-hinton.md Hinton+Osindero+Teh, DBN — layer-wise pre-training + 딥러닝 부활
5 2009-dbm-salakhutdinov.md Salakhutdinov+Hinton, DBM — all-layer undirected

시사점은 graph energy + paradigm 연결

각 논문의 "본 프로젝트 시사점" 섹션을 paradigm 관점으로 작성:

  • BMHybridGraphLinear 의 graph energy 등가성: E = -Σ w_ij s_i s_j vs eff_w = adj_outer · adj_inner · W
  • RBM bipartiteHybridGraphLinear 의 input → output channel 직접 등가
  • CD-1 의 "biased but practical" 철학 ↔ Phase 15 prune 의 magnitude proxy
  • DBN layer-wise ↔ Phase 16 후보 (Net2Net/LiGO) 의 historical precursor
  • DBM all-layer undirected ↔ Phase 14 full graph block 등가, Phase 17 (layer-wise 차등 prune) motivation

README.md 갱신

  • 디렉토리 구조 표에 energy-based/ 추가
  • 패러다임 분류 표에 새 행 추가 (graph energy reference / historical foundation)
  • 인덱스 끝에 Energy-based 섹션 신규 (5편 한 줄 요약 표)

CI / 머지 게이트 점검

변경 영향 범위

  • 영향 패키지/모듈: docs/papers/ (5 신규 + README 갱신). 코드 변경 없음.
  • 위험도: Low — 순수 문서 추가. CI 영향 Format Check / Build / Test 와 무관.

Required Status Checks

  • Commit Lint
  • PR Title Lint
  • Linked Issue Check
  • Format Check (코드 무변경)
  • Build / Test (코드 무변경)
  • Lint (make papers-lint 50개 frontmatter 통과)

롤백 계획

  • 순수 문서 추가만 — revert 시 영향 없음

다음 단계 (Phase 16 진입)

본 정리로 paradigm 의 graph energy historical roots 확립. Phase 16 (RigL/Net2Net) 진행 시 다음 시사점 활용 예정:

  • DBN/DBM layer-wise → Phase 16 의 progressive grow 방법론
  • CD statistics difference → Phase 16 의 edge importance metric 대안
  • DBM layer-wise differentiation → Phase 17 (layer-wise prune target) motivation

- docs/papers/energy-based/ 신규 디렉토리 + 5편 md 작성:
  - 1985-boltzmann-machine-hinton: Ackley+Hinton+Sejnowski, BM 원조 (energy + Gibbs sampling + contrastive update)
  - 1986-harmony-theory-smolensky: Smolensky, RBM 원조 (bipartite + conditional independence)
  - 2002-contrastive-divergence-hinton: Hinton, CD-1 (RBM 실용 학습 breakthrough)
  - 2006-dbn-hinton: Hinton+Osindero+Teh, DBN (layer-wise pre-training + 딥러닝 부활)
  - 2009-dbm-salakhutdinov: Salakhutdinov+Hinton, DBM (all-layer undirected)
- README.md 갱신: 디렉토리 구조 + 패러다임 분류 표 + Energy-based 인덱스 섹션 추가
- 시사점은 graph energy + paradigm 연결 관점으로 작성:
  - BM 의 E = -Σ w_ij s_i s_j 와 HybridGraphLinear 의 adj·W 구조 등가성
  - RBM bipartite 와 HybridGraphLinear input↔output channel 의 일치
  - CD-1 의 "biased but practical" 과 Phase 15 prune 의 magnitude proxy 철학
  - DBN/DBM 의 layer-wise growth 와 Phase 16 후보 (Net2Net/LiGO) 의 historical precursor
  - DBM 의 layer-wise differentiation 과 Phase 17 (attention vs FFN 차등 prune) motivation
- papers-lint 통과 (45 → 50편)
Copilot AI review requested due to automatic review settings May 27, 2026 01:12
@coderabbitai

coderabbitai Bot commented May 27, 2026

Copy link
Copy Markdown

Warning

Review limit reached

@juhy0987, we couldn't start this review because you've reached your PR review rate limit.

More reviews will be available in 34 minutes and 4 seconds. Learn how PR review limits work.

Your organization has run out of usage credits. Purchase more in the billing tab.

⌛ How to resolve this issue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

We recommend that you space out your commits to avoid hitting the rate limit.

🚦 How do rate limits work?

CodeRabbit enforces hourly rate limits for each developer per organization.

Our paid plans include higher PR review limits than trial, open-source, and free plans. In all cases, reviews become available again over time. During sustained high-volume PR review activity, CodeRabbit may temporarily slow when the next review becomes available.

Please see our Fair Usage Limits Policy for further information.

ℹ️ Review info
⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 0b288566-8572-45ab-a5ca-ec89fe5bd976

📥 Commits

Reviewing files that changed from the base of the PR and between 400e2d8 and b86cf63.

📒 Files selected for processing (6)
  • docs/papers/README.md
  • docs/papers/energy-based/1985-boltzmann-machine-hinton.md
  • docs/papers/energy-based/1986-harmony-theory-smolensky.md
  • docs/papers/energy-based/2002-contrastive-divergence-hinton.md
  • docs/papers/energy-based/2006-dbn-hinton.md
  • docs/papers/energy-based/2009-dbm-salakhutdinov.md
✨ Finishing Touches
🧪 Generate unit tests (beta)
  • Create PR with unit tests
  • Commit unit tests in branch docs/#73/energy-based-papers

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands and usage tips.

@juhy0987 juhy0987 added the documentation Improvements or additions to documentation label May 27, 2026

@gemini-code-assist gemini-code-assist Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request adds a new "energy-based" category to the papers documentation, introducing five new markdown files that summarize historical energy-based models (Boltzmann Machine, Harmonium/RBM, Contrastive Divergence, Deep Belief Network, and Deep Boltzmann Machine) and connect them to the project's GraphLM paradigm. The feedback identifies a factual error in the Smolensky 1986 summary, which incorrectly attributes the Contrastive Divergence algorithm to Hinton 1985 instead of Hinton 2002, and provides a direct suggestion to correct the year.

Comment thread docs/papers/energy-based/1986-harmony-theory-smolensky.md Outdated

Copilot AI left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Boltzmann machine 계열 5편을 docs/papers/energy-based/ 신규 카테고리로 추가하고, GraphLM의 graph energy 관점 reference로 README 인덱스를 확장하는 문서 전용 PR입니다.

Changes:

  • energy-based/ 카테고리 신설 및 BM/RBM/CD/DBN/DBM 논문 요약 5편 추가
  • docs/papers/README.md의 패러다임 분류, 디렉터리 구조, 논문 인덱스 갱신
  • 각 요약에 GraphLM HybridGraphLinear, Phase 15 prune, Phase 16/17 후보와의 연결점 정리

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated 3 comments.

Show a summary per file
File Description
docs/papers/README.md energy-based 카테고리와 5편 논문 인덱스 추가
docs/papers/energy-based/1985-boltzmann-machine-hinton.md Boltzmann Machine 원조 논문 요약 추가
docs/papers/energy-based/1986-harmony-theory-smolensky.md Harmonium/RBM 기원 논문 요약 추가
docs/papers/energy-based/2002-contrastive-divergence-hinton.md Contrastive Divergence 논문 요약 추가
docs/papers/energy-based/2006-dbn-hinton.md Deep Belief Network 논문 요약 추가
docs/papers/energy-based/2009-dbm-salakhutdinov.md Deep Boltzmann Machine 논문 요약 추가

Comment thread docs/papers/energy-based/1985-boltzmann-machine-hinton.md Outdated
Comment thread docs/papers/energy-based/1986-harmony-theory-smolensky.md Outdated
Comment thread docs/papers/energy-based/2006-dbn-hinton.md Outdated
- gemini #3307828144 + Copilot #3307833931 (중복): 1986 Smolensky md 의 CD 연도 1985 → 2002 (Hinton 2002 가 contrastive divergence). BM 의 positive/negative phase rule (1985) 과 RBM 의 CD-1 (2002) 구분 명확화
- Copilot #3307833906: 1985 BM md 의 prune 등가성 — `w_ij → -∞` 는 강한 negative interaction (두 unit 동시 1 강하게 억제) 이라 mask=0 (contribution 0) 과 다름. `w_ij = 0` 으로 수정 + 추가 설명
- Copilot #3307833938: 2006 DBN md 의 variational bound vs function preservation 등가화 오류 — DBN의 bound 는 data likelihood lower bound 보장 (generative quality), Net2Net 의 FP 는 forward output 정확 보존 (학습 안정성). 다른 종류 보장으로 분리 + spirit 만 공통 유지
@juhy0987 juhy0987 self-assigned this May 27, 2026
@juhy0987
juhy0987 merged commit 0b27027 into main May 27, 2026
9 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

documentation Improvements or additions to documentation

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[DOCS] Boltzmann machine 계열 논문 5편 정리 (energy-based 신규 카테고리)

2 participants