⚠️ 已归档 (Archived)

本仓库已归档，不再维护。

请使用最新版知识库：xuanshu-knowledge-base

• 📊 524篇论文（本仓库237篇的2.2倍）
• 📁 30个分类（本仓库11个的2.7倍）
• 🧠 含AGI认知地图、认知架构调研、前沿补充

本仓库保留作为历史参考。

---

📚 Materials Science ML Knowledge Base

237 papers · 11 categories · ~9,000 lines of structured knowledge · Built for materials science students starting ML from zero

---

✨ Why This Exists

Chinese-language materials science ML knowledge is extremely scarce. If you're a materials science undergrad wanting to learn ML for materials research, you'll hit two walls:

1. Papers are unreadable — English ML papers are dense with jargon. Materials students struggle to cross the terminology gap.
2. Code doesn't run — Papers come with code, but environment setup, data formats, and hyperparameters are minefields.

This knowledge base tries to bridge the "materials person learns ML" information gap — systematically organizing core materials + ML crossover knowledge in Chinese, building a bridge for those who come after.

---

📊 Coverage

KB	Files	Size	Content
KB0 Global Index	1	2,855 lines	~200 papers, full index, 11 categories
KB1 Materials ML Core	8	~2,448 lines	Databases → descriptors → property prediction → high-throughput → GNNs
KB2 Theory & Foundations	3	~887 lines	ML fundamentals / DL optimization / Bayesian methods
KB3 Models & Tools	4	~1,477 lines	CNN vision / RNN sequences / RL / tooling & frameworks
KB4 Language & Generation	2	~775 lines	Generative models / NLP & LLMs
KB5 Entertainment	1	~349 lines	Zenless Zone Zero / Steam / anime / music

Accuracy: ~99% (7 errors fixed, 3 unconfirmed)

---

🎯 Who It's For

• Materials science undergrads / grad students starting ML from zero
• Anyone wanting to quickly grasp "what can materials × AI do"
• Researchers who need a paper roadmap (which 3 to read first, not 200)

---

🗂️ Structure

materials-kb/
├── KB0_全量论文索引/          # Global index — quick lookup
│   └── KB0_全量论文索引.md      (2,855 lines, 200+ papers)
├── KB1_材料ML核心/             # Materials science ML methodology
│   ├── 01_材料数据库.md          (Materials Project / AFLOW / OQMD)
│   ├── 02_材料表示与描述符.md     (Composition / structure / electronic descriptors)
│   ├── 03_性质预测.md            (Band gap / modulus / stability prediction)
│   ├── 04_高通量筛选与逆向设计.md
│   ├── 05_特定材料系统.md         (Perovskites / batteries / catalysis)
│   ├── 06_综述论文.md
│   ├── 07_新兴方向.md
│   └── 08_图神经网络.md          (CGCNN / ALIGNN / MEGNet)
├── KB2_理论与基础/             # ML theoretical foundations
│   ├── 01_ML基础.md
│   ├── 02_DL优化.md
│   └── 03_贝叶斯方法.md
├── KB3_模型与工具/             # Specific model architectures & tools
│   ├── 01_CNN视觉.md
│   ├── 02_RNN序列.md
│   ├── 03_强化学习.md
│   └── 04_工具框架.md
├── KB4_语言与生成/             # Generative models & LLMs
│   ├── 01_生成模型.md
│   └── 02_NLP与LLM.md
├── KB5_小叶娱乐库/             # Entertainment (demonstration)
│   └── KB5_小叶娱乐库.md
├── scripts/
│   ├── kb_validate.py          # KB format validator
│   └── kb_stats.py             # KB statistics
└── fine-tune/
    ├── qa-extraction.md         # QA pair extraction methodology
    └── sft-dataset-format.md    # SFT dataset format spec

---

🚀 Usage

Method 1: Read Directly

Clone and read in order: KB0 → KB2 → KB3 → KB1. KB0 is the map, KB2/KB3 are foundations, KB1 is application.

Method 2: As Ollama RAG Knowledge Base

Upload KB files as knowledge documents in Open WebUI. Models can retrieve directly.

Method 3: Fine-Tuning Dataset Construction

Extract QA pairs from KB → format conversion → LLaMA-Factory fine-tuning → custom materials-science-specialized model. See fine-tune/qa-extraction.md.

---

📈 Statistics

Generated by scripts/kb_stats.py (run python scripts/kb_stats.py for latest):

• Total entries, code snippets, cross-references
• Entry distribution by KB
• Terminology coverage

---

🔍 Niche

Search GitHub for "materials science ML Chinese knowledge base" and you'll find virtually nothing. Existing materials databases (Materials Project, AFLOW) are data warehouses, not knowledge bases — they provide raw data, not answers to "why this paper matters", "how this method differs from that one", or "which 3 papers should a beginner read first".

This repo fills that gap: not a paper list, but a navigation map. This knowledge base is the concrete backing for the "Materials Science × AI intersection" differentiator in the 熔炉协议 (Crucible Protocol).

---

🤝 Contributing

Materials science students welcome!

How to Contribute

1. Fork this repo
2. Add entries to the relevant KB file (keep formatting consistent)
3. Run python scripts/kb_validate.py to validate
4. Submit a PR

Entry Format

### [Paper Title](link)
- **Authors**: Author et al. (Year)
- **Core Method**: Method name
- **Material System**: What material
- **Key Result**: One-line core finding
- **Code**: [link]()
- **For Beginners**: Readability rating (⭐ ~ ⭐⭐⭐)

---

📄 License

MIT License — see LICENSE

---

"Materials are the universe's coded letters to humanity. Machine learning is the decoder key." — xuanhan