跳转至

全局概览 · AI 全领域地图 (跨 25 章)

一句话总纲: 本套书覆盖数学基础 → 机器学习 → 深度学习 → 多模态 → 应用 → 工程 → 前沿 → 对齐评估 → Agent → 面试, 25 章构成 AI 工程师"从理论到落地到面试"的完整地图。本文档是跨章学习路径 + 领域全景。

配套文档: 整本书脉络地图.md — 25 章完整骨架, 含 7 主题地图 / 跨章节索引 / 50+ 论文清单 / 9 角色速成路径。

中译: 本教辅资料由 Steven Chen (godlockin) 基于自研智能体框架 Kallax 配合多模型协作完成。原文：HenryNdubuaku/maths-cs-ai-compendium，中译：godlockin/maths-cs-ai-compendium。

版本: v1.0 (2026-07-31) — 覆盖 Ch21-25 (对齐评估/Agent/数值优化/面试)

一 · 25 章全景图

flowchart TB
    subgraph 基础 [数学/工具]
        C01[Ch01 线性代数]
        C02[Ch02 矩阵论]
        C03[Ch03 概率与梯度]
        C04[Ch04 统计]
        C05[Ch05 贝叶斯与信息论]
    end
    subgraph 核心 [机器学习]
        C06[Ch06 经典ML与RL]
        C07[Ch07 NLP与Transformer]
        C08[Ch08 CNN与Diffusion]
    end
    subgraph 多模态 [感知与生成]
        C09[Ch09 语音]
        C10[Ch10 多模态]
        C11[Ch11 自主系统]
    end
    subgraph 专题 [专题与系统]
        C12[Ch12 GNN]
        C13[Ch13 系统与OS]
        C14[Ch14 数据结构与算法]
    end
    subgraph 工程 [工程与生产]
        C15[Ch15 软件工程]
        C16[Ch16 SIMD/GPU]
        C17[Ch17 AI 推理]
        C18[Ch18 ML 系统设计]
    end
    subgraph 应用与前沿 [落地与未来]
        C19[Ch19 应用 AI]
        C20[Ch20 前沿 AI]
    end
    subgraph 高阶 [对齐·评估·Agent·优化·面试]
        C21[Ch21 对齐安全可解释性]
        C22[Ch22 LLM Evaluation]
        C23[Ch23 AI Agent]
        C24[Ch24 数值分析与凸优化]
        C25[Ch25 AI 面试指南]
    end

    C01 --> C02 --> C03 --> C04 --> C05 --> C06 --> C07 --> C08 --> C09 --> C10
    C10 --> C11 --> C12 --> C13 --> C14 --> C15 --> C16 --> C17 --> C18 --> C19
    C19 --> C20 --> C21 --> C22 --> C23 --> C24 --> C25

    C07 -.-> C21
    C21 -.-> C22
    C22 -.-> C23
    C06 -.-> C24
    C18 -.-> C25

二 · 学习路径 (3 条)

路径 A: 算法研究员 (Ch01-Ch08)

顺序 章节 关键能力
1 Ch01 线性代数 向量空间, 范数
2 Ch02 矩阵论 SVD, 谱
3 Ch03 概率与梯度 优化
4 Ch04 统计 估计, 检验
5 Ch05 贝叶斯 先验, 后验
6 Ch06 ML + RL 经典 ML
7 Ch07 NLP + Transformer LLM
8 Ch08 CNN + Diffusion CV + 生成

路径 B: 应用工程师 (Ch07, Ch11-Ch18)

顺序 章节 关键能力
1 Ch07 Transformer LLM 基础
2 Ch11 自主系统 VLA
3 Ch12 GNN 图数据
4 Ch13 系统 OS, 架构
5 Ch14 算法 内功
6 Ch15 工程 工程素养
7 Ch16 SIMD/GPU 性能
8 Ch17 推理 部署
9 Ch18 系统设计 平台

路径 C: 全栈 / AGI 研究员 (全部)

按 1-20 顺序通读, 是 6 个月速成路径。

三 · 跨章核心知识串联

3.1 数学到 ML

Ch01 向量 → Ch02 矩阵 → Ch03 梯度
                     ↓
                Ch04 统计 → Ch05 贝叶斯 → Ch06 ML 算法
                                       ↓
                                Ch07-Ch08 深度学习

3.2 Transformer 全景

Ch07 §04 Transformer
   ↓
   ├─→ Ch10 §05 多模态 (CLIP / BLIP)
   ├─→ Ch11 §05 VLA (Pi0, OpenVLA)
   ├─→ Ch16 §04 Flash Attention 加速
   ├─→ Ch17 §02/03 vLLM / TGI 部署
   ├─→ Ch19 §04 Agent (RAG + Tool)
   └─→ Ch20 §05 BMI 解码

3.3 图与网络

Ch12 §02 图论
   ↓
   ├─→ Ch12 §03 GNN (GCN / SAGE / GIN)
   ├─→ Ch12 §04 GAT / Graphormer
   ├─→ Ch12 §05 3D / 等变 (AlphaFold)
   └─→ Ch18 §05 推荐系统 (PinSage)

3.4 系统栈

Ch13 OS / Ch14 算法
   ↓
Ch15 DevOps (Docker / K8s)
   ↓
Ch16 SIMD / GPU
   ↓
Ch17 推理 (量化 / vLLM)
   ↓
Ch18 ML 平台 (Feature Store / 监控)

四 · AI 全领域地图 (2024 视角)

mindmap
  root((AI 2024))
    模型
      LLM
        GPT / Claude / Gemini
        开源: Llama / Mistral / Qwen
      多模态
        GPT-4o / Gemini Pro Vision
        视觉: SAM / DINOv2
        语音: Whisper / VALL-E
      专用
        AlphaFold (蛋白)
        Codex (代码)
        DALL-E (图像)
    基础设施
      硬件
        H100 / B200 (NVIDIA)
        TPU v5e (Google)
        神经形态 (Loihi)
      软件
        PyTorch + Triton
        vLLM / TGI
        LangChain / LlamaIndex
      云
        AWS / GCP / Azure
        K8s / Docker
    应用
      金融
        量化
        风控
      生命科学
        蛋白设计
        药物发现
      Agent
        AutoGPT
        Claude Computer Use
      医疗
        AlphaFold 临床
        影像分割
    前沿
      量子 ML
      神经形态
      太空数据中心
      去中心化 AI
      脑机接口

五 · 章节依赖图

章节 前置 后续
Ch01 线性代数 — 全部
Ch02 矩阵论 Ch01 Ch12, Ch08, Ch14
Ch03 概率 Ch01, Ch02 全部 ML
Ch04 统计 Ch03 全部
Ch05 贝叶斯 Ch03, Ch04 Ch06, Ch18
Ch06 ML Ch03, Ch04 Ch07, Ch11
Ch07 Transformer Ch06 Ch10, Ch11, Ch17, Ch19
Ch08 CNN Ch06 Ch09, Ch10
Ch09 语音 Ch07, Ch08 Ch10
Ch10 多模态 Ch07, Ch08, Ch09 Ch11
Ch11 自主系统 Ch06, Ch07 Ch18
Ch12 GNN Ch02, Ch06 Ch14, Ch18
Ch13 系统 — Ch15, Ch16
Ch14 算法 Ch13 Ch15, Ch17
Ch15 工程 Ch14 Ch17, Ch18
Ch16 SIMD Ch13 Ch17
Ch17 推理 Ch15, Ch16 Ch18, Ch19
Ch18 系统设计 Ch15, Ch17 Ch19
Ch19 应用 Ch07-Ch18 —
Ch20 前沿 全部 —

六 · 领域完整清单 (按学科)

6.1 数学

  • 线性代数 (Ch01)
  • 矩阵论 (Ch02)
  • 概率统计 (Ch03-04)
  • 贝叶斯 / 信息论 (Ch05)
  • 离散数学 (Ch13 §01)

6.2 机器学习

  • 经典 ML (Ch06 §01-02)
  • 强化学习 (Ch06 §04)
  • 深度学习 (Ch06 §03, Ch07-08)

6.3 自然语言

  • 文本预处理 (Ch07 §01)
  • 词嵌入 (Ch07 §02)
  • RNN / LSTM (Ch07 §03)
  • Transformer (Ch07 §04)
  • 预训练 LLM (Ch07 §05)
  • 函数式编程 (Ch06 §04, FP 视角)

6.4 计算机视觉

  • CNN (Ch08 §01)
  • 现代架构 (Ch08 §02)
  • 目标检测 (Ch08 §03)
  • Diffusion (Ch08 §04)
  • 视频 / 3D (Ch08 §05)
  • VLA (Ch11 §05)

6.5 语音

  • 声学模型 (Ch09 §02)
  • ASR (Ch09 §03)
  • TTS (Ch09 §05)

6.6 多模态

  • 视觉语言 (Ch10 §03)
  • 跨模态对齐 (Ch10 §04)
  • 多模态 LLM (Ch10 §05)

6.7 自主系统

  • 感知 / SLAM (Ch11 §01)
  • 决策 (Ch11 §02)
  • 控制 (Ch11 §03)
  • VLA (Ch11 §05)

6.8 图网络

  • 几何深度学习 (Ch12 §01)
  • 图论 (Ch12 §02)
  • GNN (Ch12 §03)
  • GAT (Ch12 §04)
  • 3D / 等变 (Ch12 §05)

6.9 系统

  • 离散数学 (Ch13 §01)
  • 计算机架构 (Ch13 §02)
  • OS (Ch13 §03)
  • 并发并行 (Ch13 §04)
  • 编程语言 (Ch13 §05)
  • 数据结构与算法 (Ch14)

6.10 工程

  • Linux (Ch15 §01)
  • Git (Ch15 §02)
  • 代码设计 (Ch15 §03)
  • 测试 (Ch15 §04)
  • DevOps (Ch15 §05)
  • SIMD / GPU (Ch16)
  • AI 推理 (Ch17)
  • ML 系统设计 (Ch18)

6.11 应用

  • 金融 (Ch19 §01)
  • 蛋白设计 (Ch19 §02)
  • 药物发现 (Ch19 §03)
  • Agent (Ch19 §04)
  • 医疗 (Ch19 §05)

6.12 前沿

  • 量子 ML (Ch20 §01)
  • 神经形态 (Ch20 §02)
  • 太空数据中心 (Ch20 §03)
  • 去中心化 AI (Ch20 §04)
  • 脑机接口 (Ch20 §05)

七 · 重要模型 / 工具 / 论文时间线

timeline
    2012 : AlexNet, ImageNet
    2014 : GAN
    2017 : Transformer / AlphaGo Zero
    2018 : BERT / GPT-1
    2019 : GPT-2 / AlphaFold 1
    2020 : GPT-3 / AlphaFold 2
    2021 : DALL-E / GitHub Copilot
    2022 : ChatGPT / AlphaTensor / Stable Diffusion
    2023 : GPT-4 / Llama 2 / AlphaFold 3
    2024 : GPT-4o / Claude 3 / Llama 3 / Gemini 1.5 / OpenAI o1 / AlphaFold 3
    2025 : DeepSeek-V3 (MoE + FP8) / Claude 4 / Llama 4 / Gemini 2.5 / GPT-5 / Agent 平台 / Mamba-2 / DiT
    2026H1 : Claude 4.5 Opus / GPT-5.1 / Gemini 3 / 开源 DeepSeek-V4 / 推理 Scaling Law 成熟 / Embodied AI 工业落地

八 · 教辅索引

章节 信息图 大纲 思维导图 闪卡 阶段测试 讲解稿
Ch01-10 ✅ ✅ ✅ ✅ ✅ ✅ 总览 + 5节
Ch07 讲解 ✅ ✅ ✅ ✅ ✅ ✅ 总览 + 5节 + §06 2024-2026 前沿专题
Ch08 讲解 ✅ ✅ ✅ ✅ ✅ ✅ 总览 + 5节 + §06 2024-2026 多模态生成前沿
Ch11 自主系统 ✅ ✅ ✅ ✅ ✅ ✅ 5
Ch09 语音 ✅ ✅ ✅ ✅ ✅ ✅ 总览 + 5节 + §06 2024-2026 语音前沿
Ch12 GNN ✅ ✅ ✅ ✅ ✅ ✅ 总览 + 5节 + §06 2024-2026 3D GNN 前沿
Ch13 系统 ✅ ✅ ✅ ✅ ✅ ✅ 5
Ch14 数据结构 ✅ ✅ ✅ ✅ ✅ ✅ 6
Ch15 软件工程 ✅ ✅ ✅ ✅ ✅ ✅ 5
Ch16-SIMD/GPU ✅ ✅ ✅ ✅ ✅ ✅ 1 总览
Ch17 AI 推理 ✅ ✅ ✅ ✅ ✅ ✅ 总览 + 5节 + §02 2024-2026 高效架构前沿 + §06 大模型微调/蒸馏/量化专题
Ch18 ML 系统 ✅ ✅ ✅ ✅ ✅ ✅ 1 总览
Ch19 应用 AI ✅ ✅ ✅ ✅ ✅ ✅ 1 总览
Ch20 前沿 AI ✅ ✅ ✅ ✅ ✅ ✅ 1 总览

共: 202+7=209 个教学材料文件 (新增 7 个 2024-2026 前沿专题)

九 · 总结: AI 全栈工程师成长路线

阶段 1: 基础 (1-2 月)

  • Ch01-05 数学
  • Ch06 ML

阶段 2: 深度学习 (2-4 月)

  • Ch07 Transformer
  • Ch08 CNN / Diffusion

阶段 3: 多模态 + 自主 (4-6 月)

  • Ch09-11 感知

阶段 4: 专题 (6-8 月)

  • Ch12 GNN
  • Ch14 算法

阶段 5: 工程化 (8-10 月)

  • Ch13 系统
  • Ch15 DevOps
  • Ch16 SIMD / GPU
  • Ch17 推理

阶段 6: 系统设计 (10-12 月)

  • Ch18 ML 系统设计

阶段 7: 应用落地 (12-15 月)

  • Ch19 应用 AI

阶段 8: 前沿展望 (持续)

  • Ch20 前沿 AI

十 · 一句话总结

本套书 = AI 全栈工程师 6 个月速成手册。从数学基础到应用前沿, 20 章覆盖 60+ 子领域, 100+ 主流模型, 50+ 工具, 220+ 论文 (按章节主题分组), 形成完整知识地图。是 AI 工程师"理论 + 实战 + 前沿"三合一的宝典。

附录: 推荐学习资源

教材

  • Bishop, "Pattern Recognition and Machine Learning"
  • Goodfellow, "Deep Learning"
  • Sutton & Barto, "Reinforcement Learning"
  • Russell & Norvig, "Artificial Intelligence: A Modern Approach"

论文 (按章节主题分组, 共 220+ 篇)

Ch01-05 数学基础

  1. Strassen, "Gaussian Elimination is not Optimal", 1969
  2. Golub & Kahan, "Singular Value Decomposition", 1965
  3. Trefethen & Bau, "Numerical Linear Algebra", 1997
  4. Horn & Johnson, "Matrix Analysis", 1985
  5. Cover & Thomas, "Elements of Information Theory", 1991
  6. Jaynes, "Probability Theory: The Logic of Science", 2003
  7. Bernardo & Smith, "Bayesian Theory", 1994
  8. Pearl, "Causality", 2009
  9. Shalev-Shwartz & Ben-David, "Understanding Machine Learning", 2014
  10. Boyd & Vandenberghe, "Convex Optimization", 2004
  11. Bertsekas, "Nonlinear Programming", 1999
  12. Nocedal & Wright, "Numerical Optimization", 2006

Ch06 经典 ML & RL

  1. Rosenblatt, "The Perceptron", 1958
  2. Quinlan, "C4.5 Decision Trees", 1993
  3. Cortes & Vapnik, "Support-Vector Networks", 1995
  4. Breiman, "Random Forests", Machine Learning 2001
  5. Friedman, "Greedy Function Approximation (Gradient Boosting)", 1999
  6. Chen & Guestrin, "XGBoost", KDD 2016
  7. Ke et al., "LightGBM", NeurIPS 2017
  8. Hochreiter & Schmidhuber, "LSTM", Neural Computation 1997
  9. Williams & Zipser, "Backpropagation", 1989
  10. LeCun et al., "Gradient-Based Learning Applied to Document Recognition", 1998
  11. Krizhevsky et al., "ImageNet Classification with Deep CNNs", NeurIPS 2012
  12. Mnih et al., "Human-level Control through Deep RL", Nature 2015
  13. Silver et al., "Mastering the Game of Go without Human Knowledge (AlphaGo Zero)", Nature 2017
  14. Schulman et al., "PPO", arXiv 2017
  15. Lillicrap et al., "DDPG", ICLR 2016
  16. Schulman et al., "TRPO", ICML 2015
  17. Haarnoja et al., "Soft Actor-Critic (SAC)", 2018
  18. Fujimoto et al., "TD3", 2018
  19. Sutton, "The Bitter Lesson", 2019

Ch07 NLP & Transformer

  1. Mikolov et al., "Word2Vec", NeurIPS 2013
  2. Pennington et al., "GloVe", EMNLP 2014
  3. Peters et al., "ELMo", NAACL 2018
  4. Vaswani et al., "Attention Is All You Need", NeurIPS 2017
  5. Devlin et al., "BERT", NAACL 2019
  6. Radford et al., "GPT-1", 2018
  7. Radford et al., "GPT-2", 2019
  8. Brown et al., "GPT-3", NeurIPS 2020
  9. Raffel et al., "T5", JMLR 2020
  10. Su et al., "RoPE", 2021
  11. Press et al., "ALiBi", ICLR 2022
  12. Shazeer, "GLU Variants Improve Transformer", 2020
  13. Su et al., "One Billion Word Benchmark", 2019
  14. Kaplan et al., "Scaling Laws for Neural LMs", 2020
  15. Hoffmann et al., "Chinchilla Scaling", 2022
  16. Ouyang et al., "InstructGPT / RLHF", NeurIPS 2022
  17. Christiano et al., "Deep RL from Human Preferences", NeurIPS 2017
  18. Stiennon et al., "Learning to Summarize with Human Feedback", 2020
  19. Rafailov et al., "DPO", NeurIPS 2023
  20. Bai et al., "Constitutional AI", 2022
  21. Lee et al., "RLAIF", 2023
  22. DeepSeek-AI, "GRPO", 2024
  23. Chen et al., "Self-Play Fine-Tuning (SPIN)", 2024
  24. Glaese et al., "Sparrow", 2022
  25. Touvron et al., "LLaMA", 2023
  26. Touvron et al., "LLaMA 2 / 3", 2023-2024
  27. Bai et al., "Qwen / Qwen 2", 2023-2024
  28. Yang et al., "Yi", 2024
  29. Bi et al., "DeepSeek LLM", 2024

Ch08 CNN & Diffusion

  1. LeCun et al., "LeNet-5", 1998
  2. Simonyan & Zisserman, "VGG", ICLR 2015
  3. Szegedy et al., "GoogLeNet", CVPR 2015
  4. He et al., "ResNet", CVPR 2016
  5. Huang et al., "DenseNet", CVPR 2017
  6. Howard et al., "MobileNet", 2017
  7. Tan & Le, "EfficientNet", ICML 2019
  8. Dosovitskiy et al., "ViT", ICLR 2021
  9. Liu et al., "Swin Transformer", ICCV 2021
  10. He et al., "Mask R-CNN", ICCV 2017
  11. Redmon et al., "YOLO", CVPR 2016
  12. Liu et al., "SSD", ECCV 2016
  13. Carion et al., "DETR", ECCV 2020
  14. Ronneberger et al., "U-Net", MICCAI 2015
  15. Long et al., "FCN", CVPR 2015
  16. Ho et al., "DDPM", NeurIPS 2020
  17. Song et al., "DDIM", ICLR 2021
  18. Nichol & Dhariwal, "Improved DDPM", ICML 2021
  19. Rombach et al., "Latent Diffusion / SD", CVPR 2022
  20. Ho et al., "Classifier-Free Guidance", 2022
  21. Peebles & Xie, "DiT", ICCV 2023
  22. Chen et al., "PixArt-α", 2023
  23. Esser et al., "SD3 / MM-DiT", 2024
  24. Labs, "FLUX", 2024
  25. Zhao et al., "Sana", 2024
  26. OpenAI, "Sora", 2024
  27. Google, "Veo / Veo 2", 2024
  28. Meta, "Movie Gen", 2024
  29. Stability AI, "Stable Video Diffusion", 2023
  30. Blattmann et al., "Align your Latents (Video)", CVPR 2023

Ch09-10 语音与多模态

  1. Vaswani et al., "Listen, Attend and Spell", ICASSP 2016
  2. Baevski et al., "wav2vec 2.0", NeurIPS 2020
  3. Radford et al., "Whisper", ICML 2023
  4. van den Oord et al., "WaveNet", 2016
  5. Wang et al., "Tacotron 2", ICASSP 2018
  6. Ren et al., "FastSpeech 2", ICML 2021
  7. Wang et al., "VALL-E", 2023
  8. Radford et al., "CLIP", ICML 2021
  9. Li et al., "BLIP / BLIP-2", 2022-2023
  10. Liu et al., "LLaVA", NeurIPS 2023
  11. Alayrac et al., "Flamingo", NeurIPS 2022
  12. Bai et al., "Qwen-VL", 2023
  13. Chen et al., "InternVL", 2024
  14. DeepSeek, "Janus-Pro", 2025
  15. Hurst et al., "GPT-4o System Card", 2024
  16. Wang et al., "Gemini 1.5 / 2.0 / 2.5", 2024-2025
  17. Anthropic, "Claude 3.5 / 4 / 4.5", 2024-2025
  18. Meta, "Llama 3 / 4 (Multimodal)", 2024-2025
  19. Su et al., "OneFormer", 2022
  20. Kirillov et al., "SAM", ICCV 2023

Ch11 自主系统

  1. Cadena et al., "Past, Present, and Future of SLAM", 2016
  2. Mur-Artal & Tardós, "ORB-SLAM2", IEEE TR-R 2017
  3. Thrun et al., "Probabilistic Robotics", 2005
  4. Kolve et al., "AI2-THOR", 2017
  5. Savva et al., "Habitat", ICCV 2019
  6. Brohan et al., "RT-2", 2023
  7. Brohan et al., "OpenVLA", 2024
  8. Black et al., "Pi0", 2024
  9. Tian et al., "Octo", 2024
  10. Leal-Taixé et al., "MOT Challenge", 2015
  11. Redmon & Farhadi, "YOLOv3", 2018
  12. Wang et al., "YOLOv8", 2023

Ch12 GNN

  1. Scarselli et al., "Graph Neural Network", IEEE TNN 2008
  2. Kipf & Welling, "GCN", ICLR 2017
  3. Hamilton et al., "GraphSAGE", NeurIPS 2017
  4. Xu et al., "GIN", ICLR 2019
  5. Veličković et al., "GAT", ICLR 2018
  6. Ying et al., "Graphormer", NeurIPS 2021
  7. Gilmer et al., "MPNN", ICML 2017
  8. Bronstein et al., "Geometric Deep Learning", 2021
  9. Jumper et al., "AlphaFold 2", Nature 2021
  10. Abramson et al., "AlphaFold 3", Nature 2024
  11. Ying et al., "PinSage", KDD 2018
  12. Zhang et al., "GNN recommender", 2019

Ch13-14 系统与算法

  1. Hennessy & Patterson, "Computer Architecture", 6th Ed, 2017
  2. Bryant & O'Hallaron, "Computer Systems: A Programmer's Perspective", 2016
  3. Tanenbaum, "Modern Operating Systems", 2014
  4. Cormen et al., "Introduction to Algorithms (CLRS)", 4th Ed, 2022
  5. Lamport, "Time, Clocks, and Ordering", CACM 1978
  6. Dijkstra, "Cooperating Sequential Processes", 1968
  7. Hoare, "Communicating Sequential Processes", 1978
  8. Akhmediev & Ankiewicz, "Solitons", 1997
  9. Lamport, "Paxos", 1998
  10. Ongaro & Ousterhout, "Raft", USENIX ATC 2014

Ch16 SIMD / GPU

  1. Hennessy & Patterson, "GPU Computing", 2017
  2. NVIDIA, "CUDA Programming Guide", 2024
  3. Vasilyev et al., "AVX-512", 2018
  4. ARM, "SVE / SVE2", 2021
  5. Khronos, "SYCL / OpenCL", 2024
  6. Chetlur et al., "cuDNN", 2014
  7. Paszke et al., "PyTorch", NeurIPS 2019
  8. Bradbury et al., "JAX", 2018
  9. Tillet et al., "Triton", MAPL 2019

Ch17 AI 推理与高效架构

  1. Kwon et al., "PagedAttention / vLLM", SOSP 2023
  2. Dao et al., "Flash Attention", NeurIPS 2022
  3. Shah et al., "Flash Attention 3", 2024
  4. Gu & Dao, "Mamba", 2023
  5. Dao & Gu, "Mamba-2 / SSD", 2024
  6. DeepSeek-AI, "MLA / DeepSeek-V2", 2024
  7. DeepSeek-AI, "DeepSeek-V3 (MoE + FP8)", 2025
  8. AI21, "Jamba (Mamba-Transformer)", 2024
  9. TII, "Falcon Mamba", 2024
  10. Ma et al., "BitNet b1.58", 2024
  11. Frantar et al., "GPTQ", ICLR 2023
  12. Lin et al., "AWQ", 2024
  13. Xiao et al., "SmoothQuant", 2023
  14. Liu et al., "KIVI", 2024
  15. Zheng et al., "SGLang", 2024
  16. NVIDIA, "TensorRT-LLM", 2024
  17. Cai et al., "Medusa", 2024
  18. Li et al., "EAGLE / EAGLE-2", 2024
  19. Fu et al., "Lookahead Decoding", 2024
  20. Microsoft, "RetNet", 2023
  21. Peng et al., "RWKV", 2023
  22. Poli et al., "Hyena", 2023

Ch18-19 ML 系统与应用

  1. Sculley et al., "Hidden Technical Debt in ML Systems", NeurIPS 2015
  2. Polyzotis et al., "Data Management Challenges in Production ML", SIGMOD 2018
  3. Schelter et al., "On Challenges in ML Model Management", 2018
  4. Liu et al., "ML Production Systems", 2020
  5. Zaharia et al., "MLflow", 2018
  6. Biewald, "Weights & Biases", 2020
  7. Jumper et al., "AlphaFold 2", Nature 2021
  8. Abramson et al., "AlphaFold 3", Nature 2024
  9. Watson et al., "AlphaFold", Nature 2023
  10. Boiko et al., "Emergent Autonomous Scientific Research", 2023
  11. Romera-Paredes et al., "FunSearch", Nature 2023
  12. Anthropic, "Computer Use", 2024
  13. OpenAI, "Operator / Agent", 2024-2025
  14. Anthropic, "MCP (Model Context Protocol)", 2024
  15. Park et al., "Generative Agents", UIST 2023
  16. AutoGPT, "AutoGPT", 2023
  17. Schick et al., "Toolformer", NeurIPS 2023
  18. Yao et al., "ReAct", ICLR 2023
  19. Wei et al., "Chain-of-Thought (CoT)", NeurIPS 2022

Ch20 前沿

  1. Biamonte et al., "Quantum Machine Learning", Nature 2017
  2. Schuld & Petruccione, "Quantum ML", 2018
  3. Davies et al., "Loihi / Intel Neuromorphic", IEEE Micro 2018
  4. Pehle et al., "BrainScaleS-2", 2022
  5. Marković et al., "Photonic Computing", 2020
  6. Hinton et al., "Forward-Forward Algorithm", 2022
  7. Possible, "Open Source AGI", 2024
  8. Musk et al., "xAI", 2023-2025
  9. LeCun, "JEPA / World Models", 2024
  10. Hassabis et al., "AlphaProof", Nature 2024
  11. DeepMind, "AlphaGeometry", Nature 2024

通用前沿与综述

  1. OpenAI, "GPT-4 Technical Report", 2023
  2. Anthropic, "Claude 4 / 4.5 Opus", 2025
  3. OpenAI, "o1 / o3", 2024-2025
  4. Snell et al., "Scaling LLM Test-Time Compute", 2024
  5. Brown et al., "Large Language Monkeys", 2024
  6. Wei et al., "Emergent Abilities of LLMs", 2022
  7. Schaeffer et al., "Are Emergent Abilities of LLMs a Mirage?", NeurIPS 2023
  8. Bender et al., "On the Dangers of Stochastic Parrots", FAccT 2021
  9. Anthropic, "Sleeper Agents", 2024
  10. Hubinger et al., "Alignment Faking", 2024
  11. Liang et al., "Holistic Evaluation of LLMs", 2023
  12. Zheng et al., "Judging LLM-as-a-Judge", NeurIPS 2023
  13. Wang et al., "Self-Consistency", ICLR 2023
  14. Madaan et al., "Self-Refine", NeurIPS 2023
  15. Khattab et al., "DSPy", ICLR 2024

工具

  • PyTorch, TensorFlow
  • HuggingFace Transformers
  • LangChain, LlamaIndex
  • Weights & Biases, MLflow

在线课程

  • CS231n (Stanford)
  • CS224n (Stanford)
  • DeepMind x UCL
  • fast.ai