第五讲：完善图像部分内容与配图（按多轮评审意见）

- 结构：删去引言/本讲小结，仅保留 5.1/5.2/5.3 三部分（含学习目标）
- 语言：改为研究生教材口吻、软化绝对化表述、修正中文引号
- 5.1 表示学习：补充数据原始形式(像素/振幅/词表)、向量空间与相似度、卷积(公式)、token 概念，并强化文本表征(基准语义空间)
- 图示：新增/重画多张 TikZ 配图并对着图讲——三模态原始形式、向量空间、卷积、文本表示流程、三类融合(Emu/BLIP-2/LLaVA)、通用结构(含文本输入)、视觉语言模型及三种连接模块(线性投影/Q-Former/门控交叉注意力)、扩散去噪、图像编辑
- 代表工作与发展脉络梳理，参考文献改用 \cite/\bibitem 汇入全书参考资料，逐条核对并补齐(含 Emu 等)
- 逐图高分辨率核验无重叠；XeLaTeX 编译通过(28 页)
语音/音频小节保留 \todochen 占位待陈丹补充。

aml_notes.tex

@@ -26,6 +26,8 @@
 \usepackage{multirow}
 \usepackage{multirow}
 \usepackage{caption}
 \usepackage{caption}
 \usepackage{subcaption}
 \usepackage{subcaption}
+\usepackage{tikz}
+\usetikzlibrary{arrows.meta,positioning,fit,backgrounds,calc,shapes.geometric}
 \graphicspath{{figures/}{images/}}
 \graphicspath{{figures/}{images/}}
 
 
 % 颜色与强调框
 % 颜色与强调框
@@ -121,6 +123,15 @@
 \newcommand{\blankline}{\par\noindent\rule{\linewidth}{0.4pt}\par}
 \newcommand{\blankline}{\par\noindent\rule{\linewidth}{0.4pt}\par}
 \newcommand{\keyword}[1]{\textbf{\color{CourseBlue}#1}}
 \newcommand{\keyword}[1]{\textbf{\color{CourseBlue}#1}}
 
 
+% 直观理解框：用类比/比喻帮助学生建立直觉
+\definecolor{CoursePurple}{HTML}{6D28D9}
+\newtcolorbox{intuition}{
+  colback=CourseGray,
+  colframe=CoursePurple,
+  title=直观理解,
+  fonttitle=\bfseries
+}
+
 % 协作占位：标记"待陈丹补充（语音/音频部分）"的内容，便于后续合并时定位
 % 协作占位：标记"待陈丹补充（语音/音频部分）"的内容，便于后续合并时定位
 \newcommand{\todochen}[1]{%
 \newcommand{\todochen}[1]{%
   \par\noindent
   \par\noindent
@@ -323,12 +334,41 @@
 
 
 这里可以列出教材、论文、网站或其他阅读材料。
 这里可以列出教材、论文、网站或其他阅读材料。
 
 
-\begin{thebibliography}{9}
-  \bibitem{sample-book}
-  作者. \emph{书名}. 出版社, 年份.
-
-  \bibitem{sample-web}
-  网站名称. 文章标题. \url{https://example.com}
+\begin{thebibliography}{99}
+  % —— 第五部分 多模态机器学习方法与模型 ——
+  \bibitem{alexnet} Krizhevsky A, Sutskever I, Hinton G E. ImageNet Classification with Deep Convolutional Neural Networks. NeurIPS, 2012.
+  \bibitem{vgg} Simonyan K, Zisserman A. Very Deep Convolutional Networks for Large-Scale Image Recognition (VGG). ICLR, 2015.
+  \bibitem{resnet} He K, Zhang X, Ren S, Sun J. Deep Residual Learning for Image Recognition. CVPR, 2016.
+  \bibitem{vit} Dosovitskiy A, et al. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale. ICLR, 2021.
+  \bibitem{simclr} Chen T, et al. A Simple Framework for Contrastive Learning of Visual Representations (SimCLR). ICML, 2020.
+  \bibitem{moco} He K, et al. Momentum Contrast for Unsupervised Visual Representation Learning (MoCo). CVPR, 2020.
+  \bibitem{dino} Caron M, et al. Emerging Properties in Self-Supervised Vision Transformers (DINO). ICCV, 2021.
+  \bibitem{mae} He K, et al. Masked Autoencoders Are Scalable Vision Learners (MAE). CVPR, 2022.
+  \bibitem{beit} Bao H, Dong L, Piao S, Wei F. BEiT: BERT Pre-Training of Image Transformers. ICLR, 2022.
+  \bibitem{clip} Radford A, et al. Learning Transferable Visual Models From Natural Language Supervision (CLIP). ICML, 2021.
+  \bibitem{align} Jia C, et al. Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision (ALIGN). ICML, 2021.
+  \bibitem{siglip} Zhai X, et al. Sigmoid Loss for Language Image Pre-Training (SigLIP). ICCV, 2023.
+  \bibitem{flamingo} Alayrac J B, et al. Flamingo: a Visual Language Model for Few-Shot Learning. NeurIPS, 2022.
+  \bibitem{blip2} Li J, et al. BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models. ICML, 2023.
+  \bibitem{llava} Liu H, et al. Visual Instruction Tuning (LLaVA). NeurIPS, 2023.
+  \bibitem{instructblip} Dai W, et al. InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning. NeurIPS, 2023.
+  \bibitem{qwenvl} Bai J, et al. Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyond. arXiv:2308.12966, 2023.
+  \bibitem{emu} Sun Q, et al. Emu: Generative Pretraining in Multimodality. ICLR, 2024.
+  \bibitem{gan} Goodfellow I, et al. Generative Adversarial Nets. NeurIPS, 2014.
+  \bibitem{vae} Kingma D P, Welling M. Auto-Encoding Variational Bayes (VAE). ICLR, 2014.
+  \bibitem{ddpm} Ho J, Jain A, Abbeel P. Denoising Diffusion Probabilistic Models (DDPM). NeurIPS, 2020.
+  \bibitem{cfg} Ho J, Salimans T. Classifier-Free Diffusion Guidance. arXiv:2207.12598, 2022.
+  \bibitem{dalle} Ramesh A, et al. Zero-Shot Text-to-Image Generation (DALL·E). ICML, 2021.
+  \bibitem{glide} Nichol A, et al. GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models. ICML, 2022.
+  \bibitem{dalle2} Ramesh A, et al. Hierarchical Text-Conditional Image Generation with CLIP Latents (DALL·E 2). arXiv:2204.06125, 2022.
+  \bibitem{imagen} Saharia C, et al. Photorealistic Text-to-Image Diffusion Models with Deep Language Understanding (Imagen). NeurIPS, 2022.
+  \bibitem{ldm} Rombach R, et al. High-Resolution Image Synthesis with Latent Diffusion Models (Stable Diffusion). CVPR, 2022.
+  \bibitem{sdedit} Meng C, et al. SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations. ICLR, 2022.
+  \bibitem{p2p} Hertz A, et al. Prompt-to-Prompt Image Editing with Cross Attention Control. ICLR, 2023.
+  \bibitem{ip2p} Brooks T, Holynski A, Efros A A. InstructPix2Pix: Learning to Follow Image Editing Instructions. CVPR, 2023.
+  \bibitem{controlnet} Zhang L, Rao A, Agrawala M. Adding Conditional Control to Text-to-Image Diffusion Models (ControlNet). ICCV, 2023.
+  \bibitem{dreambooth} Ruiz N, et al. DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generation. CVPR, 2023.
+  \bibitem{codi} Tang Z, et al. Any-to-Any Generation via Composable Diffusion (CoDi). NeurIPS, 2023.
 \end{thebibliography}
 \end{thebibliography}
 
 
 \end{document}
 \end{document}