AI prompts
base on 深度学习经典、新论文逐段精读 # 深度学习论文精读
## 录制完成的论文
| 日期 | 标题 | 封面 | 时长 | 视频（播放数） |
| --: | -- | -- | --: | -- |
| 9/04/24 | Llama 3.1论文精读 · 5. 模型训练过程 | <img src="imgs/llama3-process.jpg" width="200px"/> | 10:41| [](https://www.bilibili.com/video/BV1c8HbeaEXi)<br /> |
| 8/28/24 | Llama 3.1论文精读 · 4. 训练infra | <img src="imgs/llama3-training-infra.webp" width="200px"/> | 25:04| [](https://www.bilibili.com/video/BV1b4421f7fa)<br />[](https://www.youtube.com/watch?v=6XidEHVjS1A) |
| 8/13/24 | Llama 3.1论文精读 · 3. 模型 | <img src="imgs/llama3-model.webp" width="200px"/> | 26:14| [](https://www.bilibili.com/video/BV1Q4421Z7Tj)<br />[](https://www.youtube.com/watch?v=G6gF-5g1Gg4) |
| 8/05/24 | [Llama 3.1论文精读 · 2. 预训练数据](https://arxiv.org/pdf/2407.21783) | <img src="imgs/llama3-pretrain-data.jpg" width="200px"/> | 23:37| [](https://www.bilibili.com/video/BV1u142187S5)[](https://www.youtube.com/watch?v=wXFr3zIE8FM)|
| 7/31/24 | Llama 3.1论文精读 · 1. 导言 | <img src="imgs/llama3-intro.jpg" width="200px"/> | 18:53| [](https://www.bilibili.com/video/BV1WM4m1y7Uh)<br />[](https://www.youtube.com/watch?v=-PztagF3wQE) |
| 3/30/23 | [GPT-4](https://openai.com/research/gpt-4) | <img src="imgs/gpt4.jpg" width="200px"/> | 1:20:38 | [](https://www.bilibili.com/video/BV1vM4y1U7b5)<br />[](https://youtu.be/K0SZ9mdygTw) |
| 3/23/23 | 大模型时代下做科研的四个思路 | <img src="imgs/limited-resources.jpg" width="200px"/> | 1:06:29 | [](https://www.bilibili.com/video/BV1oX4y1d7X6)<br />[](https://youtu.be/sh79Z8i15PI) |
| 3/10/23 | [Anthropic LLM](https://arxiv.org/pdf/2204.05862.pdf) | <img src="imgs/anthropic_lm.jpg" width="200px"/> | 1:01:51 | [](https://www.bilibili.com/video/BV1XY411B7nM)<br />[](https://youtu.be/iqX0pgNDon0) |
| 1/20/23 | [Helm](https://arxiv.org/pdf/2211.09110.pdf) 全面语言模型评测 | <img src="imgs/helm.jpg" width="200px"/> | 1:23:37 | [](https://www.bilibili.com/video/BV1z24y1B7uX)<br />[](https://youtu.be/WgFEw9U3BXA) |
| 1/11/23 | 多模态论文串讲·下 | <img src="imgs/multimodal-2.jpg" width="200px"/> | 1:03:29 | [](https://www.bilibili.com/video/BV1fA411Z772) <br />[](https://youtu.be/S1le41J76lQ) |
| 12/29/22 | [Instruct GPT](https://arxiv.org/pdf/2203.02155.pdf) | <img src="imgs/instruct-gpt.jpg" width="200px"/> | 1:07:10 | [](https://www.bilibili.com/video/BV1hd4y187CR) <br />[](https://youtu.be/zfIGAwD1jOQ) |
| 12/19/22 | [Neural Corpus Indexer](https://arxiv.org/pdf/2206.02743.pdf) 文档检索 | <img src="imgs/nci.jpg" width="200px"/> | 55:47 | [](https://www.bilibili.com/video/BV1Se411w7Sn) <br />[](https://youtu.be/QRffZMSGJyU) |
| 12/12/22 | 多模态论文串讲·上 | <img src="imgs/multimodal-1.jpg" width="200px"/> | 1:12:27 | [](https://www.bilibili.com/video/BV1Vd4y1v77v) <br />[](https://youtu.be/6pzBOQAXUB8) |
| 11/14/22 | [OpenAI Whisper](https://cdn.openai.com/papers/whisper.pdf) 精读 | <img src="imgs/whisper.jpg" width="200px"/> | 1:12:16 | [](https://www.bilibili.com/video/BV1VG4y1t74x) <br />[](https://youtu.be/3eXCJd32UnM) |
| 11/07/22 | 在讲 OpenAI Whisper 前先做了一个剪视频小工具 | <img src="imgs/autocut.jpg" width="200px"/> | 23:39 | [](https://www.bilibili.com/video/BV1Pe4y1t7de) <br />[](https://youtu.be/PwVlvCPDnrI) |
| 10/23/22 | [Chain of Thought](https://arxiv.org/pdf/2201.11903.pdf) 论文、代码和资源 | <img src="imgs/cot.jpg" width="200px"/> | 33:21 | [](https://www.bilibili.com/video/BV1t8411e7Ug)<br />[](https://youtu.be/H4J59iG3t5o) |
| 9/17/22 | CLIP 改进工作串讲（下） | <img src="imgs/clipx-part2.jpg" width="200px"/> | 1:04:26 | [](https://www.bilibili.com/video/BV1gg411U7n4)<br />[](https://youtu.be/ugJeBivv65s) |
| 9/2/22 | CLIP 改进工作串讲（上） | <img src="imgs/clipx-part1.jpg" width="200px"/> | 1:14:43 | [](https://www.bilibili.com/video/BV1FV4y1p7Lm)<br />[](https://youtu.be/x4CDhZz_Dvg) |
| 7/29/22 | [ViLT](https://arxiv.org/pdf/2102.03334.pdf) 论文精读 | <img src="imgs/vilt.jpg" width="200px"/> | 1:03:26 | [](https://www.bilibili.com/video/BV14r4y1j74y)<br />[](https://youtu.be/ug8YvZOjOCE) |
| 7/22/22 | 理由、论据和担保【[研究的艺术](https://press.uchicago.edu/ucp/books/book/chicago/C/bo23521678.html)·四】 | <img src="imgs/craft_research_p4.jpg" width="200px"/> | 44:14 | [](https://www.bilibili.com/video/BV1SB4y1a75c) |
| 7/15/22 | 如何讲好故事、故事里的论点【[研究的艺术](https://press.uchicago.edu/ucp/books/book/chicago/C/bo23521678.html)·三】| <img src="imgs/craft_research_p3.jpg" width="200px"/> | 43:56 |[](https://www.bilibili.com/video/BV1WB4y1v7ST)|
| 7/8/22 | [DALL·E 2](https://arxiv.org/pdf/2204.06125.pdf) 逐段精读 | <img src="imgs/dalle2.jpg" width="200px"/> | 1:27:54 |[](https://www.bilibili.com/video/BV17r4y1u77B)<br />[](https://youtu.be/hO57mntSMl0)|
| 7/1/22 | 明白问题的重要性【[研究的艺术](https://press.uchicago.edu/ucp/books/book/chicago/C/bo23521678.html)·二】| <img src="imgs/craft_research_p2.jpg" width="200px"/> | 1:03:40 |[](https://www.bilibili.com/video/BV11S4y1v7S2/)|
| 6/24/22 | 跟读者建立联系【[研究的艺术](https://press.uchicago.edu/ucp/books/book/chicago/C/bo23521678.html)·一】 | <img src="imgs/craft_research_p1.jpg" width="200px"/> | 45:01 | [](https://www.bilibili.com/video/BV1hY411T7vy/) |
| 6/17/22 | [Zero](https://arxiv.org/pdf/1910.02054.pdf) 逐段精读 | <img src="imgs/zero.jpg" width="200px"/> | 52:21 | [](https://www.bilibili.com/video/BV1tY411g7ZT/) |
| 6/10/22 | [DETR](https://arxiv.org/pdf/2005.12872.pdf) 逐段精读 | <img src="imgs/detr.jpg" width="200px"/> | 54:22 | [](https://www.bilibili.com/video/BV1GB4y1X72R/) |
| 6/3/22 | [Megatron LM](https://arxiv.org/pdf/1909.08053.pdf) 逐段精读 | <img src="imgs/megatron_lm.jpg" width="200px"/> | 56:07 | [](https://www.bilibili.com/video/BV1nB4y1R7Yz/) |
| 5/27/22 | [GPipe](https://proceedings.neurips.cc/paper/2019/file/093f65e080a295f8076b1c5722a46aa2-Paper.pdf) 逐段精读 | <img src="imgs/gpipe.jpg" width="200px"/> | 58:47 | [](https://www.bilibili.com/video/BV1v34y1E7zu/) <br />[](https://youtu.be/eXjRpS_BTbs) |
| 5/5/22 | [Pathways](https://arxiv.org/pdf/2203.12533.pdf) 逐段精读 | <img src="imgs/pathways.jpg" width="200px"/> | 1:02:13 | [](https://www.bilibili.com/video/BV1xB4y1m7Xi/) <br />[](https://youtu.be/8hS1ZtgG0wU) |
| 4/28/22 | [视频理解论文串讲](https://arxiv.org/pdf/2012.06567.pdf)（下） | <img src="imgs/video-survey-p2.jpg" width="200px"/> | 1:08:32 | [](https://www.bilibili.com/video/BV11Y411P7ep/) <br />[](https://youtu.be/J2YC0-k57NM) |
| 4/21/22 | [参数服务器（Parameter Server）](https://www.usenix.org/system/files/conference/osdi14/osdi14-paper-li_mu.pdf) 逐段精读 | <img src="imgs/ps.jpg" width="200px"/> | 1:37:40 | [](https://www.bilibili.com/video/BV1YA4y197G8/) <br />[](https://youtu.be/xt-AwUrDxQk) |
| 4/14/22 | [视频理解论文串讲](https://arxiv.org/pdf/2012.06567.pdf)（上） | <img src="imgs/video-survey-p1.jpg" width="200px"/> | 51:15 | [](https://www.bilibili.com/video/BV1fL4y157yA/) <br />[](https://youtu.be/gK7AGO6okhc) |
| 3/31/22 | [I3D](https://arxiv.org/pdf/1705.07750.pdf) 论文精读 | <img src="imgs/i3d.jpg" width="200px"/> | 52:31 | [](https://www.bilibili.com/video/BV1tY4y1p7hq/) <br />[](https://youtu.be/9lIkKiAn6uE) |
| 3/24/22 | 斯坦福 2022 年 [AI 指数报告](https://aiindex.stanford.edu/wp-content/uploads/2022/03/2022-AI-Index-Report_Master.pdf) 精读 | <img src="imgs/ai_index_22.jpg" width="200px"/> | 1:19:56 | [](https://www.bilibili.com/video/BV1s44y1N7eu/) <br />[](https://youtu.be/K8h_xjQ6ufY) |
| 3/17/22 | [AlphaCode](https://storage.googleapis.com/deepmind-media/AlphaCode/competition_level_code_generation_with_alphacode.pdf) 论文精读 | <img src="imgs/alphacode.jpg" width="200px"/> | 44:00 | [](https://www.bilibili.com/video/BV1ab4y1s7rc/) <br />[](https://youtu.be/t8Gzkca9pW4) |
| 3/10/22 | [OpenAI Codex](https://arxiv.org/pdf/2107.03374.pdf) 论文精读 | <img src="imgs/codex.jpg" width="200px"/> | 47:58 | [](https://www.bilibili.com/video/BV1iY41137Zi/) <br />[](https://www.zhihu.com/zvideo/1490959755963666432)<br />[](https://youtu.be/oZriUGkQSNM) |
| 3/3/22 | [GPT](https://s3-us-west-2.amazonaws.com/openai-assets/research-covers/language-unsupervised/language_understanding_paper.pdf), [GPT-2](https://d4mucfpksywv.cloudfront.net/better-language-models/language_models_are_unsupervised_multitask_learners.pdf), [GPT-3](https://arxiv.org/abs/2005.14165) 精读 | <img src="imgs/gpt3.jpg" width="200px"/> | 1:29:58 | [](https://www.bilibili.com/video/BV1AF411b7xQ/)<br />[](https://youtu.be/t70Bl3w7bxY) |
| 2/24/22 | [Two-Stream](https://proceedings.neurips.cc/paper/2014/file/00ec53c4682d36f5c4359f4ae7bd7ba1-Paper.pdf) 逐段精读 | <img src="imgs/twostream.jpg" width="200px"/> | 52:57 | [](https://www.bilibili.com/video/BV1mq4y1x7RU/)<br />[](https://youtu.be/vuqwKP2iDe0) |
| 2/10/22 | [CLIP](https://openai.com/blog/clip/) 逐段精读 | <img src="imgs/clip.jpg" width="200px"/> | 1:38:25 | [](https://www.bilibili.com/video/BV1SL4y1s7LQ/)<br />[](https://www.zhihu.com/zvideo/1475706654562299904) <br />[](https://youtu.be/OZF1t_Hieq8) |
| 2/6/22 | 你（被）吐槽过[论文不够 novel](https://perceiving-systems.blog/en/post/novelty-in-science) 吗？| <img src="imgs/novelty.jpg" width="200px"/> | 14:11 | [](https://www.bilibili.com/video/BV1ea41127Bq/) <br />[](https://www.zhihu.com/zvideo/1475719090198876161) |
| 1/23/22 | [AlphaFold 2](https://www.nature.com/articles/s41586-021-03819-2.pdf) 精读 | <img src="imgs/alphafold_2.jpg" width="200px"/> | 1:15:28 | [](https://www.bilibili.com/video/BV1oR4y1K7Xr/) <br />[](https://www.zhihu.com/zvideo/1469132410537717760) <br />[](https://youtu.be/Oy3OCoGUr-w) |
| 1/18/22 | 如何判断（你自己的）研究工作的价值 | <img src="imgs/research_value.jpg" width="200px"/> | 9:59 | [](https://www.bilibili.com/video/BV1oL411c7Us/) <br />[](https://www.zhihu.com/zvideo/1475716940051869696) |
| 1/15/22 | [Swin Transformer](https://arxiv.org/pdf/2103.14030.pdf) 精读 | <img src="imgs/swin_transformer.jpg" width="200px"/> | 1:00:21 | [](https://www.bilibili.com/video/BV13L4y1475U/) <br />[](https://www.zhihu.com/zvideo/1466282983652691968) <br />[](https://youtu.be/luP3-Fs0QCo) |
| 1/7/22 | [指导数学直觉](https://www.nature.com/articles/s41586-021-04086-x.pdf) | <img src="imgs/math_conj.jpg" width="200px"/> | 52:51 | [](https://www.bilibili.com/video/BV1YZ4y1S72j/) <br />[](https://www.zhihu.com/zvideo/1464060386375299072) <br />[](https://youtu.be/czFGjvhtss8) |
| 1/5/22 | AlphaFold 2 预告 | <img src="imgs/alphafold_2_preview.jpg" width="200px"/> | 03:28 | [](https://www.bilibili.com/video/BV1Eu411U7Te/) |
| 12/20/21 | [对比学习](#contrastive_learning)论文综述 | <img src="imgs/contrastive.jpg" width="200px"/> | 1:32:01 | [](https://www.bilibili.com/video/BV19S4y1M7hm/) <br />[](https://www.zhihu.com/zvideo/1460828005077164032) <br />[](https://www.youtube.com/watch?v=1pvxufGRuW4) |
| 12/15/21 | [MoCo](https://arxiv.org/pdf/1911.05722.pdf) 逐段精读 | <img src="imgs/mocov1.jpg" width="200px"/> | 1:24:11 | [](https://www.bilibili.com/video/BV1C3411s7t9/) <br />[](https://www.zhihu.com/zvideo/1454723120678936576) <br />[](https://www.youtube.com/watch?v=1pvxufGRuW4) |
| 12/9/21 | 如何找研究想法 1 | <img src="imgs/mae_idea.jpg" width="200px"/> | 5:34 | [](https://www.bilibili.com/video/BV1qq4y1z7F2/) |
| 12/8/21 | [MAE](https://arxiv.org/pdf/2111.06377.pdf) 逐段精读 | <img src="imgs/mae.jpg" width="200px"/> | 47:04 | [](https://www.bilibili.com/video/BV1sq4y1q77t/) <br />[](https://www.zhihu.com/zvideo/1452458167968251904) <br />[](https://youtu.be/mYlX2dpdHHM) |
| 11/29/21 | [ViT](https://arxiv.org/pdf/2010.11929.pdf) 逐段精读 | <img src="imgs/vit.jpg" width="200px"/> | 1:11:30 | [](https://www.bilibili.com/video/BV15P4y137jb/) <br />[](https://www.zhihu.com/zvideo/1449195245754380288) <br />[](https://youtu.be/FRFt3x0bO94) |
| 11/18/21 | [BERT](https://arxiv.org/abs/1810.04805) 逐段精读 | <img src="imgs/bert.jpg" width="200px"/> | 45:49 | [](https://www.bilibili.com/video/BV1PL411M7eQ/) <br />[](https://www.zhihu.com/zvideo/1445340200976785408) <br />[](https://youtu.be/ULD3uIb2MHQ) |
| 11/9/21 | [GAN](https://papers.nips.cc/paper/2014/file/5ca3e9b122f61f8f06494c97b1afccf3-Paper.pdf) 逐段精读 | <img src="imgs/gan.jpg" width="200px"/> | 46:16 | [](https://www.bilibili.com/video/BV1rb4y187vD/) <br />[](https://www.zhihu.com/zvideo/1442091389241159681) <br />[](https://www.youtube.com/watch?v=g_0HtlrLiDo) |
| 11/3/21 | 零基础多图详解 [图神经网络](https://distill.pub/2021/gnn-intro/)（GNN/GCN） | <img src="imgs/gnn.jpg" width="200px"/> | 1:06:19 | [](https://www.bilibili.com/video/BV1iT4y1d7zP/) <br />[](https://www.zhihu.com/zvideo/1439540657619087360) <br />[](https://youtu.be/sejA2PtCITw) |
| 10/27/21 | [Transformer](https://arxiv.org/abs/1706.03762) 逐段精读<br> （视频中提到的文献 [^transformer]) |<img src="imgs/transformer.jpg" width="200px"/> | 1:27:05 | [](https://www.bilibili.com/video/BV1pu411o7BE/) <br />[](https://www.zhihu.com/zvideo/1437034536677404672) <br />[](https://youtu.be/nzqlFIcCSWQ) |
| 10/22/21 | [ResNet](https://arxiv.org/abs/1512.03385) 论文逐段精读 | <img src="imgs/resnet-2.jpg" width="200px"/> | 53:46 | [](https://www.bilibili.com/video/BV1P3411y7nn/) <br />[](https://www.zhihu.com/zvideo/1434795406001180672) <br />[](https://www.youtube.com/watch?v=pWMnzCX4cwQ) |
| 10/21/21 | 撑起计算机视觉半边天的 [ResNet](https://arxiv.org/abs/1512.03385) | <img src="imgs/resnet-1.jpg" width="200px"/> | 11:50 | [](https://www.bilibili.com/video/BV1Fb4y1h73E/) <br />[](https://www.zhihu.com/zvideo/1434787226101751808) <br />[](https://www.youtube.com/watch?v=NnSldWhSqvY) |
| 10/15/21 | [AlexNet](https://papers.nips.cc/paper/2012/file/c399862d3b9d6b76c8436e924a68c45b-Paper.pdf) 论文逐段精读 | <img src="imgs/alexnet-2.jpg" width="200px"/> | 55:21 | [](https://www.bilibili.com/video/BV1hq4y157t1/) <br />[](https://www.zhihu.com/zvideo/1432354207483871232) <br />[](https://www.youtube.com/watch?v=wYmlILPsLlY) |
| 10/14/21 | 9年后重读深度学习奠基作之一：[AlexNet](https://papers.nips.cc/paper/2012/file/c399862d3b9d6b76c8436e924a68c45b-Paper.pdf) | <img src="imgs/alexnet-1.jpg" width="200px"/> | 19:59 | [](https://www.bilibili.com/video/BV1ih411J7Kz/) <br />[](https://www.zhihu.com/zvideo/1432155856322920448) <br />[](https://www.youtube.com/watch?v=vdYH0fE6thY) |
| 10/06/21 | 如何读论文 | <img src="imgs/read-paper.jpg" width="200px"/> | 06:39 | [](https://www.bilibili.com/video/BV1H44y1t75x/) <br />[](https://www.zhihu.com/zvideo/1428973951632969728) <br />[](https://www.youtube.com/watch?v=txjl_Q4jCyQ&list=PLFXJ6jwg0qW-7UM8iUTj3qKqdhbQULP5I&index=1) |
[^transformer]: 1 [斯坦福100+作者的200+页综述](https://arxiv.org/abs/2108.07258)，2 [对LayerNorm的新研究](https://arxiv.org/pdf/1911.07013.pdf)，3 [对Attention在Transformer里面作用的研究](https://arxiv.org/abs/2103.03404)
## 所有论文
包括已经录制完成和之后将要介绍的论文。选取的原则是10年内深度学习里有影响力文章（必读文章），或者近期比较有意思的文章。当然这十年里重要的工作太多了，不可能一一过一遍。在选取的时候我会偏向一些之前 [直播课](https://c.d2l.ai/zh-v2/) 中没讲到过的。 欢迎大家在 [讨论区](https://github.com/mli/paper-reading/discussions) 里提供建（点）议（歌）。
总论文数 67，录制完成数 32
（这里引用采用的是 semanticscholar，是因为它提供 [API](https://api.semanticscholar.org/api-docs/graph#operation/get_graph_get_paper) 可以自动获取，不用手动更新。）
### 计算机视觉 - CNN
| 已录制 | 年份 | 名字 | 简介 | 引用 |
| ------ | ---- | ------------------------------------------------------------ | -------------------- | ------------------------------------------------------------ |
| ✅ | 2012 | [AlexNet](https://papers.nips.cc/paper/2012/file/c399862d3b9d6b76c8436e924a68c45b-Paper.pdf) | 深度学习热潮的奠基作 | [](https://www.semanticscholar.org/paper/ImageNet-classification-with-deep-convolutional-Krizhevsky-Sutskever/abd1c342495432171beb7ca8fd9551ef13cbd0ff) |
| | 2014 | [VGG](https://arxiv.org/pdf/1409.1556.pdf) | 使用 3x3 卷积构造更深的网络 | [](https://www.semanticscholar.org/paper/Very-Deep-Convolutional-Networks-for-Large-Scale-Simonyan-Zisserman/eb42cf88027de515750f230b23b1a057dc782108) |
| | 2014 | [GoogleNet](https://arxiv.org/pdf/1409.4842.pdf) | 使用并行架构构造更深的网络 | [](https://www.semanticscholar.org/paper/Going-deeper-with-convolutions-Szegedy-Liu/e15cf50aa89fee8535703b9f9512fca5bfc43327) |
| ✅ | 2015 | [ResNet](https://arxiv.org/pdf/1512.03385.pdf) | 构建深层网络都要有的残差连接。 |[](https://www.semanticscholar.org/paper/Deep-Residual-Learning-for-Image-Recognition-He-Zhang/2c03df8b48bf3fa39054345bafabfeff15bfd11d) |
| | 2017 | [MobileNet](https://arxiv.org/pdf/1704.04861.pdf) | 适合终端设备的小CNN |[](https://www.semanticscholar.org/paper/MobileNets%3A-Efficient-Convolutional-Neural-Networks-Howard-Zhu/3647d6d0f151dc05626449ee09cc7bce55be497e) |
| | 2019 | [EfficientNet](https://arxiv.org/pdf/1905.11946.pdf) | 通过架构搜索得到的CNN |[](https://www.semanticscholar.org/paper/EfficientNet%3A-Rethinking-Model-Scaling-for-Neural-Tan-Le/4f2eda8077dc7a69bb2b4e0a1a086cf054adb3f9) |
| | 2021 | [Non-deep networks](https://arxiv.org/pdf/2110.07641.pdf) | 让不深的网络也能在ImageNet刷到SOTA | [](https://www.semanticscholar.org/paper/Non-deep-Networks-Goyal-Bochkovskiy/0d7f6086772079bc3e243b7b375a9ca1a517ba8b) |
### 计算机视觉 - Transformer
| 已录制 | 年份 | 名字 | 简介 | 引用 |
| ------ | ---- | ------------------------------------------------------------ | -------------------- | ------------------------------------------------------------ |
| ✅ | 2020 | [ViT](https://arxiv.org/pdf/2010.11929.pdf) | Transformer杀入CV界 |[](https://www.semanticscholar.org/paper/An-Image-is-Worth-16x16-Words%3A-Transformers-for-at-Dosovitskiy-Beyer/7b15fa1b8d413fbe14ef7a97f651f47f5aff3903) |
| ✅ | 2021 | [Swin Transformer](https://arxiv.org/pdf/2103.14030.pdf) | 多层次的Vision Transformer | [](https://www.semanticscholar.org/paper/Swin-Transformer%3A-Hierarchical-Vision-Transformer-Liu-Lin/c8b25fab5608c3e033d34b4483ec47e68ba109b7) |
| | 2021 | [MLP-Mixer](https://arxiv.org/pdf/2105.01601.pdf) | 使用MLP替换self-attention |[](https://www.semanticscholar.org/paper/MLP-Mixer%3A-An-all-MLP-Architecture-for-Vision-Tolstikhin-Houlsby/2def61f556f9a5576ace08911496b7c7e4f970a4) |
| ✅ | 2021 | [MAE](https://arxiv.org/pdf/2111.06377.pdf) | BERT的CV版 |[](https://www.semanticscholar.org/paper/Masked-Autoencoders-Are-Scalable-Vision-Learners-He-Chen/c1962a8cf364595ed2838a097e9aa7cd159d3118) |
### 生成模型
| 已录制 | 年份 | 名字 | 简介 | 引用 |
| ------ | ---- | ------------------------------------------------- | ------------ | ------------------------------------------------------------ |
| ✅ | 2014 | [GAN](https://papers.nips.cc/paper/2014/file/5ca3e9b122f61f8f06494c97b1afccf3-Paper.pdf) | 生成模型的开创工作 |[](https://www.semanticscholar.org/paper/Generative-Adversarial-Nets-Goodfellow-Pouget-Abadie/54e325aee6b2d476bbbb88615ac15e251c6e8214) |
| | 2015 | [DCGAN](https://arxiv.org/pdf/1511.06434.pdf) | 使用CNN的GAN |[](https://www.semanticscholar.org/paper/Unsupervised-Representation-Learning-with-Deep-Radford-Metz/8388f1be26329fa45e5807e968a641ce170ea078) |
| | 2016 | [pix2pix](https://arxiv.org/pdf/1611.07004.pdf) | |[](https://www.semanticscholar.org/paper/Image-to-Image-Translation-with-Conditional-Isola-Zhu/8acbe90d5b852dadea7810345451a99608ee54c7) |
| | 2016 | [SRGAN](https://arxiv.org/pdf/1609.04802.pdf) | 图片超分辨率 |[](https://www.semanticscholar.org/paper/Photo-Realistic-Single-Image-Super-Resolution-Using-Ledig-Theis/df0c54fe61f0ffb9f0e36a17c2038d9a1964cba3) |
| | 2017 | [WGAN](https://arxiv.org/abs/1701.07875) | 训练更加容易 |[](https://www.semanticscholar.org/paper/Wasserstein-GAN-Arjovsky-Chintala/2f85b7376769473d2bed56f855f115e23d727094) |
| | 2017 | [CycleGAN](https://arxiv.org/abs/1703.10593) | |[](https://www.semanticscholar.org/paper/Unpaired-Image-to-Image-Translation-Using-Networks-Zhu-Park/c43d954cf8133e6254499f3d68e45218067e4941) |
| | 2018 | [StyleGAN](https://arxiv.org/abs/1812.04948) | |[](https://www.semanticscholar.org/paper/A-Style-Based-Generator-Architecture-for-Generative-Karras-Laine/ceb2ebef0b41e31c1a21b28c2734123900c005e2) |
| | 2019 | [StyleGAN2](https://arxiv.org/pdf/1912.04958.pdf) | |[](https://www.semanticscholar.org/paper/Analyzing-and-Improving-the-Image-Quality-of-Karras-Laine/f3e3d1f86a534a3654d0ee263142e44f4e2c61e9) |
| | 2020 | [DDPM](https://arxiv.org/pdf/2006.11239.pdf) | Diffusion Models |[](https://www.semanticscholar.org/paper/Denoising-Diffusion-Probabilistic-Models-Ho-Jain/289db3be7bf77e06e75541ba93269de3d604ac72) |
| | 2021 | [Improved DDPM](https://arxiv.org/pdf/2102.09672.pdf) | 改进的 DDPM |[](https://www.semanticscholar.org/paper/Improved-Denoising-Diffusion-Probabilistic-Models-Nichol-Dhariwal/de18baa4964804cf471d85a5a090498242d2e79f) |
| | 2021 | [Guided Diffusion Models](https://arxiv.org/pdf/2105.05233.pdf) | 号称超越 GAN |[](https://www.semanticscholar.org/paper/Diffusion-Models-Beat-GANs-on-Image-Synthesis-Dhariwal-Nichol/64ea8f180d0682e6c18d1eb688afdb2027c02794) |
| | 2021 | [StyleGAN3](https://arxiv.org/pdf/2106.12423.pdf) | |[](https://www.semanticscholar.org/paper/Alias-Free-Generative-Adversarial-Networks-Karras-Aittala/c1ff08b59f00c44f34dfdde55cd53370733a2c19) |
| ✅ | 2022 | [DALL.E 2](https://arxiv.org/pdf/2204.06125.pdf) | CLIP + Diffusion models，文本生成图像新高度 |[](https://www.semanticscholar.org/paper/Hierarchical-Text-Conditional-Image-Generation-with-Ramesh-Dhariwal/c57293882b2561e1ba03017902df9fc2f289dea2) |
### 计算机视觉 - Object Detection
| 已录制 | 年份 | 名字 | 简介 | 引用 |
| ------ | ---- | ------------------------------------------------- | ------------ | ------------------------------------------------------------ |
| | 2014 | [R-CNN](https://arxiv.org/pdf/1311.2524v5.pdf) | Two-stage |[](https://www.semanticscholar.org/paper/2f4df08d9072fc2ac181b7fced6a245315ce05c8) |
| | 2015 | [Fast R-CNN](http://arxiv.org/abs/1504.08083v2) | |[](https://www.semanticscholar.org/paper/7ffdbc358b63378f07311e883dddacc9faeeaf4b) |
| | 2015 | [Faster R-CNN](http://arxiv.org/abs/1506.01497v3) | |[](https://www.semanticscholar.org/paper/424561d8585ff8ebce7d5d07de8dbf7aae5e7270) |
| | 2016 | [SSD](http://arxiv.org/abs/1512.02325v5) | Single stage |[](https://www.semanticscholar.org/paper/4d7a9197433acbfb24ef0e9d0f33ed1699e4a5b0) |
| | 2016 | [YOLO](http://arxiv.org/abs/1506.02640v5) | |[](https://www.semanticscholar.org/paper/f8e79ac0ea341056ef20f2616628b3e964764cfd) |
| | 2017 | [Mask R-CNN](http://arxiv.org/abs/1703.06870v3) | |[](https://www.semanticscholar.org/paper/ea99a5535388196d0d44be5b4d7dd02029a43bb2) |
| | 2017 | [YOLOv2](http://arxiv.org/abs/1612.08242v1) | |[](https://www.semanticscholar.org/paper/7d39d69b23424446f0400ef603b2e3e22d0309d6) |
| | 2018 | [YOLOv3](http://arxiv.org/abs/1804.02767v1) | |[](https://www.semanticscholar.org/paper/e4845fb1e624965d4f036d7fd32e8dcdd2408148) |
| | 2019 | [CenterNet](https://arxiv.org/pdf/1904.07850.pdf) | Anchor free |[](https://www.semanticscholar.org/paper/Objects-as-Points-Zhou-Wang/6a2e2fd1b5bb11224daef98b3fb6d029f68a73f2) |
| ✅ | 2020 | [DETR](https://arxiv.org/pdf/2005.12872.pdf) | Transformer |[](https://www.semanticscholar.org/paper/End-to-End-Object-Detection-with-Transformers-Carion-Massa/962dc29fdc3fbdc5930a10aba114050b82fe5a3e) |
<a name="contrastive_learning"></a>
### 计算机视觉 - 对比学习
| 已录制 | 年份 | 名字 | 简介 | 引用 |
| ------ | ---- | ------------------------------------------------------------ | -------------------- | ------------------------------------------------------------ |
| ✅ | 2018 | [InstDisc](https://arxiv.org/pdf/1805.01978.pdf) | 提出实例判别和memory bank做对比学习 |[](https://www.semanticscholar.org/paper/Unsupervised-Feature-Learning-via-Non-parametric-Wu-Xiong/155b7782dbd713982a4133df3aee7adfd0b6b304) |
| ✅ | 2018 | [CPC](https://arxiv.org/pdf/1807.03748.pdf) | 对比预测编码，图像语音文本强化学习全都能做 | [](https://www.semanticscholar.org/paper/Representation-Learning-with-Contrastive-Predictive-Oord-Li/b227f3e4c0dc96e5ac5426b85485a70f2175a205) |
| ✅ | 2019 | [InvaSpread](https://arxiv.org/pdf/1904.03436.pdf) | 一个编码器的端到端对比学习 |[](https://www.semanticscholar.org/paper/Unsupervised-Embedding-Learning-via-Invariant-and-Ye-Zhang/e4bde6fe33b6c2cf9d1647ac0b041f7d1ba29c5b) |
| ✅ | 2019 | [CMC](https://arxiv.org/pdf/1906.05849.pdf) | 多视角下的对比学习 |[](https://www.semanticscholar.org/paper/Contrastive-Multiview-Coding-Tian-Krishnan/97f4d09175705be4677d675fa27e55defac44800) |
| ✅ | 2019 | [MoCov1](https://arxiv.org/pdf/1911.05722.pdf) | 无监督训练效果也很好 | [](https://www.semanticscholar.org/paper/Momentum-Contrast-for-Unsupervised-Visual-Learning-He-Fan/ec46830a4b275fd01d4de82bffcabe6da086128f) |
| ✅ | 2020 | [SimCLRv1](https://arxiv.org/pdf/2002.05709.pdf) | 简单的对比学习 (数据增强 + MLP head + 大batch训练久) |[](https://www.semanticscholar.org/paper/A-Simple-Framework-for-Contrastive-Learning-of-Chen-Kornblith/34733eaf66007516347a40ad5d9bbe1cc9dacb6b) |
| ✅ | 2020 | [MoCov2](https://arxiv.org/pdf/2003.04297.pdf) | MoCov1 + improvements from SimCLRv1 |[](https://www.semanticscholar.org/paper/Improved-Baselines-with-Momentum-Contrastive-Chen-Fan/a1b8a8df281bbaec148a897927a49ea47ea31515) |
| ✅ | 2020 | [SimCLRv2](https://arxiv.org/pdf/2006.10029.pdf) | 大的自监督预训练模型很适合做半监督学习 |[](https://www.semanticscholar.org/paper/Big-Self-Supervised-Models-are-Strong-Learners-Chen-Kornblith/3e7f5f4382ac6f9c4fef6197dd21abf74456acd1) |
| ✅ | 2020 | [BYOL](https://arxiv.org/pdf/2006.07733.pdf) | 不需要负样本的对比学习 | [](https://www.semanticscholar.org/paper/Bootstrap-Your-Own-Latent%3A-A-New-Approach-to-Grill-Strub/38f93092ece8eee9771e61c1edaf11b1293cae1b) |
| ✅ | 2020 | [SWaV](https://arxiv.org/pdf/2006.09882.pdf) | 聚类对比学习 | [](https://www.semanticscholar.org/paper/Unsupervised-Learning-of-Visual-Features-by-Cluster-Caron-Misra/10161d83d29fc968c4612c9e9e2b61a2fc25842e) |
| ✅ | 2020 | [SimSiam](https://arxiv.org/pdf/2011.10566.pdf) | 化繁为简的孪生表征学习 |[](https://www.semanticscholar.org/paper/Exploring-Simple-Siamese-Representation-Learning-Chen-He/0e23d2f14e7e56e81538f4a63e11689d8ac1eb9d) |
| ✅ | 2021 | [MoCov3](https://arxiv.org/pdf/2104.02057.pdf) | 如何更稳定的自监督训练ViT |[](https://www.semanticscholar.org/paper/An-Empirical-Study-of-Training-Self-Supervised-Chen-Xie/739ceacfafb1c4eaa17509351b647c773270b3ae) |
| ✅ | 2021 | [DINO](https://arxiv.org/pdf/2104.14294.pdf) | transformer加自监督在视觉也很香 |[](https://www.semanticscholar.org/paper/Emerging-Properties-in-Self-Supervised-Vision-Caron-Touvron/ad4a0938c48e61b7827869e4ac3baffd0aefab35) |
### 计算机视觉 - 视频理解
| 已录制 | 年份 | 名字 | 简介 | 引用 |
| ------ | ---- | ------------------------------------------------------------ | -------------------- | ------------------------------------------------------------ |
| ✅ | 2014 | [DeepVideo](https://cs.stanford.edu/people/karpathy/deepvideo/) | 提出sports1M数据集，用深度学习做视频理解 |[](https://www.semanticscholar.org/paper/Large-Scale-Video-Classification-with-Convolutional-Karpathy-Toderici/6d4c9c923e9f145d1c01a2de2afc38ec23c44253) |
| ✅ | 2014 | [Two-stream](https://arxiv.org/pdf/1406.2199.pdf) | 引入光流做时序建模，神经网络首次超越手工特征 |[](https://www.semanticscholar.org/paper/Two-Stream-Convolutional-Networks-for-Action-in-Simonyan-Zisserman/67dccc9a856b60bdc4d058d83657a089b8ad4486) |
| ✅ | 2014 | [C3D](https://arxiv.org/pdf/1412.0767.pdf) | 比较深的3D-CNN做视频理解 |[](https://www.semanticscholar.org/paper/Learning-Spatiotemporal-Features-with-3D-Networks-Tran-Bourdev/d25c65d261ea0e6a458be4c50c40ffe5bc508f77) |
| ✅ | 2015 | [Beyond-short-snippets](https://arxiv.org/pdf/1503.08909.pdf) | 尝试使用LSTM |[](https://www.semanticscholar.org/paper/Beyond-short-snippets%3A-Deep-networks-for-video-Ng-Hausknecht/5418b2a482720e013d487a385c26fae0f017c6a6) |
| ✅ | 2016 | [Convolutional fusion](https://arxiv.org/pdf/1604.06573.pdf) | 做early fusion来加强时空间建模 |[](https://www.semanticscholar.org/paper/Convolutional-Two-Stream-Network-Fusion-for-Video-Feichtenhofer-Pinz/9d9aced120e530484609164c836da64548693484) |
| ✅ | 2016 | [TSN](https://arxiv.org/pdf/1608.00859.pdf) | 超级有效的视频分段建模，bag of tricks in video |[](https://www.semanticscholar.org/paper/Temporal-Segment-Networks%3A-Towards-Good-Practices-Wang-Xiong/ea3d7de6c0880e14455b9acb28f1bc1234321456) |
| ✅ | 2017 | [I3D](https://arxiv.org/pdf/1705.07750.pdf) | 提出Kinetics数据集，膨胀2D网络到3D，开启3D-CNN时代 |[](https://www.semanticscholar.org/paper/Quo-Vadis%2C-Action-Recognition-A-New-Model-and-the-Carreira-Zisserman/b61a3f8b80bbd44f24544dc915f52fd30bbdf485) |
| ✅ | 2017 | [R2+1D](https://arxiv.org/pdf/1711.11248.pdf) | 拆分3D卷积核，使3D网络容易优化 |[](https://www.semanticscholar.org/paper/A-Closer-Look-at-Spatiotemporal-Convolutions-for-Tran-Wang/89c3050522a0bb9820c32dc7444e003ef0d3e2e4) |
| ✅ | 2017 | [Non-local](https://arxiv.org/pdf/1711.07971.pdf) | 引入自注意力做视觉问题 |[](https://www.semanticscholar.org/paper/Non-local-Neural-Networks-Wang-Girshick/8899094797e82c5c185a0893896320ef77f60e64) |
| ✅ | 2018 | [SlowFast](https://arxiv.org/pdf/1812.03982.pdf) | 快慢两支提升效率 |[](https://www.semanticscholar.org/paper/SlowFast-Networks-for-Video-Recognition-Feichtenhofer-Fan/8b47b9c3c35b2b2a78bff7822605b3040f87d699) |
| ✅ | 2021 | [TimeSformer](https://arxiv.org/pdf/2102.05095.pdf) | 视频中第一个引入transformer，开启video transformer时代 |[](https://www.semanticscholar.org/paper/Is-Space-Time-Attention-All-You-Need-for-Video-Bertasius-Wang/c143ea9e30b1f2d93a9c060253845423f9e60e1f) |
### 多模态学习
| 已录制 | 年份 | 名字 | 简介 | 引用 |
| ------ | ---- | ------------------------------------------------------------ | -------------------- | ------------------------------------------------------------ |
| ✅ | 2021 | [CLIP](https://openai.com/blog/clip/) | 图片和文本之间的对比学习 |[](https://www.semanticscholar.org/paper/Learning-Transferable-Visual-Models-From-Natural-Radford-Kim/6f870f7f02a8c59c3e23f407f3ef00dd1dcf8fc4) |
| ✅ | 2021 | [ViLT](https://arxiv.org/pdf/2102.03334.pdf) | 第一个摆脱了目标检测的视觉文本模型 |[](https://www.semanticscholar.org/paper/ViLT%3A-Vision-and-Language-Transformer-Without-or-Kim-Son/0839722fb5369c0abaff8515bfc08299efc790a1) |
| ✅ | 2021 | [ViLD](https://arxiv.org/pdf/2104.13921.pdf) | CLIP蒸馏帮助开集目标检测 |[](https://www.semanticscholar.org/paper/Open-vocabulary-Object-Detection-via-Vision-and-Gu-Lin/cf9b8da26d9b92e75ba49616ed2a1033f59fce14) |
| ✅ | 2021 | [GLIP](https://arxiv.org/pdf/2112.03857.pdf) | 联合目标检测和文本定位 |[](https://www.semanticscholar.org/paper/Grounded-Language-Image-Pre-training-Li-Zhang/5341b412383c43f4a693ad63ec4489e3ec7688c8) |
| ✅ | 2021 | [CLIP4Clip](https://arxiv.org/pdf/2104.08860.pdf) | 拿CLIP直接做视频文本retrieval |[](https://www.semanticscholar.org/paper/CLIP4Clip%3A-An-Empirical-Study-of-CLIP-for-End-to-Luo-Ji/281ad83e06d731d5d686acf07cd701576f1188c4) |
| ✅ | 2021 | [ActionCLIP](https://arxiv.org/pdf/2109.08472.pdf) | 用多模态对比学习有监督的做视频动作分类 |[](https://www.semanticscholar.org/paper/ActionCLIP%3A-A-New-Paradigm-for-Video-Action-Wang-Xing/dc05240a06326b5b1664f7e8c95c330b08cd0349) |
| ✅ | 2021 | [PointCLIP](https://arxiv.org/pdf/2112.02413.pdf) | 3D变2D，巧妙利用CLIP做点云 |[](https://www.semanticscholar.org/paper/PointCLIP%3A-Point-Cloud-Understanding-by-CLIP-Zhang-Guo/f3ce9ba3fcec362b70263a7ed63d9404975496a0) |
| ✅ | 2022 | [LSeg](https://arxiv.org/pdf/2201.03546.pdf) | 有监督的开集分割 |[](https://www.semanticscholar.org/paper/Language-driven-Semantic-Segmentation-Li-Weinberger/cc9826c222ac1e81b4b374dd9e0df130f298b1e8) |
| ✅ | 2022 | [GroupViT](https://arxiv.org/pdf/2202.11094.pdf) | 只用图像文本对也能无监督做分割 |[](https://www.semanticscholar.org/paper/GroupViT%3A-Semantic-Segmentation-Emerges-from-Text-Xu-Mello/0b5f27a5766c5d1394a6282ad94fec21d620bd6b) |
| ✅ | 2022 | [CLIPasso](https://arxiv.org/pdf/2202.05822.pdf) | CLIP跨界生成简笔画 |[](https://www.semanticscholar.org/paper/CLIPasso%3A-Semantically-Aware-Object-Sketching-Vinker-Pajouheshgar/9dec819778bebae4a468c7813f7638534c826f52) |
| ✅ | 2022 | [DepthCLIP](https://arxiv.org/pdf/2207.01077.pdf) | 用文本跨界估计深度 |[](https://www.semanticscholar.org/paper/Can-Language-Understand-Depth-Zhang-Zeng/9d0afe58801fe9e5537902e853d6e9e385340a92) |
### 自然语言处理 - Transformer
| 已录制 | 年份 | 名字 | 简介 | 引用 |
| ------ | ---- | ------------------------------------------------------------ | -------------------- | ------------------------------------------------------------ |
| ✅ | 2017 | [Transformer](https://arxiv.org/abs/1706.03762) | 继MLP、CNN、RNN后的第四大类架构 |[](https://www.semanticscholar.org/paper/Attention-is-All-you-Need-Vaswani-Shazeer/204e3073870fae3d05bcbc2f6a8e263d9b72e776) |
| ✅ | 2018 | [GPT](https://s3-us-west-2.amazonaws.com/openai-assets/research-covers/language-unsupervised/language_understanding_paper.pdf) | 使用 Transformer 解码器来做预训练 |[](https://www.semanticscholar.org/paper/Improving-Language-Understanding-by-Generative-Radford-Narasimhan/cd18800a0fe0b668a1cc19f2ec95b5003d0a5035) |
| ✅ | 2018 | [BERT](https://arxiv.org/abs/1810.04805) | Transformer一统NLP的开始 |[](https://www.semanticscholar.org/paper/BERT%3A-Pre-training-of-Deep-Bidirectional-for-Devlin-Chang/df2b0e26d0599ce3e70df8a9da02e51594e0e992) |
| ✅ | 2019 | [GPT-2](https://d4mucfpksywv.cloudfront.net/better-language-models/language_models_are_unsupervised_multitask_learners.pdf) | 更大的 GPT 模型，朝着zero-shot learning迈了一大步 |[](https://www.semanticscholar.org/paper/Language-Models-are-Unsupervised-Multitask-Learners-Radford-Wu/9405cc0d6169988371b2755e573cc28650d14dfe) |
| ✅ | 2020 | [GPT-3](https://arxiv.org/abs/2005.14165) | 100倍更大的 GPT-2，few-shot learning效果显著 |[](https://www.semanticscholar.org/paper/Language-Models-are-Few-Shot-Learners-Brown-Mann/6b85b63579a916f705a8e10a49bd8d849d91b1fc) |
| ✅ | 2024 | [Llama 3.1](https://arxiv.org/pdf/2407.21783) | 强大的Meta开源模型 - 动态扩展，多模态学习，零样本学习，高效计算 |[](https://www.semanticscholar.org/paper/4176a4cecfaef26b2c503827493867e703f3411a) |
### 系统
| 已录制 | 年份 | 名字 | 简介 | 引用 |
| ------ | ---- | ------------------------------------------------------------ | -------------------- | ------------------------------------------------------------ |
| ✅ | 2014 | [参数服务器](https://www.usenix.org/system/files/conference/osdi14/osdi14-paper-li_mu.pdf) | 支持千亿参数的传统机器学习模型 |[](https://www.semanticscholar.org/paper/Scaling-Distributed-Machine-Learning-with-the-Li-Andersen/0de0c3240bda7972bd0a3c8369ebc4b4f2e4f9c2) |
| ✅ | 2018 | [GPipe](https://proceedings.neurips.cc/paper/2019/file/093f65e080a295f8076b1c5722a46aa2-Paper.pdf) | 流水线（Pipeline）并行 |[](https://www.semanticscholar.org/paper/GPipe%3A-Efficient-Training-of-Giant-Neural-Networks-Huang-Cheng/c18663fea10c8a303d045fd2c1f33cacf9b73ca3) |
| ✅ | 2019 | [Megatron-LM](https://arxiv.org/pdf/1909.08053.pdf) | 张量（Tensor）并行 |[](https://www.semanticscholar.org/paper/Megatron-LM%3A-Training-Multi-Billion-Parameter-Using-Shoeybi-Patwary/8323c591e119eb09b28b29fd6c7bc76bd889df7a) |
| ✅ | 2019 | [Zero](https://arxiv.org/pdf/1910.02054.pdf) | 参数分片 |[](https://www.semanticscholar.org/paper/ZeRO%3A-Memory-optimizations-Toward-Training-Trillion-Rajbhandari-Rasley/00c957711b12468cb38424caccdf5291bb354033) |
| ✅ | 2022 | [Pathways](https://arxiv.org/pdf/2203.12533.pdf) | 将Jax拓展到上千TPU核上 |[](https://www.semanticscholar.org/paper/Pathways%3A-Asynchronous-Distributed-Dataflow-for-ML-Barham-Chowdhery/512e9aa873a1cd7eb61ce1f25ca7df6acb7e2352) |
### 图神经网络
| 已录制 | 年份 | 名字 | 简介 | 引用 |
| ------ | ---- | ------------------------------------------------------------ | -------------------- | ------------------------------------------------------------ |
| ✅ | 2021 | [图神经网络介绍](https://distill.pub/2021/gnn-intro/) | GNN的可视化介绍 |[](https://www.semanticscholar.org/paper/A-Gentle-Introduction-to-Graph-Neural-Networks-S%C3%A1nchez-Lengeling-Reif/2c0e0440882a42be752268d0b64243243d752a74) |
### 优化算法
| 已录制 | 年份 | 名字 | 简介 | 引用 |
| ------ | ---- | ------------------------------------------------------------ | -------------------- | ------------------------------------------------------------ |
| | 2014 | [Adam](https://arxiv.org/abs/1412.6980) | 深度学习里最常用的优化算法之一 |[](https://www.semanticscholar.org/paper/Adam%3A-A-Method-for-Stochastic-Optimization-Kingma-Ba/a6cb366736791bcccc5c8639de5a8f9636bf87e8) |
| | 2016 | [为什么超大的模型泛化性不错](https://arxiv.org/abs/1611.03530) | |[](https://www.semanticscholar.org/paper/Understanding-deep-learning-requires-rethinking-Zhang-Bengio/54ddb00fa691728944fd8becea90a373d21597cf) |
| | 2017 | [为什么Momentum有效](https://distill.pub/2017/momentum/) | Distill的可视化介绍 |[](https://www.semanticscholar.org/paper/Why-Momentum-Really-Works-Goh/3e8ccf9d3d843c9855c5d76ab66d3e775384da72) |
### 新领域应用
| 已录制 | 年份 | 名字 | 简介 | 引用 |
| ------ | ---- | ------------------------------------------------------------ | -------------------- | ------------------------------------------------------------ |
| | 2016 | [AlphaGo](https://storage.googleapis.com/deepmind-media/alphago/AlphaGoNaturePaper.pdf) | 强化学习出圈 |[](https://www.semanticscholar.org/paper/Mastering-the-game-of-Go-with-deep-neural-networks-Silver-Huang/846aedd869a00c09b40f1f1f35673cb22bc87490) |
| | 2020 | [AlphaFold](https://discovery.ucl.ac.uk/id/eprint/10089234/1/343019_3_art_0_py4t4l_convrt.pdf) | 赢得比赛的的蛋白质3D结构预测 |[](https://www.semanticscholar.org/paper/Improved-protein-structure-prediction-using-from-Senior-Evans/3a083d843f891b3574494c385699c21766ce8b7a) |
| ✅ | 2021 | [AlphaFold 2](https://www.nature.com/articles/s41586-021-03819-2.pdf) | 原子级别精度的蛋白质3D结构预测 |[](https://www.semanticscholar.org/paper/Highly-accurate-protein-structure-prediction-with-Jumper-Evans/dc32a984b651256a8ec282be52310e6bd33d9815) |
| ✅ | 2021 | [Codex](https://arxiv.org/pdf/2107.03374.pdf) | 使用注释生成代码 |[](https://www.semanticscholar.org/paper/Evaluating-Large-Language-Models-Trained-on-Code-Chen-Tworek/acbdbf49f9bc3f151b93d9ca9a06009f4f6eb269) |
| ✅ | 2021 | [指导数学直觉](https://www.nature.com/articles/s41586-021-04086-x.pdf) | 分析不同数学物体之前的联系来帮助发现新定理 |[](https://www.semanticscholar.org/paper/Advancing-mathematics-by-guiding-human-intuition-AI-Davies-Velickovic/f672b8fb430606fee0bb368f16603531ce1e90c4) |
| ✅ | 2022 | [AlphaCode](https://storage.googleapis.com/deepmind-media/AlphaCode/competition_level_code_generation_with_alphacode.pdf) | 媲美一般程序员的编程解题水平 |[](https://www.semanticscholar.org/paper/Competition-Level-Code-Generation-with-AlphaCode-Li-Choi/5cbe278b65a81602a864184bbca37de91448a5f5) |
", Assign "at most 3 tags" to the expected json: {"id":"5796","tags":[]} "only from the tags list I provide: [{"id":77,"name":"3d"},{"id":89,"name":"agent"},{"id":17,"name":"ai"},{"id":54,"name":"algorithm"},{"id":24,"name":"api"},{"id":44,"name":"authentication"},{"id":3,"name":"aws"},{"id":27,"name":"backend"},{"id":60,"name":"benchmark"},{"id":72,"name":"best-practices"},{"id":39,"name":"bitcoin"},{"id":37,"name":"blockchain"},{"id":1,"name":"blog"},{"id":45,"name":"bundler"},{"id":58,"name":"cache"},{"id":21,"name":"chat"},{"id":49,"name":"cicd"},{"id":4,"name":"cli"},{"id":64,"name":"cloud-native"},{"id":48,"name":"cms"},{"id":61,"name":"compiler"},{"id":68,"name":"containerization"},{"id":92,"name":"crm"},{"id":34,"name":"data"},{"id":47,"name":"database"},{"id":8,"name":"declarative-gui "},{"id":9,"name":"deploy-tool"},{"id":53,"name":"desktop-app"},{"id":6,"name":"dev-exp-lib"},{"id":59,"name":"dev-tool"},{"id":13,"name":"ecommerce"},{"id":26,"name":"editor"},{"id":66,"name":"emulator"},{"id":62,"name":"filesystem"},{"id":80,"name":"finance"},{"id":15,"name":"firmware"},{"id":73,"name":"for-fun"},{"id":2,"name":"framework"},{"id":11,"name":"frontend"},{"id":22,"name":"game"},{"id":81,"name":"game-engine "},{"id":23,"name":"graphql"},{"id":84,"name":"gui"},{"id":91,"name":"http"},{"id":5,"name":"http-client"},{"id":51,"name":"iac"},{"id":30,"name":"ide"},{"id":78,"name":"iot"},{"id":40,"name":"json"},{"id":83,"name":"julian"},{"id":38,"name":"k8s"},{"id":31,"name":"language"},{"id":10,"name":"learning-resource"},{"id":33,"name":"lib"},{"id":41,"name":"linter"},{"id":28,"name":"lms"},{"id":16,"name":"logging"},{"id":76,"name":"low-code"},{"id":90,"name":"message-queue"},{"id":42,"name":"mobile-app"},{"id":18,"name":"monitoring"},{"id":36,"name":"networking"},{"id":7,"name":"node-version"},{"id":55,"name":"nosql"},{"id":57,"name":"observability"},{"id":46,"name":"orm"},{"id":52,"name":"os"},{"id":14,"name":"parser"},{"id":74,"name":"react"},{"id":82,"name":"real-time"},{"id":56,"name":"robot"},{"id":65,"name":"runtime"},{"id":32,"name":"sdk"},{"id":71,"name":"search"},{"id":63,"name":"secrets"},{"id":25,"name":"security"},{"id":85,"name":"server"},{"id":86,"name":"serverless"},{"id":70,"name":"storage"},{"id":75,"name":"system-design"},{"id":79,"name":"terminal"},{"id":29,"name":"testing"},{"id":12,"name":"ui"},{"id":50,"name":"ux"},{"id":88,"name":"video"},{"id":20,"name":"web-app"},{"id":35,"name":"web-server"},{"id":43,"name":"webassembly"},{"id":69,"name":"workflow"},{"id":87,"name":"yaml"}]" returns me the "expected json"