Repository: mli/paper-reading
Branch: main
Commit: 4c0c431e877f
Files: 3
Total size: 87.5 KB
Directory structure:
gitextract_qq6_x0jl/
├── .gitignore
├── LICENSE
└── README.md
================================================
FILE CONTENTS
================================================
================================================
FILE: .gitignore
================================================
imgs/*.png
.DS_Store
convert.sh
================================================
FILE: LICENSE
================================================
Apache License
Version 2.0, January 2004
http://www.apache.org/licenses/
TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
1. Definitions.
"License" shall mean the terms and conditions for use, reproduction,
and distribution as defined by Sections 1 through 9 of this document.
"Licensor" shall mean the copyright owner or entity authorized by
the copyright owner that is granting the License.
"Legal Entity" shall mean the union of the acting entity and all
other entities that control, are controlled by, or are under common
control with that entity. For the purposes of this definition,
"control" means (i) the power, direct or indirect, to cause the
direction or management of such entity, whether by contract or
otherwise, or (ii) ownership of fifty percent (50%) or more of the
outstanding shares, or (iii) beneficial ownership of such entity.
"You" (or "Your") shall mean an individual or Legal Entity
exercising permissions granted by this License.
"Source" form shall mean the preferred form for making modifications,
including but not limited to software source code, documentation
source, and configuration files.
"Object" form shall mean any form resulting from mechanical
transformation or translation of a Source form, including but
not limited to compiled object code, generated documentation,
and conversions to other media types.
"Work" shall mean the work of authorship, whether in Source or
Object form, made available under the License, as indicated by a
copyright notice that is included in or attached to the work
(an example is provided in the Appendix below).
"Derivative Works" shall mean any work, whether in Source or Object
form, that is based on (or derived from) the Work and for which the
editorial revisions, annotations, elaborations, or other modifications
represent, as a whole, an original work of authorship. For the purposes
of this License, Derivative Works shall not include works that remain
separable from, or merely link (or bind by name) to the interfaces of,
the Work and Derivative Works thereof.
"Contribution" shall mean any work of authorship, including
the original version of the Work and any modifications or additions
to that Work or Derivative Works thereof, that is intentionally
submitted to Licensor for inclusion in the Work by the copyright owner
or by an individual or Legal Entity authorized to submit on behalf of
the copyright owner. For the purposes of this definition, "submitted"
means any form of electronic, verbal, or written communication sent
to the Licensor or its representatives, including but not limited to
communication on electronic mailing lists, source code control systems,
and issue tracking systems that are managed by, or on behalf of, the
Licensor for the purpose of discussing and improving the Work, but
excluding communication that is conspicuously marked or otherwise
designated in writing by the copyright owner as "Not a Contribution."
"Contributor" shall mean Licensor and any individual or Legal Entity
on behalf of whom a Contribution has been received by Licensor and
subsequently incorporated within the Work.
2. Grant of Copyright License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
copyright license to reproduce, prepare Derivative Works of,
publicly display, publicly perform, sublicense, and distribute the
Work and such Derivative Works in Source or Object form.
3. Grant of Patent License. Subject to the terms and conditions of
this License, each Contributor hereby grants to You a perpetual,
worldwide, non-exclusive, no-charge, royalty-free, irrevocable
(except as stated in this section) patent license to make, have made,
use, offer to sell, sell, import, and otherwise transfer the Work,
where such license applies only to those patent claims licensable
by such Contributor that are necessarily infringed by their
Contribution(s) alone or by combination of their Contribution(s)
with the Work to which such Contribution(s) was submitted. If You
institute patent litigation against any entity (including a
cross-claim or counterclaim in a lawsuit) alleging that the Work
or a Contribution incorporated within the Work constitutes direct
or contributory patent infringement, then any patent licenses
granted to You under this License for that Work shall terminate
as of the date such litigation is filed.
4. Redistribution. You may reproduce and distribute copies of the
Work or Derivative Works thereof in any medium, with or without
modifications, and in Source or Object form, provided that You
meet the following conditions:
(a) You must give any other recipients of the Work or
Derivative Works a copy of this License; and
(b) You must cause any modified files to carry prominent notices
stating that You changed the files; and
(c) You must retain, in the Source form of any Derivative Works
that You distribute, all copyright, patent, trademark, and
attribution notices from the Source form of the Work,
excluding those notices that do not pertain to any part of
the Derivative Works; and
(d) If the Work includes a "NOTICE" text file as part of its
distribution, then any Derivative Works that You distribute must
include a readable copy of the attribution notices contained
within such NOTICE file, excluding those notices that do not
pertain to any part of the Derivative Works, in at least one
of the following places: within a NOTICE text file distributed
as part of the Derivative Works; within the Source form or
documentation, if provided along with the Derivative Works; or,
within a display generated by the Derivative Works, if and
wherever such third-party notices normally appear. The contents
of the NOTICE file are for informational purposes only and
do not modify the License. You may add Your own attribution
notices within Derivative Works that You distribute, alongside
or as an addendum to the NOTICE text from the Work, provided
that such additional attribution notices cannot be construed
as modifying the License.
You may add Your own copyright statement to Your modifications and
may provide additional or different license terms and conditions
for use, reproduction, or distribution of Your modifications, or
for any such Derivative Works as a whole, provided Your use,
reproduction, and distribution of the Work otherwise complies with
the conditions stated in this License.
5. Submission of Contributions. Unless You explicitly state otherwise,
any Contribution intentionally submitted for inclusion in the Work
by You to the Licensor shall be under the terms and conditions of
this License, without any additional terms or conditions.
Notwithstanding the above, nothing herein shall supersede or modify
the terms of any separate license agreement you may have executed
with Licensor regarding such Contributions.
6. Trademarks. This License does not grant permission to use the trade
names, trademarks, service marks, or product names of the Licensor,
except as required for reasonable and customary use in describing the
origin of the Work and reproducing the content of the NOTICE file.
7. Disclaimer of Warranty. Unless required by applicable law or
agreed to in writing, Licensor provides the Work (and each
Contributor provides its Contributions) on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
implied, including, without limitation, any warranties or conditions
of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
PARTICULAR PURPOSE. You are solely responsible for determining the
appropriateness of using or redistributing the Work and assume any
risks associated with Your exercise of permissions under this License.
8. Limitation of Liability. In no event and under no legal theory,
whether in tort (including negligence), contract, or otherwise,
unless required by applicable law (such as deliberate and grossly
negligent acts) or agreed to in writing, shall any Contributor be
liable to You for damages, including any direct, indirect, special,
incidental, or consequential damages of any character arising as a
result of this License or out of the use or inability to use the
Work (including but not limited to damages for loss of goodwill,
work stoppage, computer failure or malfunction, or any and all
other commercial damages or losses), even if such Contributor
has been advised of the possibility of such damages.
9. Accepting Warranty or Additional Liability. While redistributing
the Work or Derivative Works thereof, You may choose to offer,
and charge a fee for, acceptance of support, warranty, indemnity,
or other liability obligations and/or rights consistent with this
License. However, in accepting such obligations, You may act only
on Your own behalf and on Your sole responsibility, not on behalf
of any other Contributor, and only if You agree to indemnify,
defend, and hold each Contributor harmless for any liability
incurred by, or claims asserted against, such Contributor by reason
of your accepting any such warranty or additional liability.
END OF TERMS AND CONDITIONS
APPENDIX: How to apply the Apache License to your work.
To apply the Apache License to your work, attach the following
boilerplate notice, with the fields enclosed by brackets "[]"
replaced with your own identifying information. (Don't include
the brackets!) The text should be enclosed in the appropriate
comment syntax for the file format. We also recommend that a
file or class name and description of purpose be included on the
same "printed page" as the copyright notice for easier
identification within third-party archives.
Copyright [yyyy] [name of copyright owner]
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
================================================
FILE: README.md
================================================
# 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） |
| --: | -- | -- | --: | -- |
| 1/10/25 | [OpenAI Sora](https://openai.com/index/video-generation-models-as-world-simulators/) 上
(包含Movie Gen和HunyuanVideo) | 
| 1:04:18 | [](https://www.bilibili.com/video/BV1VdcxesEAt/?share_source=copy_web&vd_source=5d037e935914fc22e2e978cdccf5cdfe)
[](https://youtu.be/5MGq7dSOghY?si=lY-OsadDsTeKf-ub) |
| 9/04/24 | Llama 3.1论文精读 · 5. 模型训练过程 | 
| 10:41| [](https://www.bilibili.com/video/BV1c8HbeaEXi)
|
| 8/28/24 | Llama 3.1论文精读 · 4. 训练infra | 
| 25:04| [](https://www.bilibili.com/video/BV1b4421f7fa)
[](https://www.youtube.com/watch?v=6XidEHVjS1A) |
| 8/13/24 | Llama 3.1论文精读 · 3. 模型 | 
| 26:14| [](https://www.bilibili.com/video/BV1Q4421Z7Tj)
[](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) | 
| 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 | 
| 18:53| [](https://www.bilibili.com/video/BV1WM4m1y7Uh)
[](https://www.youtube.com/watch?v=-PztagF3wQE) |
| 3/30/23 | [GPT-4](https://openai.com/research/gpt-4) | 
| 1:20:38 | [](https://www.bilibili.com/video/BV1vM4y1U7b5)
[](https://youtu.be/K0SZ9mdygTw) |
| 3/23/23 | 大模型时代下做科研的四个思路 | 
| 1:06:29 | [](https://www.bilibili.com/video/BV1oX4y1d7X6)
[](https://youtu.be/sh79Z8i15PI) |
| 3/10/23 | [Anthropic LLM](https://arxiv.org/pdf/2204.05862.pdf) | 
| 1:01:51 | [](https://www.bilibili.com/video/BV1XY411B7nM)
[](https://youtu.be/iqX0pgNDon0) |
| 1/20/23 | [Helm](https://arxiv.org/pdf/2211.09110.pdf) 全面语言模型评测 | 
| 1:23:37 | [](https://www.bilibili.com/video/BV1z24y1B7uX)
[](https://youtu.be/WgFEw9U3BXA) |
| 1/11/23 | 多模态论文串讲·下 | 
| 1:03:29 | [](https://www.bilibili.com/video/BV1fA411Z772)
[](https://youtu.be/S1le41J76lQ) |
| 12/29/22 | [Instruct GPT](https://arxiv.org/pdf/2203.02155.pdf) | 
| 1:07:10 | [](https://www.bilibili.com/video/BV1hd4y187CR)
[](https://youtu.be/zfIGAwD1jOQ) |
| 12/19/22 | [Neural Corpus Indexer](https://arxiv.org/pdf/2206.02743.pdf) 文档检索 | 
| 55:47 | [](https://www.bilibili.com/video/BV1Se411w7Sn)
[](https://youtu.be/QRffZMSGJyU) |
| 12/12/22 | 多模态论文串讲·上 | 
| 1:12:27 | [](https://www.bilibili.com/video/BV1Vd4y1v77v)
[](https://youtu.be/6pzBOQAXUB8) |
| 11/14/22 | [OpenAI Whisper](https://cdn.openai.com/papers/whisper.pdf) 精读 | 
| 1:12:16 | [](https://www.bilibili.com/video/BV1VG4y1t74x)
[](https://youtu.be/3eXCJd32UnM) |
| 11/07/22 | 在讲 OpenAI Whisper 前先做了一个剪视频小工具 | 
| 23:39 | [](https://www.bilibili.com/video/BV1Pe4y1t7de)
[](https://youtu.be/PwVlvCPDnrI) |
| 10/23/22 | [Chain of Thought](https://arxiv.org/pdf/2201.11903.pdf) 论文、代码和资源 | 
| 33:21 | [](https://www.bilibili.com/video/BV1t8411e7Ug)
[](https://youtu.be/H4J59iG3t5o) |
| 9/17/22 | CLIP 改进工作串讲（下） | 
| 1:04:26 | [](https://www.bilibili.com/video/BV1gg411U7n4)
[](https://youtu.be/ugJeBivv65s) |
| 9/2/22 | CLIP 改进工作串讲（上） | 
| 1:14:43 | [](https://www.bilibili.com/video/BV1FV4y1p7Lm)
[](https://youtu.be/x4CDhZz_Dvg) |
| 7/29/22 | [ViLT](https://arxiv.org/pdf/2102.03334.pdf) 论文精读 | 
| 1:03:26 | [](https://www.bilibili.com/video/BV14r4y1j74y)
[](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】 | 
| 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】| 
| 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 | 
| 1:27:54 |[](https://www.bilibili.com/video/BV17r4y1u77B)
[](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】| 
| 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】 | 
| 45:01 | [](https://www.bilibili.com/video/BV1hY411T7vy/) |
| 6/17/22 | [Zero](https://arxiv.org/pdf/1910.02054.pdf) 逐段精读 | 
| 52:21 | [](https://www.bilibili.com/video/BV1tY411g7ZT/) |
| 6/10/22 | [DETR](https://arxiv.org/pdf/2005.12872.pdf) 逐段精读 | 
| 54:22 | [](https://www.bilibili.com/video/BV1GB4y1X72R/) |
| 6/3/22 | [Megatron LM](https://arxiv.org/pdf/1909.08053.pdf) 逐段精读 | 
| 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 | 
| 58:47 | [](https://www.bilibili.com/video/BV1v34y1E7zu/)
[](https://youtu.be/eXjRpS_BTbs) |
| 5/5/22 | [Pathways](https://arxiv.org/pdf/2203.12533.pdf) 逐段精读 | 
| 1:02:13 | [](https://www.bilibili.com/video/BV1xB4y1m7Xi/)
[](https://youtu.be/8hS1ZtgG0wU) |
| 4/28/22 | [视频理解论文串讲](https://arxiv.org/pdf/2012.06567.pdf)（下） | 
| 1:08:32 | [](https://www.bilibili.com/video/BV11Y411P7ep/)
[](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 | 
| 1:37:40 | [](https://www.bilibili.com/video/BV1YA4y197G8/)
[](https://youtu.be/xt-AwUrDxQk) |
| 4/14/22 | [视频理解论文串讲](https://arxiv.org/pdf/2012.06567.pdf)（上） | 
| 51:15 | [](https://www.bilibili.com/video/BV1fL4y157yA/)
[](https://youtu.be/gK7AGO6okhc) |
| 3/31/22 | [I3D](https://arxiv.org/pdf/1705.07750.pdf) 论文精读 | 
| 52:31 | [](https://www.bilibili.com/video/BV1tY4y1p7hq/)
[](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 | 
| 1:19:56 | [](https://www.bilibili.com/video/BV1s44y1N7eu/)
[](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 | 
| 44:00 | [](https://www.bilibili.com/video/BV1ab4y1s7rc/)
[](https://youtu.be/t8Gzkca9pW4) |
| 3/10/22 | [OpenAI Codex](https://arxiv.org/pdf/2107.03374.pdf) 论文精读 | 
| 47:58 | [](https://www.bilibili.com/video/BV1iY41137Zi/)
[](https://www.zhihu.com/zvideo/1490959755963666432)
[](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 | 
| 1:29:58 | [](https://www.bilibili.com/video/BV1AF411b7xQ/)
[](https://youtu.be/t70Bl3w7bxY) |
| 2/24/22 | [Two-Stream](https://proceedings.neurips.cc/paper/2014/file/00ec53c4682d36f5c4359f4ae7bd7ba1-Paper.pdf) 逐段精读 | 
| 52:57 | [](https://www.bilibili.com/video/BV1mq4y1x7RU/)
[](https://youtu.be/vuqwKP2iDe0) |
| 2/10/22 | [CLIP](https://openai.com/blog/clip/) 逐段精读 | 
| 1:38:25 | [](https://www.bilibili.com/video/BV1SL4y1s7LQ/)
[](https://www.zhihu.com/zvideo/1475706654562299904)
[](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？| 
| 14:11 | [](https://www.bilibili.com/video/BV1ea41127Bq/)
[](https://www.zhihu.com/zvideo/1475719090198876161) |
| 1/23/22 | [AlphaFold 2](https://www.nature.com/articles/s41586-021-03819-2.pdf) 精读 | 
| 1:15:28 | [](https://www.bilibili.com/video/BV1oR4y1K7Xr/)
[](https://www.zhihu.com/zvideo/1469132410537717760)
[](https://youtu.be/Oy3OCoGUr-w) |
| 1/18/22 | 如何判断（你自己的）研究工作的价值 | 
| 9:59 | [](https://www.bilibili.com/video/BV1oL411c7Us/)
[](https://www.zhihu.com/zvideo/1475716940051869696) |
| 1/15/22 | [Swin Transformer](https://arxiv.org/pdf/2103.14030.pdf) 精读 | 
| 1:00:21 | [](https://www.bilibili.com/video/BV13L4y1475U/)
[](https://www.zhihu.com/zvideo/1466282983652691968)
[](https://youtu.be/luP3-Fs0QCo) |
| 1/7/22 | [指导数学直觉](https://www.nature.com/articles/s41586-021-04086-x.pdf) | 
| 52:51 | [](https://www.bilibili.com/video/BV1YZ4y1S72j/)
[](https://www.zhihu.com/zvideo/1464060386375299072)
[](https://youtu.be/czFGjvhtss8) |
| 1/5/22 | AlphaFold 2 预告 | 
| 03:28 | [](https://www.bilibili.com/video/BV1Eu411U7Te/) |
| 12/20/21 | [对比学习](#contrastive_learning)论文综述 | 
| 1:32:01 | [](https://www.bilibili.com/video/BV19S4y1M7hm/)
[](https://www.zhihu.com/zvideo/1460828005077164032)
[](https://www.youtube.com/watch?v=1pvxufGRuW4) |
| 12/15/21 | [MoCo](https://arxiv.org/pdf/1911.05722.pdf) 逐段精读 | 
| 1:24:11 | [](https://www.bilibili.com/video/BV1C3411s7t9/)
[](https://www.zhihu.com/zvideo/1454723120678936576)
[](https://www.youtube.com/watch?v=1pvxufGRuW4) |
| 12/9/21 | 如何找研究想法 1 | 
| 5:34 | [](https://www.bilibili.com/video/BV1qq4y1z7F2/) |
| 12/8/21 | [MAE](https://arxiv.org/pdf/2111.06377.pdf) 逐段精读 | 
| 47:04 | [](https://www.bilibili.com/video/BV1sq4y1q77t/)
[](https://www.zhihu.com/zvideo/1452458167968251904)
[](https://youtu.be/mYlX2dpdHHM) |
| 11/29/21 | [ViT](https://arxiv.org/pdf/2010.11929.pdf) 逐段精读 | 
| 1:11:30 | [](https://www.bilibili.com/video/BV15P4y137jb/)
[](https://www.zhihu.com/zvideo/1449195245754380288)
[](https://youtu.be/FRFt3x0bO94) |
| 11/18/21 | [BERT](https://arxiv.org/abs/1810.04805) 逐段精读 | 
| 45:49 | [](https://www.bilibili.com/video/BV1PL411M7eQ/)
[](https://www.zhihu.com/zvideo/1445340200976785408)
[](https://youtu.be/ULD3uIb2MHQ) |
| 11/9/21 | [GAN](https://papers.nips.cc/paper/2014/file/5ca3e9b122f61f8f06494c97b1afccf3-Paper.pdf) 逐段精读 | 
| 46:16 | [](https://www.bilibili.com/video/BV1rb4y187vD/)
[](https://www.zhihu.com/zvideo/1442091389241159681)
[](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） | 
| 1:06:19 | [](https://www.bilibili.com/video/BV1iT4y1d7zP/)
[](https://www.zhihu.com/zvideo/1439540657619087360)
[](https://youtu.be/sejA2PtCITw) |
| 10/27/21 | [Transformer](https://arxiv.org/abs/1706.03762) 逐段精读
（视频中提到的文献 [^transformer]) |
| 1:27:05 | [](https://www.bilibili.com/video/BV1pu411o7BE/)
[](https://www.zhihu.com/zvideo/1437034536677404672)
[](https://youtu.be/nzqlFIcCSWQ) |
| 10/22/21 | [ResNet](https://arxiv.org/abs/1512.03385) 论文逐段精读 | 
| 53:46 | [](https://www.bilibili.com/video/BV1P3411y7nn/)
[](https://www.zhihu.com/zvideo/1434795406001180672)
[](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) | 
| 11:50 | [](https://www.bilibili.com/video/BV1Fb4y1h73E/)
[](https://www.zhihu.com/zvideo/1434787226101751808)
[](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 | 
| 55:21 | [](https://www.bilibili.com/video/BV1hq4y157t1/)
[](https://www.zhihu.com/zvideo/1432354207483871232)
[](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) | 
| 19:59 | [](https://www.bilibili.com/video/BV1ih411J7Kz/)
[](https://www.zhihu.com/zvideo/1432155856322920448)
[](https://www.youtube.com/watch?v=vdYH0fE6thY) |
| 10/06/21 | 如何读论文 | 
| 06:39 | [](https://www.bilibili.com/video/BV1H44y1t75x/)
[](https://www.zhihu.com/zvideo/1428973951632969728)
[](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) |
| ✅ | 2024 | [Sora](https://openai.com/index/video-generation-models-as-world-simulators/) | 开启视频生成热潮 | |
| ✅ | 2024 | [Movie Gen](https://arxiv.org/pdf/2410.13720) | 精确的文本指导视频编辑、个性化视频生成 | |
| ✅ | 2025 | [HunyuanVideo](https://arxiv.org/pdf/2412.03603) | 开源视频生成框架 | |
### 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) |
### 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) |