@wsl8297: Discovered a deep learning paper reading project on GitHub: paper-reading. Author Mu Shen reads classic and new deep learning papers paragraph by paragraph, recorded into video explanations, has been updated for over 3 years. GitHub: https://github.com/mli/paper-reading...
Summary
Mu Shen's deep learning paper reading project on GitHub includes in-depth reading videos of major papers such as GPT-4, Llama 3.1, Sora, etc. Each video is about 1 hour, suitable for AI researchers and developers to deeply understand classic papers.
Discovered a deep learning paper reading project on GitHub: paper-reading. Author Mu Shen reads classic and new deep learning papers paragraph by paragraph, records video explanations, and has been updated for over 3 years. GitHub: https://github.com/mli/paper-reading... The project includes in-depth reading videos of significant papers such as GPT-4, Llama 3.1, Sora, DALL·E 2, Instruct GPT, Whisper, Chain of Thought, etc. Each video is about 1 hour of deep explanation, breaking down the paper content paragraph by paragraph. The videos are simultaneously updated on Bilibili and YouTube, and also include series such as multimodal paper overviews and CLIP improvement work overviews. Besides paper reading, there are also sharing of research ideas in the era of large models, research methodology, etc. Suitable for AI researchers and developers who want to deeply understand classic papers and keep up with cutting-edge progress.
View Cached Full Text
Cached at: 05/15/26, 10:59 AM
Deep Learning Paper Reading Project: paper-reading
Discovered a deep learning paper reading project on GitHub: paper-reading. The author, Mu Shen, does a paragraph-by-paragraph deep reading of classic and new deep learning papers, recording video explanations that have been updated for over 3 years. GitHub: https://github.com/mli/paper-reading… The project includes in-depth reading videos for major papers like GPT-4, Llama 3.1, Sora, DALL·E 2, Instruct GPT, Whisper, Chain of Thought, etc. Each video is about an hour of deep explanation, breaking down the paper paragraph by paragraph. Videos are also updated simultaneously on Bilibili and YouTube, and include series such as multimodal paper overviews and CLIP improvement work overviews. In addition to paper reading, there are also sharing of research ideas in the era of large models, research methodology, and more. Suitable for AI researchers and developers who want to deeply understand classic papers and keep up with cutting-edge progress.
mli/paper-reading
Source: https://github.com/mli/paper-reading
Deep Learning Paper Reading
Recorded Papers
| Date | Title | Cover | Duration | Video (Views) |
|---|---|---|---|---|
| 1/10/25 | OpenAI Sora (https://openai.com/index/video-generation-models-as-world-simulators/) Part 1 (including Movie Gen and HunyuanVideo) | 1:04:18 | bilibili (https://www.bilibili.com/video/BV1VdcxesEAt/?share_source=copy_web&vd_source=5d037e935914fc22e2e978cdccf5cdfe) | |
| 9/04/24 | Llama 3.1 Paper Reading · 5. Model Training Process | 10:41 | bilibili (https://www.bilibili.com/video/BV1c8HbeaEXi) | |
| 8/28/24 | Llama 3.1 Paper Reading · 4. Training Infrastructure | 25:04 | bilibili (https://www.bilibili.com/video/BV1b4421f7fa) | |
| 8/13/24 | Llama 3.1 Paper Reading · 3. Model | 26:14 | bilibili (https://www.bilibili.com/video/BV1Q4421Z7Tj) | |
| 8/05/24 | Llama 3.1 Paper Reading · 2. Pre-training Data (https://arxiv.org/pdf/2407.21783) | 23:37 | bilibili (https://www.bilibili.com/video/BV1u142187S5) | |
| 7/31/24 | Llama 3.1 Paper Reading · 1. Introduction | 18:53 | bilibili (https://www.bilibili.com/video/BV1WM4m1y7Uh) | |
| 3/30/23 | GPT-4 (https://openai.com/research/gpt-4) | 1:20:38 | bilibili (https://www.bilibili.com/video/BV1vM4y1U7b5) | |
| 3/23/23 | Four Research Ideas in the Era of Large Models | 1:06:29 | bilibili (https://www.bilibili.com/video/BV1oX4y1d7X6) | |
| 3/10/23 | Anthropic LLM (https://arxiv.org/pdf/2204.05862.pdf) | 1:01:51 | bilibili (https://www.bilibili.com/video/BV1XY411B7nM) | |
| 1/20/23 | Helm (https://arxiv.org/pdf/2211.09110.pdf) Comprehensive Language Model Evaluation | 1:23:37 | bilibili (https://www.bilibili.com/video/BV1z24y1B7uX) | |
| 1/11/23 | Multimodal Paper Overview · Part 2 | 1:03:29 | bilibili (https://www.bilibili.com/video/BV1fA411Z772) | |
| 12/29/22 | Instruct GPT (https://arxiv.org/pdf/2203.02155.pdf) | 1:07:10 | bilibili (https://www.bilibili.com/video/BV1hd4y187CR) | |
| 12/19/22 | Neural Corpus Indexer (https://arxiv.org/pdf/2206.02743.pdf) Document Retrieval | 55:47 | bilibili (https://www.bilibili.com/video/BV1Se411w7Sn) | |
| 12/12/22 | Multimodal Paper Overview · Part 1 | 1:12:27 | bilibili (https://www.bilibili.com/video/BV1Vd4y1v77v) | |
| 11/14/22 | OpenAI Whisper (https://cdn.openai.com/papers/whisper.pdf) In-depth Reading | 1:12:16 | bilibili (https://www.bilibili.com/video/BV1VG4y1t74x) | |
| 11/07/22 | Before Talking About OpenAI Whisper, I Made a Little Video Editing Tool | 23:39 | bilibili (https://www.bilibili.com/video/BV1Pe4y1t7de) | |
| 10/23/22 | Chain of Thought (https://arxiv.org/pdf/2201.11903.pdf) Paper, Code, and Resources | 33:21 | bilibili (https://www.bilibili.com/video/BV1t8411e7Ug) | |
| 9/17/22 | CLIP Improvement Work Overview (Part 2) | 1:04:26 | bilibili (https://www.bilibili.com/video/BV1gg411U7n4) | |
| 9/2/22 | CLIP Improvement Work Overview (Part 1) | 1:14:43 | bilibili (https://www.bilibili.com/video/BV1FV4y1p7Lm) | |
| 7/29/22 | ViLT (https://arxiv.org/pdf/2102.03334.pdf) Paper In-depth Reading | 1:03:26 | bilibili (https://www.bilibili.com/video/BV14r4y1j74y) | |
| 7/22/22 | Reasons, Evidence, and Warrants [The Craft of Research (https://press.uchicago.edu/ucp/books/book/chicago/C/bo23521678.html) · 4] | 44:14 | bilibili (https://www.bilibili.com/video/BV1SB4y1a75c) | |
| 7/15/22 | How to Tell a Good Story, Arguments in a Story [The Craft of Research (https://press.uchicago.edu/ucp/books/book/chicago/C/bo23521678.html) · 3] | 43:56 | bilibili (https://www.bilibili.com/video/BV1WB4y1v7ST) | |
| 7/8/22 | DALL·E 2 (https://arxiv.org/pdf/2204.06125.pdf) Paragraph-by-Paragraph Reading | 1:27:54 | bilibili (https://www.bilibili.com/video/BV17r4y1u77B) | |
| 7/1/22 | Understanding the Importance of the Problem [The Craft of Research (https://press.uchicago.edu/ucp/books/book/chicago/C/bo23521678.html) · 2] | 1:03:40 | bilibili (https://www.bilibili.com/video/BV11S4y1v7S2/) | |
| 6/24/22 | Connecting with Readers [The Craft of Research (https://press.uchicago.edu/ucp/books/book/chicago/C/bo23521678.html) · 1] | 45:01 | bilibili (https://www.bilibili.com/video/BV1hY411T7vy/) | |
| 6/17/22 | Zero (https://arxiv.org/pdf/1910.02054.pdf) Paragraph-by-Paragraph Reading | 52:21 | bilibili (https://www.bilibili.com/video/BV1tY411g7ZT/) | |
| 6/10/22 | DETR (https://arxiv.org/pdf/2005.12872.pdf) Paragraph-by-Paragraph Reading | 54:22 | bilibili (https://www.bilibili.com/video/BV1GB4y1X72R/) | |
| 6/3/22 | Megatron LM (https://arxiv.org/pdf/1909.08053.pdf) Paragraph-by-Paragraph Reading | 56:07 | bilibili (https://www.bilibili.com/video/BV1nB4y1R7Yz/) | |
| 5/27/22 | GPipe (https://proceedings.neurips.cc/paper/2019/file/093f65e080a295f8076b1c5722a46aa2-Paper.pdf) Paragraph-by-Paragraph Reading | 58:47 | bilibili (https://www.bilibili.com/video/BV1v34y1E7zu/) | |
| 5/5/22 | Pathways (https://arxiv.org/pdf/2203.12533.pdf) Paragraph-by-Paragraph Reading | 1:02:13 | bilibili (https://www.bilibili.com/video/BV1xB4y1m7Xi/) | |
| 4/28/22 | Video Understanding Paper Overview (https://arxiv.org/pdf/2012.06567.pdf) (Part 2) | 1:08:32 | bilibili (https://www.bilibili.com/video/BV11Y411P7ep/) | |
| 4/21/22 | Parameter Server (https://www.usenix.org/system/files/conference/osdi14/osdi14-paper-li_mu.pdf) Paragraph-by-Paragraph Reading | 1:37:40 | bilibili (https://www.bilibili.com/video/BV1YA4y197G8/) | |
| 4/14/22 | Video Understanding Paper Overview (https://arxiv.org/pdf/2012.06567.pdf) (Part 1) | 51:15 | bilibili (https://www.bilibili.com/video/BV1fL4y157yA/) | |
| 3/31/22 | I3D (https://arxiv.org/pdf/1705.07750.pdf) Paper In-depth Reading | 52:31 | bilibili (https://www.bilibili.com/video/BV1tY4y1p7hq/) | |
| 3/24/22 | Stanford 2022 AI Index Report (https://aiindex.stanford.edu/wp-content/uploads/2022/03/2022-AI-Index-Report_Master.pdf) In-depth Reading | 1:19:56 | bilibili (https://www.bilibili.com/video/BV1s44y1N7eu/) | |
| 3/17/22 | AlphaCode (https://storage.googleapis.com/deepmind-media/AlphaCode/competition_level_code_generation_with_alphacode.pdf) Paper In-depth Reading | 44:00 | bilibili (https://www.bilibili.com/video/BV1ab4y1s7rc/) | |
| 3/10/22 | OpenAI Codex (https://arxiv.org/pdf/2107.03374.pdf) Paper In-depth Reading | 47:58 | bilibili (https://www.bilibili.com/video/BV1iY41137Zi/) zhihu (https://www.zhihu.com/zvideo/1490959755963666432) | |
| 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) In-depth Reading | 1:29:58 | bilibili (https://www.bilibili.com/video/BV1AF411b7xQ/) | |
| 2/24/22 | Two-Stream (https://proceedings.neurips.cc/paper/2014/file/00ec53c4682d36f5c4359f4ae7bd7ba1-Paper.pdf) Paragraph-by-Paragraph Reading | 52:57 | bilibili (https://www.bilibili.com/video/BV1mq4y1x7RU/) | |
| 2/10/22 | CLIP (https://openai.com/blog/clip/) Paragraph-by-Paragraph Reading | 1:38:25 | bilibili (https://www.bilibili.com/video/BV1SL4y1s7LQ/) zhihu (https://www.zhihu.com/zvideo/1475706654562299904) | |
| 2/6/22 | Have You Been (or Complained About) Papers Not Being Novel (https://perceiving-systems.blog/en/post/novelty-in-science) Enough? | 14:11 | bilibili (https://www.bilibili.com/video/BV1ea41127Bq/) zhihu (https://www.zhihu.com/zvideo/1475719090198876161) | |
| 1/23/22 | AlphaFold 2 (https://www.nature.com/articles/s41586-021-03819-2.pdf) In-depth Reading | 1:15:28 | bilibili (https://www.bilibili.com/video/BV1oR4y1K7Xr/) zhihu (https://www.zhihu.com/zvideo/1469132410537717760) | |
| 1/18/22 | How to Judge the Value of (Your Own) Research Work | 9:59 | bilibili (https://www.bilibili.com/video/BV1oL411c7Us/) zhihu (https://www.zhihu.com/zvideo/1475716940051869696) | |
| 1/15/22 | Swin Transformer (https://arxiv.org/pdf/2103.14030.pdf) In-depth Reading | 1:00:21 | bilibili (https://www.bilibili.com/video/BV13L4y1475U/) zhihu (https://www.zhihu.com/zvideo/1466282983652691968) | |
| 1/7/22 | Guiding Mathematical Intuition (https://www.nature.com/articles/s41586-021-04086-x.pdf) | 52:51 | bilibili (https://www.bilibili.com/video/BV1YZ4y1S72j/) zhihu (https://www.zhihu.com/zvideo/1464060386375299072) | |
| 1/5/22 | AlphaFold 2 Preview | 03:28 | bilibili (https://www.bilibili.com/video/BV1Eu411U7Te/) | |
| 12/20/21 | Contrastive Learning Paper Survey | 1:32:01 | bilibili (https://www.bilibili.com/video/BV19S4y1M7hm/) zhihu (https://www.zhihu.com/zvideo/1460828005077164032) | |
| 12/15/21 | MoCo (https://arxiv.org/pdf/1911.05722.pdf) Paragraph-by-Paragraph Reading | 1:24:11 | bilibili (https://www.bilibili.com/video/BV1C3411s7t9/) zhihu (https://www.zhihu.com/zvideo/1454723120678936576) | |
| 12/9/21 | How to Find Research Ideas 1 | 5:34 | bilibili (https://www.bilibili.com/video/BV1qq4y1z7F2/) | |
| 12/8/21 | MAE (https://arxiv.org/pdf/2111.06377.pdf) Paragraph-by-Paragraph Reading | 47:04 | bilibili (https://www.bilibili.com/video/BV1sq4y1q77t/) zhihu (https://www.zhihu.com/zvideo/1452458167968251904) | |
| 11/29/21 | ViT (https://arxiv.org/pdf/2010.11929.pdf) Paragraph-by-Paragraph Reading | 1:11:30 | bilibili (https://www.bilibili.com/video/BV15P4y137jb/) zhihu (https://www.zhihu.com/zvideo/1449195245754380288) | |
| 11/18/21 | BERT (https://arxiv.org/abs/1810.04805) Paragraph-by-Paragraph Reading | 45:49 | bilibili (https://www.bilibili.com/video/BV1PL411M7eQ/) zhihu (https://www.zhihu.com/zvideo/1445340200976785408) | |
| 11/9/21 | GAN (https://papers.nips.cc/paper/2014/file/5ca3e9b122f61f8f06494c97b1afccf3-Paper.pdf) Paragraph-by-Paragraph Reading | 46:16 | bilibili (https://www.bilibili.com/video/BV1rb4y187vD/) zhihu (https://www.zhihu.com/zvideo/1442091389241159681) | |
| 11/3/21 | Zero-Based Multi-Figure Detailed Explanation of Graph Neural Networks (https://distill.pub/2021/gnn-intro/) (GNN/GCN) | 1:06:19 | bilibili (https://www.bilibili.com/video/BV1iT4y1d7zP/) zhihu (https://www.zhihu.com/zvideo/1439540657619087360) | |
| 10/27/21 | Transformer (https://arxiv.org/abs/1706.03762) Paragraph-by-Paragraph Reading (References mentioned in video 1) | 1:27:05 | bilibili (https://www.bilibili.com/video/BV1pu411o7BE/) zhihu (https://www.zhihu.com/zvideo/1437034536677404672) | |
| 10/22/21 | ResNet (https://arxiv.org/abs/1512.03385) Paragraph-by-Paragraph Reading | 53:46 | bilibili (https://www.bilibili.com/video/BV1P3411y7nn/) zhihu (https://www.zhihu.com/zvideo/1434795406001180672) | |
| 10/21/21 | ResNet (https://arxiv.org/abs/1512.03385): The Backbone of Computer Vision | 11:50 | bilibili (https://www.bilibili.com/video/BV1Fb4y1h73E/) zhihu (https://www.zhihu.com/zvideo/1434787226101751808) | |
| 10/15/21 | AlexNet (https://papers.nips.cc/paper/2012/file/c399862d3b9d6b76c8436e924a68c45b-Paper.pdf) Paragraph-by-Paragraph Reading | 55:21 | bilibili (https://www.bilibili.com/video/BV1hq4y157t1/) zhihu (https://www.zhihu.com/zvideo/1432354207483871232) | |
| 10/14/21 | Rereading a Foundational Work of Deep Learning 9 Years Later: AlexNet (https://papers.nips.cc/paper/2012/file/c399862d3b9d6b76c8436e924a68c45b-Paper.pdf) | 19:59 | bilibili (https://www.bilibili.com/video/BV1ih411J7Kz/) zhihu (https://www.zhihu.com/zvideo/1432155856322920448) | |
| 10/06/21 | How to Read a Paper | 06:39 | bilibili (https://www.bilibili.com/video/BV1H44y1t75x/) zhihu (https://www.zhihu.com/zvideo/1428973951632969728) |
1
1 Stanford 200+ page survey with 100+ authors (https://arxiv.org/abs/2108.07258), 2 New research on LayerNorm (https://arxiv.org/pdf/1911.07013.pdf), 3 Research on the role of Attention in Transformers (https://arxiv.org/abs/2103.03404)
All Papers
Includes papers already recorded and those to be introduced later. The selection principle is influential papers in deep learning within the last 10 years (must-read papers), or recent interesting papers. Of course, there are too many important works in these ten years to cover one by one. When selecting, I will lean towards those not covered in previous live classes (https://c.d2l.ai/zh-v2/).
Feel free to provide suggestions (requests) in the discussion area (https://github.com/mli/paper-reading/discussions).
Total papers: 67, Recorded: 32 (Citations here use Semantic Scholar because it provides an API (https://api.semanticscholar.org/api-docs/graph#operation/get_graph_get_paper) to automatically fetch without manual updates.)
Computer Vision - CNN
| Recorded | Year | Name | Description | Citations |
|---|---|---|---|---|
| ✅ | 2012 | AlexNet (https://papers.nips.cc/paper/2012/file/c399862d3b9d6b76c8436e924a68c45b-Paper.pdf) | Foundational work of the deep learning boom | citation (https://www.semanticscholar.org/paper/ImageNet-classification-with-deep-convolutional-Krizhevsky-Sutskever/abd1c342495432171beb7ca8fd9551ef13cbd0ff) |
| 2014 | VGG (https://arxiv.org/pdf/1409.1556.pdf) | Deeper networks using 3x3 convolutions | citation (https://www.semanticscholar.org/paper/Very-Deep-Convolutional-Networks-for-Large-Scale-Simonyan-Zisserman/eb42cf88027de515750f230b23b1a057dc782108) | |
| 2014 | GoogleNet (https://arxiv.org/pdf/1409.4842.pdf) | Deeper networks using parallel architectures | citation (https://www.semanticscholar.org/paper/Going-deeper-with-convolutions-Szegedy-Liu/e15cf50aa89fee8535703b9f9512fca5bfc43327) | |
| ✅ | 2015 | ResNet (https://arxiv.org/pdf/1512.03385.pdf) | Residual connections essential for deep networks. | citation (https://www.semanticscholar.org/paper/Deep-Residual-Learning-for-Image-Recognition-He-Zhang/2c03df8b48bf3fa39054345bafabfeff15bfd11d) |
| 2017 | MobileNet (https://arxiv.org/pdf/1704.04861.pdf) | Small CNN suitable for mobile devices | citation (https://www.semanticscholar.org/paper/MobileNets%3A-Efficient-Convolutional-Neural-Networks-Howard-Zhu/3647d6d0f151dc05626449ee09cc7bce55be497e) | |
| 2019 | EfficientNet (https://arxiv.org/pdf/1905.11946.pdf) | CNN obtained through architecture search | citation (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) | Achieving SOTA on ImageNet with shallow networks | citation (https://www.semanticscholar.org/paper/Non-deep-Networks-Goyal-Bochkovskiy/0d7f6086772079bc3e243b7b375a9ca1a517ba8b) |
Computer Vision - Transformer
| Recorded | Year | Name | Description | Citations |
|---|---|---|---|---|
| ✅ | 2020 | ViT (https://arxiv.org/pdf/2010.11929.pdf) | Transformer enters CV | citation (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) | Hierarchical Vision Transformer | citation (https://www.semanticscholar.org/paper/Swin-Transformer%3A-Hierarchical-Vision-Transformer-Liu-Lin/c8b25fab5608c3e033d34b4483ec47e68ba109b7) |
| 2021 | MLP-Mixer (https://arxiv.org/pdf/2105.01601.pdf) | Replacing self-attention with MLPs | citation (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 version for CV | citation (https://www.semanticscholar.org/paper/Masked-Autoencoders-Are-Scalable-Vision-Learners-He-Chen/c1962a8cf364595ed2838a097e9aa7cd159d3118) |
Generative Models
| Recorded | Year | Name | Description | Citations |
|---|---|---|---|---|
| ✅ | 2014 | GAN (https://papers.nips.cc/paper/2014/file/5ca3e9b122f61f8f06494c97b1afccf3-Paper.pdf) | Pioneering work in generative models | citation (https://www.semanticscholar.org/pap |
Similar Articles
@sheriyuo: AI method for reading papers: Read http://arxiv.org/abs/xxxx.xxxxx, summarize the core idea of this paper
Recommends using AI for reading papers and references the three-pass method from the classic 'How to Read a Paper' technique.
@GitHub_Daily: To dive deep into model research, you can't just stay at the application layer—you need to understand how the underlying system is trained and optimized. I stumbled upon LLMSys-PaperList, a carefully curated collection of papers related to large model systems. It is continuously updated from 2022 to the latest top conference papers in 2026, and organized by categories such as training, inference, multimodality...
A carefully curated collection of papers related to large model systems, covering training, inference, multimodality, and more. It is continuously updated and includes technical reports, frameworks, and courses, making it a valuable reference for researchers and developers.
@Phoenixyin13: Today, I will thoroughly teach you how to read a paper efficiently and with high quality in the AI era. S. Keshav's 'How to Read a Paper' is indeed a classic that has stood the test of time in academia, directly solving the problem with a three-step method. Check and see if you usually read papers like this: First pass, funnel. A…
A tweet introduces how to combine Keshav's three-pass method to efficiently read academic papers in the AI era, emphasizing AI tools for compressing noise, explaining complex concepts, and adversarial training, while reminding readers to maintain critical thinking and academic taste.
@Jolyne_AI: There's an open-source tool on GitHub called daily-arXiv-ai-enhanced that turns 'paper chasing' into a daily automated routine. It fetches the latest arXiv papers daily and uses DeepSeek and other large models to generate Chinese summaries, helping you quickly stay on top of the latest AI research.
daily-arXiv-ai-enhanced is an open-source tool that automatically fetches the latest papers from arXiv daily via GitHub Actions, and uses DeepSeek and other large models to generate Chinese summaries, helping you quickly keep up with AI research progress.
@QingQ77: 'Dive into Deep Learning' is an excellent introductory book, but its update speed struggles to keep pace with the field's development. Since the Transformer, content like CLIP, Diffusion, vLLM, and more has proliferated. While online resources are abundant, they are highly fragmented—today you study Attention, tomorrow LoRA, the day after...
This project is a systematic deep learning notes repository covering PyTorch, Transformers, generative models, and more. It aims to address the fragmentation of learning materials and provides code implementations along with practical guides.