{
  "source_file": "0930",
  "source_platform": "个人博客",
  "author": "Christopher Olah / colah's blog",
  "title_listed": "Multimodal Neurons",
  "keywords": "多模态神经元,CLIP,可解释性",
  "domain": "图像处理",
  "content_type": "深度技术解析",
  "is_paid_flag": "否",
  "url": "https://distill.pub/2021/multimodal-neurons/",
  "publish_date": "2021-03",
  "corpus_stage": "预训练/SFT",
  "notes": "CLIP多模态神经元分析（Distill）",
  "id": "9ca46bca473e",
  "collected_at": "2026-10-02T14:28:20",
  "title": "Multimodal Neurons in Artificial Neural Networks",
  "body_markdown": "Distill\nMultimodal Neurons in Artificial Neural Networks\nAuthors\nAffiliations\nGabriel Goh\nOpenAI\nNick Cammarata †\nOpenAI\nChelsea Voss †\nOpenAI\nShan Carter\nObservable\nMichael Petrov\nOpenAI\nLudwig Schubert\nAlec Radford\nOpenAI\nChris Olah\nPublished\nMarch 4, 2021\nDOI\n10.23915/distill.00030\nAcknowledgments\nWe are deeply grateful to Sandhini Agarwal, Daniela Amodei, Dario Amodei,\n      Tom Brown, Jeff Clune, Steve Dowling, Gretchen Krueger, Brice Menard,\n      Reiichiro Nakano, Aditya Ramesh, Pranav Shyam, Ilya Sutskever and Martin\n      Wattenberg.\nAuthor Contributions\nGabriel Goh:\nResearch lead. Gabriel Goh first discovered multimodal\n      neurons, sketched out the project direction and paper outline, and did\n      much of the conceptual and engineering work that allowed the team to\n      investigate the models in a scalable way. This included developing tools\n      for understanding how concepts were built up and decomposed (that were\n      applied to emotion neurons), developing zero-shot neuron search (that\n      allowed easy discoverability of neurons), and working with Michael Petrov\n      on porting CLIP to microscope. Subsequently developed faceted feature\n      visualization, and text feature visualization.\nChris Olah:\nWorked with Gabe on the overall framing of the article,\n      actively mentored each member of the team through their work providing\n      both high and low level contributions to their sections, and contributed\n      to the text of much of the article, setting the stylistic tone. He worked\n      with Gabe on understanding the neuroscience literature and better\n      understanding the relevant neuroscience literature. Additionally, he wrote\n      the sections on region neurons and developed diversity feature\n      visualization which Gabe used to create faceted feature visualization\nAlec Radford:\nDeveloped CLIP. First observed that CLIP was learning\n      to read. Advised Gabriel Goh on project direction on a weekly basis. Upon\n      the discovery that CLIP was using text to classify images, proposed\n      typographical adversarial attacks as a promising research direction.\nShan Carter:\nWorked on initial investigation of CLIP with Gabriel\n      Goh. Did multimodal activation atlases to understand the space of\n      multimodal representations and geometry, and neuron atlases, which\n      potentially helped the arrangement and display of neurons. Provided much\n      useful advice on the visual presentation of ideas, and helped with many\n      aspects of visual design.\nMichael Petrov:\nWorked on the initial investigation of multimodal\n      neurons by implementing and scaling dataset examples. Discovered, with\n      Gabriel Goh, the original “Spider-Man” multimodal neuron in the dataset\n      examples, and many more multimodal neurons. Assisted a lot in the\n      engineering of Microscope both early on, and at the end, including helping\n      Gabriel Goh with the difficult technical challenges of porting microscope\n      to a different backend.\nChelsea Voss†:\nPerformed investigation of the typographical attacks\n      phenomena, both via linear probes and zero-shot, confirming that the\n      attacks were indeed real and state of the art. Proposed and successfully\n      found “in-the-wild” attacks in the zero-shot classifier. Subsequently\n      wrote the section “typographical attacks”. Upon completion of this part of\n      the project, investigated responses of neurons to rendered text on\n      dictionary words. Also assisted with the organization of neurons into\n      neuron cards.\nNick Cammarata†:\nDrew the connection between multimodal neurons in\n      neural networks and multimodal neurons in the brain, which became the\n      overall framing of the article. Created the conditional probability plots\n      (regional, Trump, mental health), labeling more than 1500 images,\n      discovered that negative pre-ReLU activations are often interpretable, and\n      discovered that neurons sometimes contain a distinct regime change between\n      medium and strong activations. Wrote the identity section and the emotion\n      sections, building off Gabriel’s discovery of emotion neurons and\n      discovering that “complex” emotions can be broken down into simpler ones.\n      Edited the overall text of the article and built infrastructure allowing\n      the team to collaborate in Markdown with embeddable components.\nLudwig Schubert:\nHelped with general infrastructure.\n† equal contributors\nDiscussion and Review\nReview 1 - Anonymous\nReview 2 - Anonymous\nReview 3 - Anonymous\nReferences\nInvariant visual representation by single neurons in the human brain\n[PDF]\nQuiroga, R.Q., Reddy, L., Kreiman, G., Koch, C. and Fried, I., 2005. Nature, Vol 435(7045), pp. 1102--1107. Nature Publishing Group.\nExplicit encoding of multimodal percepts by single neurons in the human brain\nQuiroga, R.Q., Kraskov, A., Koch, C. and Fried, I., 2009. Current Biology, Vol 19(15), pp. 1308--1313. Elsevier.\nLearning Transferable Visual Models From Natural Language Supervision\n[link]\nRadford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G. and Sutskever, I., 2021.\nDeep Residual Learning for Image Recognition\n[PDF]\nHe, K., Zhang, X., Ren, S. and Sun, J., 2015. CoRR, Vol abs/1512.03385.\nAttention is all you need\nVaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L. and Polosukhin, I., 2017. Advances in neural information processing systems, pp. 5998--6008.\nImproved deep metric learning with multi-class n-pair loss objective\nSohn, K., 2016. Advances in neural information processing systems, pp. 1857--1865.\nContrastive multiview coding\nTian, Y., Krishnan, D. and Isola, P., 2019. arXiv preprint arXiv:1906.05849.\nLinear algebraic structure of word senses, with applications to polysemy\nArora, S., Li, Y., Liang, Y., Ma, T. and Risteski, A., 2018. Transactions of the Association for Computational Linguistics, Vol 6, pp. 483--495. MIT Press.\nVisualizing and understanding recurrent networks\n[PDF]\nKarpathy, A., Johnson, J. and Fei-Fei, L., 2015. arXiv preprint arXiv:1506.02078.\nObject detectors emerge in deep scene cnns\n[PDF]\nZhou, B., Khosla, A., Lapedriza, A., Oliva, A. and Torralba, A., 2014. arXiv preprint arXiv:1412.6856.\nNetwork Dissection: Quantifying Interpretability of Deep Visual Representations\n[PDF]\nBau, D., Zhou, B., Khosla, A., Oliva, A. and Torralba, A., 2017. Computer Vision and Pattern Recognition.\nZoom In: An Introduction to Circuits\nOlah, C., Cammarata, N., Schubert, L., Goh, G., Petrov, M. and Carter, S., 2020. Distill, Vol 5(3), pp. e00024--001.\nMultifaceted feature visualization: Uncovering the different types of features learned by each neuron in deep neural networks\n[PDF]\nNguyen, A., Yosinski, J. and Clune, J., 2016. arXiv preprint arXiv:1602.03616.\nSparse but not ‘grandmother-cell’ coding in the medial temporal lobe\nQuiroga, R.Q., Kreiman, G., Koch, C. and Fried, I., 2008. Trends in cognitive sciences, Vol 12(3), pp. 87--91. Elsevier.\nConcept cells: the building blocks of declarative memory functions\nQuiroga, R.Q., 2012. Nature Reviews Neuroscience, Vol 13(8), pp. 587--597. Nature Publishing Group.\nEmotional expressions reconsidered: Challenges to inferring emotion from human facial movements\nBarrett, L.F., Adolphs, R., Marsella, S., Martinez, A.M. and Pollak, S.D., 2019. Psychological science in the public interest, Vol 20(1), pp. 1--68. Sage Publications Sage CA: Los Angeles, CA.\nGeographical evaluation of word embeddings\n[PDF]\nKonkol, M., Brychc{\\'\\i}n, T., Nykl, M. and Hercig, T., 2017. Proceedings of the Eighth International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pp. 224--232.\nUsing Artificial Intelligence to Augment Human Intelligence\n[link]\nCarter, S. and Nielsen, M., 2017. Distill.\nDOI: 10.23915/distill.00009\nVisualizing Representations: Deep Learning and Human Beings\n[link]\nOlah, C., 2015.\nNatural language processing (almost) from scratch\nCollobert, R., Weston, J., Bottou, L., Karlen, M., Kavukcuoglu, K. and Kuksa, P., 2011. Journal of machine learning research, Vol 12(ARTICLE), pp. 2493--2537.\nLinguistic regularities in continuous space word representations\nMikolov, T., Yih, W. and Zweig, G., 2013. Proceedings of the 2013 conference of the north american chapter of the association for computational linguistics: Human language technologies, pp. 746--751.\nMan is to computer programmer as woman is to homemaker? debiasing word embeddings\nBolukbasi, T., Chang, K., Zou, J.Y., Saligrama, V. and Kalai, A.T., 2016. Advances in neural information processing systems, pp. 4349--4357.\nIntriguing properties of neural networks\n[PDF]\nSzegedy, C., Zaremba, W., Sutskever, I., Bruna, J., Erhan, D., Goodfellow, I. and Fergus, R., 2013. arXiv preprint arXiv:1312.6199.\nVisualizing higher-layer features of a deep network\n[PDF]\nErhan, D., Bengio, Y., Courville, A. and Vincent, P., 2009. University of Montreal, Vol 1341, pp. 3.\nFeature Visualization\n[link]\nOlah, C., Mordvintsev, A. and Schubert, L., 2017. Distill.\nDOI: 10.23915/distill.00007\nHow does the brain solve visual object recognition?\nDiCarlo, J.J., Zoccolan, D. and Rust, N.C., 2012. Neuron, Vol 73(3), pp. 415--434. Elsevier.\nImagenet: A large-scale hierarchical image database\nDeng, J., Dong, W., Socher, R., Li, L., Li, K. and Fei-Fei, L., 2009. 2009 IEEE conference on computer vision and pattern recognition, pp. 248--255.\nBREEDS: Benchmarks for Subpopulation Shift\nSanturkar, S., Tsipras, D. and Madry, A., 2020. arXiv preprint arXiv:2008.04859.\nGlobal Weighted Average Pooling Bridges Pixel-level Localization and Image-level Classification\n[PDF]\nQiu, S., 2018. CoRR, Vol abs/1809.08264.\nSeparating style and content with bilinear models\nTenenbaum, J.B. and Freeman, W.T., 2000. Neural computation, Vol 12(6), pp. 1247--1283. MIT Press.\nThe feeling wheel: A tool for expanding awareness of emotions and increasing spontaneity and intimacy\nWillcox, G., 1982. Transactional Analysis Journal, Vol 12(4), pp. 274--276. SAGE Publications Sage CA: Los Angeles, CA.\nActivation atlas\nCarter, S., Armstrong, Z., Schubert, L., Johnson, I. and Olah, C., 2019. Distill, Vol 4(3), pp. e15.\nAdversarial Patch\n[PDF]\nBrown, T., Mané, D., Roy, A., Abadi, M. and Gilmer, J., 2017. arXiv preprint arXiv:1712.09665.\nSynthesizing Robust Adversarial Examples\n[PDF]\nAthalye, A., Engstrom, L., Ilyas, A. and Kwok, K., 2017. arXiv preprint arXiv:1707.07397.\nStudies of interference in serial verbal reactions.\nStroop, J.R., 1935. Journal of experimental psychology, Vol 18(6), pp. 643. Psychological Review Company.\nCurve Detectors\nCammarata, N., Goh, G., Carter, S., Schubert, L., Petrov, M. and Olah, C., 2020. Distill, Vol 5(6), pp. e00024--003.\nAn overview of early vision in inceptionv1\nOlah, C., Cammarata, N., Schubert, L., Goh, G., Petrov, M. and Carter, S., 2020. Distill, Vol 5(4), pp. e00024--002.\nDeep inside convolutional networks: Visualising image classification models and saliency maps\n[PDF]\nSimonyan, K., Vedaldi, A. and Zisserman, A., 2013. arXiv preprint arXiv:1312.6034.\nDeep neural networks are easily fooled: High confidence predictions for unrecognizable images\n[PDF]\nNguyen, A., Yosinski, J. and Clune, J., 2015. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 427--436.\nDOI: 10.1109/cvpr.2015.7298640\nInceptionism: Going deeper into neural networks\n[HTML]\nMordvintsev, A., Olah, C. and Tyka, M., 2015. Google Research Blog.\nPlug & play generative networks: Conditional iterative generation of images in latent space\n[PDF]\nNguyen, A., Clune, J., Bengio, Y., Dosovitskiy, A. and Yosinski, J., 2016. arXiv preprint arXiv:1612.00005.\nSun database: Large-scale scene recognition from abbey to zoo\nXiao, J., Hays, J., Ehinger, K.A., Oliva, A. and Torralba, A., 2010. 2010 IEEE computer society conference on computer vision and pattern recognition, pp. 3485--3492.\nThe pascal visual object classes (voc) challenge\nEveringham, M., Van Gool, L., Williams, C.K., Winn, J. and Zisserman, A., 2010. International journal of computer vision, Vol 88(2), pp. 303--338. Springer.\nFairface: Face attribute dataset for balanced race, gender, and age\nKärkkäinen, K. and Joo, J., 2019. arXiv preprint arXiv:1908.04913.\nA style-based generator architecture for generative adversarial networks\nKarras, T., Laine, S. and Aila, T., 2019. Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 4401--4410.\nUpdates and Corrections\nIf you see mistakes or want to suggest changes, please\ncreate an issue on GitHub\n.\nReuse\nDiagrams and text are licensed under Creative Commons Attribution\nCC-BY 4.0\nwith the\nsource available on GitHub\n, unless noted otherwise. The figures that have been reused from other sources don’t fall under this license and can be recognized by a note in their caption: “Figure from …”.\nCitation\nFor attribution in academic contexts, please cite this work as\nGoh, et al., \"Multimodal Neurons in Artificial Neural Networks\", Distill, 2021.\nBibTeX citation\n@article{goh2021multimodal,\n  author = {Goh, Gabriel and †, Nick Cammarata and †, Chelsea Voss and Carter, Shan and Petrov, Michael and Schubert, Ludwig and Radford, Alec and Olah, Chris},\n  title = {Multimodal Neurons in Artificial Neural Networks},\n  journal = {Distill},\n  year = {2021},\n  note = {https://distill.pub/2021/multimodal-neurons},\n  doi = {10.23915/distill.00030}\n}\nDistill\nis dedicated to clear explanations of machine learning\nAbout\nSubmit\nPrize\nArchive\nRSS\nGitHub\nTwitter\nISSN 2476-0757",
  "metadata": {
    "title": "Multimodal Neurons in Artificial Neural Networks",
    "description": null,
    "keywords": null,
    "author": null,
    "og:type": "article",
    "og:title": "Multimodal Neurons in Artificial Neural Networks",
    "og:description": "We report the existence of multimodal neurons in artificial neural networks, similar to those found in the human brain.",
    "og:url": "https://distill.pub/2021/multimodal-neurons",
    "og:image": "https://distill.pub/2021/multimodal-neurons/thumbnail.jpg",
    "og:locale": "en_US",
    "og:site_name": "Distill",
    "twitter:card": "summary_large_image",
    "twitter:title": "Multimodal Neurons in Artificial Neural Networks",
    "twitter:description": "We report the existence of multimodal neurons in artificial neural networks, similar to those found in the human brain.",
    "twitter:url": "https://distill.pub/2021/multimodal-neurons",
    "twitter:image": "https://distill.pub/2021/multimodal-neurons/thumbnail.jpg",
    "twitter:image:width": "560",
    "twitter:image:height": "295",
    "article:published": "2021-03-04",
    "article:created": "2021-03-04",
    "article:modified": "2021-03-22T15:04:13.000Z",
    "article:author": "Chris Olah"
  },
  "paywall": {
    "listed_paid": false,
    "detected": false,
    "marker": "",
    "public_only": false,
    "policy": "付费/会员内容只保留页面公开可见部分，不绕过付费墙"
  },
  "crawl": {
    "success": true,
    "status_code": 200,
    "error": ""
  },
  "code_blocks": [
    {
      "file": "attachments/9ca46bca473e/code/body-000.txt",
      "chars": 79,
      "language": "text"
    },
    {
      "file": "attachments/9ca46bca473e/code/body-001.txt",
      "chars": 381,
      "language": "text"
    }
  ],
  "images": [],
  "audio": [],
  "videos": [],
  "links": [
    {
      "href": "https://distill.pub/",
      "label": "Distill",
      "type": "a"
    },
    {
      "href": "http://gabgoh.github.io/",
      "label": "Gabriel Goh",
      "type": "a"
    },
    {
      "href": "http://openai.com/",
      "label": "OpenAI",
      "type": "a"
    },
    {
      "href": "http://nickcammarata.com/",
      "label": "Nick Cammarata †",
      "type": "a"
    },
    {
      "href": "https://csvoss.com/",
      "label": "Chelsea Voss †",
      "type": "a"
    },
    {
      "href": "http://shancarter.com/",
      "label": "Shan Carter",
      "type": "a"
    },
    {
      "href": "https://observablehq.com/",
      "label": "Observable",
      "type": "a"
    },
    {
      "href": "http://michaelpetrov.com/",
      "label": "Michael Petrov",
      "type": "a"
    },
    {
      "href": "https://schubert.io/",
      "label": "Ludwig Schubert",
      "type": "a"
    },
    {
      "href": "http://newmu.github.io/",
      "label": "Alec Radford",
      "type": "a"
    },
    {
      "href": "https://colah.github.io/",
      "label": "Chris Olah",
      "type": "a"
    },
    {
      "href": "https://doi.org/10.23915/distill.00030",
      "label": "10.23915/distill.00030",
      "type": "a"
    },
    {
      "href": "https://github.com/distillpub/post--multimodal/issues/1",
      "label": "Review 1 - Anonymous",
      "type": "a"
    },
    {
      "href": "https://github.com/distillpub/post--multimodal/issues/2",
      "label": "Review 2 - Anonymous",
      "type": "a"
    },
    {
      "href": "https://github.com/distillpub/post--multimodal/issues/3",
      "label": "Review 3 - Anonymous",
      "type": "a"
    },
    {
      "href": "http://amygdala.psychdept.arizona.edu/IntroData/Readings/week5/Quiroga-reddy-kreiman-koch-Fried+invariant-visual-single-neurons-human+Nature+2005.pdf",
      "label": "[PDF]",
      "type": "a"
    },
    {
      "href": "https://blog.openai.com/clip",
      "label": "[link]",
      "type": "a"
    },
    {
      "href": "http://arxiv.org/pdf/1512.03385.pdf",
      "label": "[PDF]",
      "type": "a"
    },
    {
      "href": "https://arxiv.org/pdf/1506.02078.pdf",
      "label": "[PDF]",
      "type": "a"
    },
    {
      "href": "https://arxiv.org/pdf/1412.6856.pdf",
      "label": "[PDF]",
      "type": "a"
    },
    {
      "href": "https://arxiv.org/pdf/1704.05796.pdf",
      "label": "[PDF]",
      "type": "a"
    },
    {
      "href": "https://arxiv.org/pdf/1602.03616.pdf",
      "label": "[PDF]",
      "type": "a"
    },
    {
      "href": "https://www.aclweb.org/anthology/I17-1023.pdf",
      "label": "[PDF]",
      "type": "a"
    },
    {
      "href": "https://distill.pub/2017/aia/",
      "label": "[link]",
      "type": "a"
    },
    {
      "href": "https://doi.org/10.23915/distill.00009",
      "label": "DOI: 10.23915/distill.00009",
      "type": "a"
    },
    {
      "href": "http://colah.github.io/posts/2015-01-Visualizing-Representations/",
      "label": "[link]",
      "type": "a"
    },
    {
      "href": "https://arxiv.org/pdf/1312.6199.pdf",
      "label": "[PDF]",
      "type": "a"
    },
    {
      "href": "https://www.researchgate.net/profile/Aaron_Courville/publication/265022827_Visualizing_Higher-Layer_Features_of_a_Deep_Network/links/53ff82b00cf24c81027da530.pdf",
      "label": "[PDF]",
      "type": "a"
    },
    {
      "href": "https://distill.pub/2017/feature-visualization",
      "label": "[link]",
      "type": "a"
    },
    {
      "href": "https://doi.org/10.23915/distill.00007",
      "label": "DOI: 10.23915/distill.00007",
      "type": "a"
    },
    {
      "href": "http://arxiv.org/pdf/1809.08264.pdf",
      "label": "[PDF]",
      "type": "a"
    },
    {
      "href": "https://arxiv.org/pdf/1712.09665.pdf",
      "label": "[PDF]",
      "type": "a"
    },
    {
      "href": "https://arxiv.org/pdf/1707.07397.pdf",
      "label": "[PDF]",
      "type": "a"
    },
    {
      "href": "https://arxiv.org/pdf/1312.6034.pdf",
      "label": "[PDF]",
      "type": "a"
    },
    {
      "href": "https://arxiv.org/pdf/1412.1897.pdf",
      "label": "[PDF]",
      "type": "a"
    },
    {
      "href": "https://doi.org/10.1109/cvpr.2015.7298640",
      "label": "DOI: 10.1109/cvpr.2015.7298640",
      "type": "a"
    },
    {
      "href": "https://research.googleblog.com/2015/06/inceptionism-going-deeper-into-neural.html",
      "label": "[HTML]",
      "type": "a"
    },
    {
      "href": "https://arxiv.org/pdf/1612.00005.pdf",
      "label": "[PDF]",
      "type": "a"
    },
    {
      "href": "https://github.com/distillpub/post--multimodal/issues/new",
      "label": "create an issue on GitHub",
      "type": "a"
    },
    {
      "href": "https://creativecommons.org/licenses/by/4.0/",
      "label": "CC-BY 4.0",
      "type": "a"
    },
    {
      "href": "https://github.com/distillpub/post--multimodal",
      "label": "source available on GitHub",
      "type": "a"
    },
    {
      "href": "https://distill.pub/about/",
      "label": "About",
      "type": "a"
    },
    {
      "href": "https://distill.pub/journal/",
      "label": "Submit",
      "type": "a"
    },
    {
      "href": "https://distill.pub/prize/",
      "label": "Prize",
      "type": "a"
    },
    {
      "href": "https://distill.pub/archive/",
      "label": "Archive",
      "type": "a"
    },
    {
      "href": "https://distill.pub/rss.xml",
      "label": "RSS",
      "type": "a"
    },
    {
      "href": "https://github.com/distillpub",
      "label": "GitHub",
      "type": "a"
    },
    {
      "href": "https://twitter.com/distillpub",
      "label": "Twitter",
      "type": "a"
    }
  ],
  "repair_version": 2,
  "repaired_at": "2026-10-14T10:52:36Z",
  "source_url": "https://distill.pub/2021/multimodal-neurons/",
  "papers": [
    {
      "status": "download_error",
      "url": "http://amygdala.psychdept.arizona.edu/IntroData/Readings/week5/Quiroga-reddy-kreiman-koch-Fried+invariant-visual-single-neurons-human+Nature+2005.pdf",
      "reason": "ConnectTimeout: HTTPConnectionPool(host='amygdala.psychdept.arizona.edu', port=80): Max retries exceeded with url: /IntroData/Readings/week5/Quiroga-reddy-kreiman-koch-Fried+invariant-visual-singl",
      "retried_at": "2026-10-20T22:24:13Z"
    },
    {
      "status": "ok",
      "url": "http://arxiv.org/pdf/1512.03385.pdf",
      "bytes": 819383,
      "reused": false,
      "local_path": "papers/repair/b02a0eb4ab960d80014f.pdf"
    },
    {
      "status": "ok",
      "url": "https://arxiv.org/pdf/1506.02078.pdf",
      "bytes": 2986762,
      "reused": false,
      "local_path": "papers/repair/62afc74dbc8d04e2b22f.pdf"
    },
    {
      "status": "ok",
      "url": "https://arxiv.org/pdf/1412.6856.pdf",
      "bytes": 7369966,
      "reused": false,
      "local_path": "papers/repair/341420a67157cd804005.pdf"
    },
    {
      "status": "ok",
      "url": "https://arxiv.org/pdf/1704.05796.pdf",
      "bytes": 3329116,
      "reused": false,
      "local_path": "papers/repair/fa291157c0972073146e.pdf"
    },
    {
      "status": "ok",
      "url": "https://arxiv.org/pdf/1602.03616.pdf",
      "bytes": 10374723,
      "reused": true,
      "local_path": "papers/repair/d18f7c8f01e3fa11cf5c.pdf"
    },
    {
      "status": "ok",
      "url": "https://www.aclweb.org/anthology/I17-1023.pdf",
      "bytes": 309867,
      "reused": false,
      "local_path": "papers/repair/ce64bcdbdb5e786e17d2.pdf"
    },
    {
      "status": "ok",
      "url": "https://arxiv.org/pdf/1312.6199.pdf",
      "bytes": 6565323,
      "reused": true,
      "local_path": "papers/repair/67aa78d89bf9969f80c2.pdf"
    },
    {
      "status": "http_403",
      "url": "https://www.researchgate.net/profile/Aaron_Courville/publication/265022827_Visualizing_Higher-Layer_Features_of_a_Deep_Network/links/53ff82b00cf24c81027da530.pdf",
      "retried_at": "2026-10-20T17:40:54Z"
    },
    {
      "status": "ok",
      "url": "http://arxiv.org/pdf/1809.08264.pdf",
      "bytes": 1734044,
      "reused": false,
      "local_path": "papers/repair/f218fb013e37f2a34530.pdf"
    },
    {
      "status": "ok",
      "url": "https://arxiv.org/pdf/1712.09665.pdf",
      "bytes": 1720993,
      "reused": false,
      "local_path": "papers/repair/a6d341b15c5090ebd080.pdf"
    },
    {
      "status": "ok",
      "url": "https://arxiv.org/pdf/1707.07397.pdf",
      "bytes": 4557212,
      "reused": false,
      "local_path": "papers/repair/bdf496ac999ecf71b370.pdf"
    },
    {
      "status": "ok",
      "url": "https://arxiv.org/pdf/1312.6034.pdf",
      "bytes": 2262025,
      "reused": true,
      "local_path": "papers/repair/5c375055da17ead61604.pdf"
    },
    {
      "status": "ok",
      "url": "https://arxiv.org/pdf/1412.1897.pdf",
      "bytes": 9985569,
      "reused": true,
      "local_path": "papers/repair/a11d2cb82fb90198cc81.pdf"
    },
    {
      "status": "ok",
      "url": "https://arxiv.org/pdf/1612.00005.pdf",
      "bytes": 25726389,
      "reused": true,
      "local_path": "papers/repair/b188d1a5bdcb285fbb45.pdf"
    }
  ],
  "actual_video_files": [],
  "body_chars": 13566,
  "expected_body_images": 0,
  "discovery_status": "ok",
  "delivery_status": "partial",
  "missing_reason": "download_error；http_403",
  "raw_html": "attachments/9ca46bca473e/pages/source.html",
  "local_html": "attachments/9ca46bca473e/pages/article.html",
  "body_base_path": "attachments/9ca46bca473e/pages/",
  "body_markdown_path": "attachments/9ca46bca473e/pages/article.md"
}