{
  "docId": "019dd923-5e88-73ef-bd5c-fd5012384ae3",
  "docSlug": "bf350dd574c19997",
  "documentTitle": "2021 Air Street Capital The State of AI Report 2021",
  "authorId": "AirStreetCapital",
  "authorName": "Air Street Capital",
  "documentKindSlug": "consulting-deck",
  "documentKindLabel": "Consulting deck",
  "sourceTypeSlug": "vc_research",
  "sourceTypeLabel": "VC research",
  "presentationDate": null,
  "orientation": "landscape",
  "aspectRatio": 1.777,
  "pageNumber": 17,
  "pageCount": 188,
  "prevPage": 16,
  "nextPage": 18,
  "slideType": "industry_trends",
  "function": "summarize",
  "density": "overcrowded",
  "nDataPoints": 15,
  "notes": "The slide highlights a shift in research focus back to non-transformer architectures like MLP-Mixer and convolutional models.",
  "elementsJson": [
    "headline_text",
    "paragraph",
    "bullet_list",
    "line_chart"
  ],
  "metadataConfidence": 0.95,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5c-fd5012384ae3/17",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5c-fd5012384ae3",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5c-fd5012384ae3.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5c-fd5012384ae3#slide-17",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "When trained properly, they are competitive with transformers on several NLP and computer vision tasks.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d25-982355b25f4a",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.3,
        "x": 0.65,
        "y": 0.45
      },
      "kind": "chart",
      "text": "Linear 5-shot ImageNet Top-1 [%] vs Training Size",
      "attrs": null,
      "subkind": "line",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "e15f3861-423b-40f6-95f2-5ad35eb272a5",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.3,
        "w": 0.6,
        "x": 0.05,
        "y": 0.35
      },
      "kind": "list",
      "text": "Google researchers set out to disentangle the effects of pre-training and architectural advancements on the performance of language models. They found that pre-training helps CNNs as much as it helps transformers. On 7 out of 8 tasks they consider, they showed that a pre-trained convolutional Seq2Seq outperforms T5, a recent SOTA transformer. However, transformers still have the edge in modeling long-range dependencies.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "8a4607a4-8f08-4085-96c9-ffdc6cb73e40",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.2,
        "w": 0.6,
        "x": 0.05,
        "y": 0.68
      },
      "kind": "list",
      "text": "Other Google researchers proposed MLP-Mixer, an all-MLP architecture for computer vision. Using MLPs for computer vision goes against the conventional wisdom (using CNNs) and recent breakthroughs (Vision Transformers). They show that MLP-Mixer scales well to large datasets and is competitive with SOTA CNNs and ViTs.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "ebf3bac6-769c-4be8-b18e-bf893f3d6c86",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "Linear 5-shot ImageNet Top-1 (%)",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d25-9f165cfa0af6",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.12,
        "w": 0.9,
        "x": 0.05,
        "y": 0.21
      },
      "kind": "paragraph",
      "text": "While pre-trained transformers have taken the ML world by storm, new research shows that convolutional neural networks (CNNs) and multi-layered perceptrons (MLPs) shouldn't be an afterthought. When trained properly, they are competitive with transformers on several NLP and computer vision tasks.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "e32a7693-b07e-405a-98b5-2184fec31818",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.5,
        "x": 0.05,
        "y": 0.14
      },
      "kind": "title",
      "text": "Beyond transformers: MLPs and CNNs make a comeback",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "db3f4565-3bed-45d3-8103-126b54bf8539",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Action Titles",
      "slug": "action-titles",
      "agent": "Architect",
      "layer": "slide",
      "matchId": "019dd95a-0fd5-7148-8ed0-d0010510acd6",
      "evidence": "Title 'MLPs and CNNs make a comeback' frames an unexpected reversal.",
      "confidence": 80
    },
    {
      "name": "AIDA Model",
      "slug": "aida-model",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "433874aa-c928-4117-a61e-fe209f367f9a",
      "evidence": "The slide presents a clear attention-grabbing headline, and lists key points that build up to a conclusion, following the AIDA model structure.",
      "confidence": 0.7
    },
    {
      "name": "Unexpected Pattern",
      "slug": "unexpected-pattern",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "019dd95a-0fd5-7148-8ed0-d6384b39b770",
      "evidence": "Frames pattern break vs the prior transformer narrative.",
      "confidence": 70
    }
  ],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 152,
      "from": 5,
      "beatId": "019dd95a-0682-776c-8e35-0affbadec38e",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Facts (What)",
      "beatSlug": "triple-take-the-facts-what",
      "evidence": "Definitions p.5-6, Exec Summary p.7, then Sections 1-3 catalog evidence.",
      "position": 1,
      "confidence": 78,
      "parentBeatName": "Setup",
      "parentBeatSlug": "setup"
    },
    {
      "to": 152,
      "from": 10,
      "beatId": "019dd95a-0682-776c-8e35-1b3ee3a1c012",
      "arcName": "The Mountain",
      "arcSlug": "mountain",
      "beatName": "Rising Action",
      "beatSlug": "mountain-rising-action",
      "evidence": "Three sections accumulating evidence of accelerating progress.",
      "position": 2,
      "confidence": 55,
      "parentBeatName": "Development",
      "parentBeatSlug": "development"
    }
  ],
  "loops": [
    {
      "to": 18,
      "from": 11,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3d-45052222b756",
      "evidence": "Multiple evidence slides (vision, audio, 3D, MLPs comeback) all converging on transformer dominance.",
      "position": 2,
      "objective": "Show transformers spreading across AI domains",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 82,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}