{
  "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": 37,
  "pageCount": 188,
  "prevPage": 36,
  "nextPage": 38,
  "slideType": "case_study",
  "function": "illustrate_case",
  "density": "dense",
  "nDataPoints": 4,
  "notes": "Includes a technical diagram of the contrastive learning architecture and the loss function formula.",
  "elementsJson": [
    "paragraph",
    "bullet_list",
    "flowchart"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5c-fd5012384ae3/37",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5c-fd5012384ae3",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5c-fd5012384ae3.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5c-fd5012384ae3#slide-37",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "ConVIRT outperforms all ImageNet-initialized models with only 10% as much labeled training data.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d27-83449a871109",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.22,
        "w": 0.65,
        "x": 0.18,
        "y": 0.72
      },
      "kind": "diagram",
      "text": "ConVIRT architecture diagram showing image and text encoders, contrastive loss calculation",
      "attrs": null,
      "subkind": "process",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "d0f4144b-8071-4e84-81ea-c1ea85cd638a",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.12,
        "w": 0.96,
        "x": 0.02,
        "y": 0.41
      },
      "kind": "list",
      "text": "During contrastive pre-training, the model learns to associate each image in a batch with its text companion, while dissociating it from the other text snippets. To have better learned representations, ConVIRT makes the task harder by using random transformations of the images and texts.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "124341ca-4b31-45f8-8111-ea1633d82a7b",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.18,
        "w": 0.96,
        "x": 0.02,
        "y": 0.21
      },
      "kind": "list",
      "text": "The canonical approach to applying deep computer vision to medical images is fine-tuning ImageNet pre-trained models or using rule-based label extraction from medical textual reports. In contrast, the ConVIRT method pre-trains directly on naturally occurring image-text pairs using a contrastive objective, without any supervision. ConVIRT outperforms all ImageNet-initialized models with only 10% as much labeled training data.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "268f19c0-4e81-48aa-bbdf-49ddd9799879",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.12,
        "w": 0.96,
        "x": 0.02,
        "y": 0.55
      },
      "kind": "list",
      "text": "ConVIRT was tested on 4 datasets spanning 4 different classification tasks: binary, multi-label binary, multi-class and anomaly detection. In 3 out of the 4 tasks, ConVIRT with only 1% training data achieved better classification results than ImageNet initialized models which used 100% training data.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "29e0ebe4-8397-4bda-9902-adec01ab27c6",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "labeled training data: 1%",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d27-8616a380e3ec",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.04,
        "w": 0.6,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "Learning medical image representations from text-image pairings",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "7a588fa9-d2d8-4828-8ea8-4f4a7a8a397d",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [],
  "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": 40,
      "from": 36,
      "name": "Aha Moment",
      "slug": "03-aha-moment",
      "bestFor": "Data-heavy sections, research findings, analytical arguments",
      "matchId": "019dd95a-07fe-70ce-8d3d-53192fc75199",
      "evidence": "Diffusion models > GANs, then CLIP, then DALL-E payoff.",
      "position": 5,
      "objective": "Reveal multimodal generative AI breakthrough",
      "structure": "The Problem/Question -> What the Data Shows -> The Insight",
      "confidence": 72,
      "description": "Establish a tension, present data, then deliver the insight that changes everything"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}