{
  "docId": "019dd923-5e88-73ef-bd5c-f6fdd8f57895",
  "docSlug": "3a892393f72c1ff5",
  "documentTitle": "2019 Air Street Capital The State of AI Report 2019",
  "authorId": "AirStreetCapital",
  "authorName": "Nathan Benaich and Ian Hogarth",
  "documentKindSlug": "consulting-deck",
  "documentKindLabel": "Consulting deck",
  "sourceTypeSlug": "vc_research",
  "sourceTypeLabel": "VC research",
  "presentationDate": null,
  "orientation": "landscape",
  "aspectRatio": 1.777,
  "pageNumber": 35,
  "pageCount": 136,
  "prevPage": 34,
  "nextPage": 36,
  "slideType": "industry_trends",
  "function": "analyze_data",
  "density": "overcrowded",
  "nDataPoints": 40,
  "notes": "Slide contains two tables comparing various neural network models across metrics like parameters, latency, and accuracy.",
  "elementsJson": [
    "headline_text",
    "action_title",
    "bullet_list",
    "data_table"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5c-f6fdd8f57895/35",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5c-f6fdd8f57895",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5c-f6fdd8f57895.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5c-f6fdd8f57895#slide-35",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "The system reaches 74.0% top-1 accuracy with 76ms latency on a Pixel phone, which is 1.5x faster than MobileNetV2.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-41ab-72d9-bbc4-eaa8665a9745",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.08,
        "w": 0.5,
        "x": 0.48,
        "y": 0.88
      },
      "kind": "disclaimer",
      "text": "Table 3. ImageNet classification performance compared with baselines. For baseline models, we directly cite the parameter size, FLOP count and top-1 accuracy on the ImageNet validation set from their original papers.",
      "attrs": null,
      "subkind": null,
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "6ba1cdaa-9d00-475a-bd9b-19cd93d76560",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.45,
        "x": 0.02,
        "y": 0.28
      },
      "kind": "list",
      "text": "Google demonstrated a multi-objective RL-based approach (MnasNet) to find high accuracy CNN models with low real-world inference latency as measured on the Google Pixel platform. The system reaches 74.0% top-1 accuracy with 76ms latency on a Pixel phone, which is 1.5x faster than MobileNetV2.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "6b6b176d-5e8d-4b44-9276-c8354dc2e13c",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.45,
        "x": 0.02,
        "y": 0.62
      },
      "kind": "list",
      "text": "Facebook proposed a differentiable neural architecture search (DNAS) framework that uses gradient-based methods to optimize CNN architectures over a layer-wise search space. FBNet-B achieves the same top-1 accuracy than MnasNet but with 23.1 ms latency and 420x smaller search cost.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "8b7a9736-1fb7-4d9d-b611-8c8460e82547",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "CPU Latency: 76ms",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-41ab-72d9-bbc4-ed99591e2b18",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.5,
        "x": 0.48,
        "y": 0.28
      },
      "kind": "table",
      "text": "Table 1: Model performance comparison (Parameters, Latency, Accuracy)",
      "attrs": null,
      "subkind": "data",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "00f1c00e-32e9-4703-9b1e-e68d639fa80f",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.35,
        "w": 0.5,
        "x": 0.48,
        "y": 0.55
      },
      "kind": "table",
      "text": "Table 2: Model performance comparison (Search cost, Latency, Accuracy)",
      "attrs": null,
      "subkind": "data",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "92508734-ef79-4c03-9b0b-baf576c301fb",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.03,
        "w": 0.96,
        "x": 0.02,
        "y": 0.22
      },
      "kind": "title",
      "text": "The pace of CNN-based automated architecture search is accelerating: Facebook ups ante vs. Google.",
      "attrs": null,
      "subkind": "action-title",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "a20e9e60-abd0-42cd-a73e-a841ec4b0291",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.04,
        "w": 0.96,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "AutoML: Designing resource-constrained networks with real device performance feedback",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "3e9eab85-ceb0-4eb0-9528-541ee381f3c5",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Action Titles",
      "slug": "action-titles",
      "agent": "Architect",
      "layer": "slide",
      "matchId": "019dd95a-0fd5-7148-8ecc-f57a7158c09f",
      "evidence": "Title: 'Facebook ups ante vs. Google' on NAS.",
      "confidence": 80
    }
  ],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 92,
      "from": 5,
      "beatId": "019dd95a-0682-776c-8e34-ce57e5bbe574",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Facts (What)",
      "beatSlug": "triple-take-the-facts-what",
      "evidence": "Definitions through Research/Talent/Industry — fact-dense case studies and benchmarks.",
      "position": 1,
      "confidence": 70,
      "parentBeatName": "Setup",
      "parentBeatSlug": "setup"
    },
    {
      "to": 92,
      "from": 9,
      "beatId": "019dd95a-0682-776c-8e34-ddbccb391af1",
      "arcName": "Voyage and Return",
      "arcSlug": "voyage-return",
      "beatName": "The Unknown",
      "beatSlug": "voyage-return-the-unknown",
      "evidence": "Research/Talent/Industry frontiers explored.",
      "position": 2,
      "confidence": 45,
      "parentBeatName": "Complication",
      "parentBeatSlug": "complication"
    }
  ],
  "loops": [
    {
      "to": 39,
      "from": 34,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3c-aa82735727fd",
      "evidence": "Sequence of 'state of the art' generative breakthroughs across modalities.",
      "position": 6,
      "objective": "Trace generative-model progress: AutoML, GANs, voice, 3D",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 70,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}