{
  "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": 30,
  "pageCount": 188,
  "prevPage": 29,
  "nextPage": 31,
  "slideType": "case_study",
  "function": "illustrate_case",
  "density": "dense",
  "nDataPoints": 9,
  "notes": "The slide highlights the XLand environment and Population Based Training as key enablers for zero-shot generalization in RL.",
  "elementsJson": [
    "headline_text",
    "paragraph",
    "bullet_list",
    "photo",
    "bar_chart_grouped"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5c-fd5012384ae3/30",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5c-fd5012384ae3",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5c-fd5012384ae3.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5c-fd5012384ae3#slide-30",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "DeepMind trained RL agents on 3.4M tasks across a diverse set of 700k games in a 3D simulated environment, and show they can generalize to radically different games without additional training.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d26-0c6c47b373eb",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.25,
        "w": 0.3,
        "x": 0.65,
        "y": 0.65
      },
      "kind": "chart",
      "text": "Test metrics progress during training",
      "attrs": null,
      "subkind": "bar-grouped",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "f8cc3040-8225-41ee-ac3d-8c198ccd8e6c",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.2,
        "w": 0.3,
        "x": 0.65,
        "y": 0.38
      },
      "kind": "image",
      "text": "Examples of XLand environments",
      "attrs": null,
      "subkind": "illustration",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "9d81eb4f-7fd1-4ca0-90c2-f468b7889848",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.45,
        "w": 0.6,
        "x": 0.03,
        "y": 0.38
      },
      "kind": "list",
      "text": "The researchers created XLand, a vast controllable environment, which allows them to dynamically adapt both how the agents train and, crucially, the games on which they train.\nThe distribution of games is learned using a hyperparameter optimization technique called Population Based Training. It allows them to find the games which have the right level of difficulty given the agents’ behaviour. This ensures the agents build evermore general capabilities.\nAs training progresses, the agents exhibit heuristic behaviours such as experimenting, changing the state of the world, and cooperation, which are uncharacteristic of usual RL agents. These learned behaviours allow them to generalize to hand-designed held-out tasks, a first in RL research.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "cd5a22c8-f5e1-437d-b853-9c16068065c3",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.15,
        "w": 0.6,
        "x": 0.03,
        "y": 0.22
      },
      "kind": "paragraph",
      "text": "RL agents have shown impressive performance on challenging individual tasks. But can they generalize to tasks they never trained on? DeepMind trained RL agents on 3.4M tasks across a diverse set of 700k games in a 3D simulated environment, and show they can generalize to radically different games without additional training.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "33f18606-0bc8-4b89-adc2-2c264d11eb82",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.5,
        "x": 0.03,
        "y": 0.14
      },
      "kind": "title",
      "text": "Zero-shot generalisation in reinforcement learning",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "fbfe8510-78d1-4d70-83e1-1638be5dacbe",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Aha! Moment",
      "slug": "aha-moment",
      "agent": "Storyteller",
      "layer": "slide",
      "matchId": "03e42107-b64a-4d1b-b3af-2d3225070c43",
      "evidence": "The slide highlights a significant achievement in zero-shot generalisation in reinforcement learning, likely to surprise or intrigue the audience.",
      "confidence": 0.8
    }
  ],
  "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": 32,
      "from": 28,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3d-4c5c6602de39",
      "evidence": "MuZero, DreamerV2, generalisation, Leela coaching, then statistics rigour critique.",
      "position": 4,
      "objective": "Survey reinforcement learning advances + concerns",
      "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
}