GRPO

{
  "author": null,
  "context": "Group Relative Policy Optimization; standard RL algorithm used in ATLAS stage 2 without modification, enabling compatible training with existing VLM pipelines.",
  "category": "ai",
  "enrichment": {
    "is_stale": true
  },
  "norm_label": "GRPO",
  "source_url": null,
  "graphify_id": "grpo",
  "source_file": "guo-atlas-2026-fulltext.md",
  "imported_from": "/Users/antonborzov/Documents/Research.nosync/papers/extract_typed_out/guo-atlas-2026-fulltext/graph.json",
  "extracted_type": "method",
  "source_location": "§2.2, §2.3",
  "graphify_file_type": "method"
}