{"id": "card_n_2be7f1c3f6e9", "kind": "note", "title": "Diffusion models — the probability-flow ODE behind generative AI", "body": "Score-based generative models noise data toward a Gaussian and learn to reverse it. The\n'probability-flow ODE' is the deterministic fluid whose time-marginals match the noising SDE — the\nsame continuity equation, velocity set by the learned score ∇log p. It is the engine under modern\nimage generation.", "source": {"label": "Concordance assay — 2026-07-08", "url": "https://narrowhighway.com/s/8a3d0e4b4c8af33851fc97658b7af1f9631e8a770708a245234f2226f076d694", "ref": "diffusion_models", "authority_tier": "engine_derived"}, "shelf": "science", "box": "fluid_probability_dynamics", "bands": ["diffusion models", "probability flow ode", "score matching", "generative ai", "machine learning"], "connections": [{"to_card_id": "card_n_8530c72a2201", "relationship": "instantiates", "origin": "assay_2026_07_08"}], "author": "engine", "created_at": "2026-07-09T21:52:14.887408+00:00", "updated_at": "2026-07-09T21:52:14.887408+00:00", "visibility": "public", "lifecycle_stage": "public", "volatility": "permanent", "surface": "secular", "metrics": {"paperclips_count": 0, "helpful_count": 0, "not_helpful_count": 0, "cite_count": 0, "walks_through_count": 0, "flagged_count": 0}, "source_hash": "ae198e3018f6021870ff9a662cf73d1a6a50c630d855cacaf5452c63199a1dbe"}