Definition
Plain language
A diagram of which facts depend on which others that's guaranteed not to silently drop any dependency.
As stated in the literature
Independence map — a Bayesian-network graph whose conditional independencies are a subset of the underlying distribution's, ensuring no encoded dependency is lost during factorization.
Why it matters: Without the I-map property, a graphical model can fool you into assuming variables are independent when they aren't, breaking any inference built on top.
For example, if X and Y truly depend on each other, a valid I-map must include an edge connecting them rather than quietly leaving them disconnected.
Heard on the show
“It's what's called an I-map.”Episode 042 — An Agentic Scientific Computing System That Actually Remembers What It Learns