Bayesian Inference Lab

Same posterior, different formulations — then deliberately worse approximations.

← original QBar demo
posterior weight likelihood prior
mode inference
—
naive replay (if run)
—
KL(exact posterior ∥ mode)
—
KL(exact predictive ∥ mode)
—

Posterior predictive · QBar is the input

Horizontal position is probability p, not x. Tap/click anywhere: the current predictive inverse CDF turns that p into an observation x, then Bayes updates. Orange dots are past observations reprojected into the current predictive.

Tap/click a probability position; QBar converts it through the predictive inverse CDF into the next observation.