nucleation

LIT-1544: Correlated Non-Parametric Latent Feature Models

1 min read

Properties5
StatusActive — read 2026-10-26 (NOTE-1184) in full, from the arXiv copy of the UAI paper; worth reading for the category layer that makes IBP features co-occur while keeping finitely many on finite data, and for the condition that bounds what such models can express (THEORY-1012). Read it knowing that its features can be switched on by categories and never off, so it does not model inhibition, and that its table supports a gain over the IBP on three of five datasets, not "in almost all cases".
Version2
Tagsprobabilistic-modeling
Cited as “paper” byNOTE-1184 — Doshi-Velez and Ghahramani — Correlated non-parametric latent feature models
Cited as “source” byTHEORY-1012 — Latent features made dependent through a layer of categories, each observation's features being a function of how many of its categories carry each feature, stay finite on finite data when a feature carried by none of an observation's categories is always off; under that condition a category can switch features on and never off, so features depend on one another by sharing a category, or, where an observation has a single category, by excluding one another, and no feature or category suppresses another

Graph View

Doshi-Velez, F. & Ghahramani, Z. (2009). Correlated non-parametric latent feature models. In Proceedings of the 25th Conference on Uncertainty in Artificial Intelligence (UAI 2009). PMLR R7: 143–150. arXiv:1205.2650.

Read 2026-10-26 in full, in NOTE-1184. The summary was checked against the reading and holds. T22 cited it for modes that “reinforce, inhibit, or depend on one another”; the reading supports the first and the last, not inhibition (THEORY-1012).

Why this is here

Cited at T22. The independence of IBP features is one source of the identifiability gaps that CASE-165 records for T23’s model, where features that co-occur are identified only as a sum; this is the established generalisation that models their correlation. It bears on QUESTION-276.

Backlinks