Rudi, Alessandro; Ciliberto, Carlo; (2021) PSD Representations for Effective Probability Models. In: Ranzato, M and Beygelzimer, A and Dauphin, Y and Liang, PS and Vaughan, JW, (eds.) Advances in Neural Information Processing Systems 34 (NeurIPS 2021). NeurIPS Proceedings: Online conference.

Preview

Text 6271_psd_representations_for_effect.pdf - Published Version Download (728kB) | Preview

Abstract

Finding a good way to model probability densities is key to probabilistic inference. An ideal model should be able to concisely approximate any probability while being also compatible with two main operations: multiplications of two models (product rule) and marginalization with respect to a subset of the random variables (sum rule). In this work, we show that a recently proposed class of positive semi-definite (PSD) models for non-negative functions is particularly suited to this end. In particular, we characterize both approximation and generalization capabilities of PSD models, showing that they enjoy strong theoretical guarantees. Moreover, we show that we can perform efficiently both sum and product rule in closed form via matrix operations, enjoying the same versatility of mixture models. Our results open the way to applications of PSD models to density estimation, decision theory, and inference.

Download activity - last month

Export as JSONXMLCSV

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as JSONCSVXML

Archive Staff Only

View Item