A sum-product algorithm with polynomials for computing exact derivatives of the likelihood in Bayesian networks

Abstract : We consider a Bayesian network with a parameter θ. It is well known that the probability of an evidence conditional on θ (the likelihood) can be computed through a sum-product of potentials. In this work we propose a polynomial version of the sum-product algorithm based on generating functions for computing both the likelihood function and all its exact derivatives. For a unidimensional parameter we obtain the derivatives up to order d with a complexity O(C × d 2) where C is the complexity for computing the likelihood alone. For a parameter of p dimensions we obtain the likelihood, the gradient and the Hessian with a complexity O(C × p 2). These complexities are similar to the numerical method with the main advantage that it computes exact derivatives instead of approximations.
Complete list of metadatas

Cited literature [12 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-02350422
Contributor : Gregory Nuel <>
Submitted on : Friday, November 8, 2019 - 8:34:04 AM
Last modification on : Sunday, November 10, 2019 - 1:30:12 AM

Identifiers

  • HAL Id : hal-02350422, version 1

Citation

Alexandra Lefebvre, Grégory Nuel. A sum-product algorithm with polynomials for computing exact derivatives of the likelihood in Bayesian networks. Proceedings of Machine Learning Research, PMLR, 2018, 72, pp.201 - 212. ⟨hal-02350422⟩

Share

Metrics

Record views

32

Files downloads

6