The complexity of approximating MAPs for belief networks with bounded probabilities
β Scribed by Ashraf M. Abdelbar; Stephen T. Hedetniemi; Sandra M. Hedetniemi
- Publisher
- Elsevier Science
- Year
- 2000
- Tongue
- English
- Weight
- 69 KB
- Volume
- 124
- Category
- Article
- ISSN
- 0004-3702
No coin nor oath required. For personal study only.
β¦ Synopsis
Probabilistic inference and maximum a posteriori (MAP) explanation are two important and related problems on Bayesian belief networks. Both problems are known to be NP-hard for both approximation and exact solution. In 1997, Dagum and Luby showed that efficiently approximating probabilistic inference is possible for belief networks in which all probabilities are bounded away from 0. In this paper, we show that the corresponding result for MAP explanation does not hold: finding, or approximating, MAPs for belief networks remains NP-hard for belief networks with probabilities bounded within the range [l, u] for any 0 l < 0.5 < u 1. Our results cover both deterministic and randomized approximation.
π SIMILAR VOLUMES
I4~, study the approximation o.f continuous mappings and dichotomies by one-hidden-layer networks from a computational point of view Our approach is based on a new approximation method specially designed for constructing "small networks. We give upper bounds on the size o.f these networks.
## Abstract We study the bounded approximation property for spaces of holomorphic functions. We show that if __U__ is a balanced open subset of a FrΓ©chetβSchwartz space or (__DFM__ )βspace __E__ , then the space βοΈ(__U__ ) of holomorphic mappings on __U__ , with the compactβopen topology, has the b