Approximating the Log-Partition Function
Name
Cosson-cosson-SM-EECS-2021-thesis.pdf
Description
Thesis PDF
Size
704.5 KB
Format
Adobe PDF
Checksum (MD5)
79ca43ed750f2b0dd156ed27c0619acd
Author(s)
Cosson, Romain
Advisor(s)
Shah, Devavrat
Date Issued
June 2021
Publisher
Massachusetts Institute of Technology
Abstract
Graphical Models are used to represent structural information on a high-dimensional joint probability distribution. Their expressiveness offers simple reductions from a large number of NP-hard problems to inference tasks such as computing the partition function (exact inference) or approximating the log-partition function (approximate inference). In this master thesis, we will motivate the need for a general constant-factor approximations of the log-partition function and prove that a variant of the well studied tree-reweighted algorithm [1] achieves constant factor guarantees. We will express the corresponding approximation ratio 𝜅(𝐺) solely as a function of the graph structure 𝐺.
MIT Department
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
Terms of Use
In Copyright - Educational Use Permitted
Copyright MIT
Persistent DSpace Link
