ReMAP performs instance-specific optimization of MRF MAP inference by using a GNN to generate label distributions and gradient descent in an over-parameterized continuous space to find low-energy discrete solutions.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
ReMAP: Neural Reparameterization for Scalable MAP Inference in Arbitrary-Order Markov Random Fields
ReMAP performs instance-specific optimization of MRF MAP inference by using a GNN to generate label distributions and gradient descent in an over-parameterized continuous space to find low-energy discrete solutions.