Marginal Weighted Maximum Log-likelihood for Efficient Learning of Perturb-and-Map models
arXiv:1811.08725
Abstract
We consider the structured-output prediction problem through probabilistic approaches and generalize the "perturb-and-MAP" framework to more challenging weighted Hamming losses, which are crucial in applications. While in principle our approach is a straightforward marginalization, it requires solving many related MAP inference problems. We show that for log-supermodular pairwise models these operations can be performed efficiently using the machinery of dynamic graph cuts. We also propose to use double stochastic gradient descent, both on the data and on the perturbations, for efficient learning. Our framework can naturally take weak supervision (e.g., partial labels) into account. We conduct a set of experiments on medium-scale character recognition and image segmentation, showing the benefits of our algorithms.
arXiv commentPublished in Proceedings of the Conference of Uncertainty in Artificial Intelligence (UAI), 2018
References in corpus (6)
- Tightening LP Relaxations for MAP using Message Passing
- On the Partition Function and Random Maximum A-Posteriori Perturbations
- Minding the Gaps for Block Frank-Wolfe Optimization of Structured SVMs
- Scalable Variational Inference in Log-supermodular Models
- Closed-Form Training of Conditional Random Fields for Large Scale Image Segmentation
- Submodular relaxation for inference in Markov random fields