Research map / Probabilistic and scientific modeling / Variational and generative sampling
Blending Learning and Inference in Conditional Random Fields
Tamir Hazan, Alexander G. Schwing, Raquel Urtasun
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| blend learning with message passing inference instead of nested inference | standard CRF learning with 20 message passing iterations · Room layout (Hedau et al. 2009) | training time to zero primal-dual gap: more than 600 seconds -> less than 50 seconds | Improved |
| use 10 message passing iterations in standard learning instead of 20 (ablation) | standard learning · Room layout (Hedau et al. 2009) | primal-dual gap: 10 iterations never close the duality gap | Hurt |
| blend learning and inference | standard learning with 20 iterations · Room layout (Hedau et al. 2009) | test set error over training time: more than 10 times faster | Improved |
| add loss-augmented (loss) term when counting numbers equal zero (ablation) | blending with counting numbers 0 · Room layout (Hedau et al. 2009) | test error: stuck at high error without loss | Improved |
| add loss when counting numbers equal one (ablation) | blending with counting numbers 1 · Room layout (Hedau et al. 2009) | test error: similar performance to loss-included setting | No clear effect |
| reduce standard learning message passing iterations from 20 down to 2 (ablation) | standard learning · Room layout (Hedau et al. 2009) | test error: approaches blending performance | No clear effect |
| update messages twice per parameter update (two rounds of message passing) (ablation) | blended learning · Room layout (Hedau et al. 2009) | convergence speed: slightly faster convergence | Improved |
Closest papers by meaning
- Discriminative Training of Conditional Random Fields with Probably Submodular Constraints (IJCV 2020 · similarity 0.87)
- A Discriminative View of MRF Pre-Processing Algorithms (ICCV 2017 · similarity 0.86)
- Efficient semidefinite-programming-based inference for binary and multi-class MRFs (NeurIPS 2020 · similarity 0.86)
- Neural Conditional Probability for Uncertainty Quantification (NeurIPS 2024 · similarity 0.85)
- Bayesian Exploration of Pre-trained Models for Low-shot Image Classification (CVPR 2024 · similarity 0.84)
- Maxitive Donsker-Varadhan Formulation for Possibilistic Variational Inference (NeurIPS 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
