Research map / Probabilistic and scientific modeling / Variational and generative sampling
Truncated Inference for Latent Variable Optimization Problems: Application to Robust Estimation and Learning
Christopher Zach, Huu Le
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 |
|---|---|---|---|
| use ReGeMM condition for confidence weights instead of IRLS | robust bundle adjustment · Dubrovnik-356 and Venice-427 bundle adjustment | final objective value | Improved |
| use ReGeMM condition instead of explicit graduated optimization | robust bundle adjustment · Dubrovnik-356 and Venice-427 bundle adjustment | final objective value | No clear effect |
| use ReGeMM instead of IRLS in run-time | robust bundle adjustment · 20 metric bundle adjustment instances | run-time: between 5% and 25% slower than IRLS | Hurt |
| use adaptive inference (ReGeMM or SuDeMM) instead of fixed 32 passes (ablation) | contrastive Hebbian learning, full batch · MNIST subset, 784-64x4-10 | objective value | No clear effect |
| use fixed 2 inference passes instead of 32 or adaptive (ablation) | contrastive Hebbian learning, full batch · MNIST subset | training speed per epoch | Improved |
| use SuDeMM test instead of ReGeMM condition (ablation) | contrastive Hebbian learning, full batch · MNIST subset | number of inference steps | Improved |
| use too few inference passes in stochastic training (ablation) | stochastic contrastive learning · MNIST, Fashion-MNIST, KMNIST, CIFAR-10 | objective value | Hurt |
| use stochastic SuDeMM adaptive precision instead of fixed passes (ablation) | stochastic contrastive learning · MNIST, Fashion-MNIST, KMNIST, CIFAR-10 | estimated true objective | Improved |
Closest papers by meaning
- Learning latent variable structured prediction models with Gaussian perturbations (NeurIPS 2018 · similarity 0.88)
- Online but Accurate Inference for Latent Variable Models with Local Gibbs Sampling (JMLR 2017 · similarity 0.86)
- Adaptive Robust Learning using Latent Bernoulli Variables (ICML 2024 · similarity 0.85)
- Preventing Posterior Collapse with delta-VAEs (ICLR 2019 · similarity 0.85)
- Variational (Gradient) Estimate of the Score Function in Energy-based Latent Variable Models (ICML 2021 · similarity 0.85)
- Posterior Inference in Latent Space for Scalable Constrained Black-box Optimization (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.
