Research map / Probabilistic and scientific modeling / Variational and generative sampling
Variational Inference for Monte Carlo Objectives
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| VIMCO per-sample learning signals (leave-one-out geometric-mean baseline) instead of NVIL | multi-sample objective training of SBN · MNIST generative, test NLL | negative log-likelihood (nats): NVIL 93.4 vs VIMCO 92.6 at 10 samples; 96.2 vs 91.9 at 50 samples | Helped |
| Increasing number of samples in the objective with VIMCO (ablation) | VIMCO multi-sample objective · MNIST generative, test NLL | negative log-likelihood (nats): 93.5 at 2 samples to 91.9 at 50 samples | Helped |
| Increasing number of samples with NVIL (ablation) | NVIL multi-sample objective · MNIST generative, test NLL | negative log-likelihood (nats): 93.6 at 2 samples to 96.2 at 50 samples | Hurt |
| VIMCO instead of RWS (wake-only update) for proposal learning | multi-sample generative training · MNIST generative, test NLL | negative log-likelihood (nats): VIMCO better at 2 samples (93.5 vs 94.6); RWS better at 50 samples (92.5 vs 91.9 favoring VIMCO, actually VIMCO lower) | Mixed |
| Learned proposal distribution instead of prior proposal (ablation) | VIMCO structured output prediction · MNIST bottom-half completion, NLL with 100 samples | negative log-likelihood (nats): 56.5 (prior) to 46.1 (learned) | Helped |
| VIMCO per-sample learning signals versus NVIL baseline (learning-signal magnitude) (ablation) | multi-sample training of SBN on MNIST · MNIST generative, learning signal RMS | learning signal magnitude: about 3 times lower than NVIL at 10 samples | Helped |
Closest papers by meaning
- Variational Inference for Sequential Data with Future Likelihood Estimates (ICML 2020 · similarity 0.92)
- Forward $\chi^2$ Divergence Based Variational Importance Sampling (ICLR 2024 · similarity 0.92)
- Variational Inference and Model Selection with Generalized Evidence Bounds (ICML 2018 · similarity 0.90)
- Variational Inference with Locally Enhanced Bounds for Hierarchical Models (ICML 2022 · similarity 0.90)
- Sticking the Landing: Simple, Lower-Variance Gradient Estimators for Variational Inference (NeurIPS 2017 · similarity 0.90)
- SoftCVI: Contrastive variational inference with self-generated soft labels (ICLR 2025 · similarity 0.88)
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.