Research map / Probabilistic and scientific modeling / Variational and generative sampling
Filtering Variational Objectives
Chris J Maddison, John Lawson, George Tucker, Nicolas Heess, Mohammad Norouzi, Andriy Mnih, Arnaud Doucet, Yee Teh
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 |
|---|---|---|---|
| Train the VRNN with FIVO rather than the strongest ELBO/IWAE baseline | VRNN · Nottingham, JSB, MuseData, Piano-midi.de, and TIMIT | test log-likelihood bound: Nottingham: -2.85 -> -2.58; JSB: -7.41 -> -6.72; MuseData: -7.13 -> -5.89; Piano-midi.de: -7.85 -> -7.43; TIMIT 64 units | Improved |
| Add a smoothing posterior (ablation) | VRNN trained with ELBO · Nottingham, JSB, MuseData, Piano-midi.de, and TIMIT train | log-likelihood bound: Nottingham: -2.40 -> -2.59; JSB: -5.48 -> -5.53; MuseData: -6.54 -> -6.48; Piano-midi.de: -6.68 -> -6.77; TIMIT: 0 -> -9 | Mixed |
| Add a smoothing posterior (ablation) | VRNN trained with IWAE · Nottingham, JSB, MuseData, Piano-midi.de, and TIMIT train | log-likelihood bound: Nottingham: -2.52 -> -2.37; JSB: -5.77 -> -4.63; MuseData: -6.54 -> -6.47; Piano-midi.de: -6.74 -> -6.74; TIMIT: 1,469 - | Improved |
| Add a smoothing posterior (ablation) | VRNN trained with FIVO · Nottingham, JSB, MuseData, Piano-midi.de, and TIMIT train | log-likelihood bound: Nottingham: -2.29 -> -2.34; JSB: -4.08 -> -3.83; MuseData: -5.80 -> -5.87; Piano-midi.de: -6.41 -> -6.34; TIMIT: 6,991 - | Mixed |
| Include the resampling-gradient terms during training (ablation) | VRNN trained with FIVO · TIMIT and polyphonic music datasets | training and held-out log-likelihood | Hurt |
| Allow the variational posterior to condition on future observations in FIVO (ablation) | VRNN trained with FIVO · sequential modeling tasks | log-likelihood bound | No clear effect |
Closest papers by meaning
- Variational Inference for Monte Carlo Objectives (ICML 2016 · similarity 0.87)
- Debiasing Evidence Approximations: On Importance-weighted Autoencoders and Jackknife Variational Inference (ICLR 2018 · similarity 0.86)
- Variational Inference and Model Selection with Generalized Evidence Bounds (ICML 2018 · similarity 0.85)
- Discretely Relaxing Continuous Variables for tractable Variational Inference (NeurIPS 2018 · similarity 0.85)
- Learning Latent Variable Models via Jarzynski-adjusted Langevin Algorithm (NeurIPS 2025 · similarity 0.84)
- Forward $\chi^2$ Divergence Based Variational Importance Sampling (ICLR 2024 · similarity 0.83)
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.
