Research map / Probabilistic and scientific modeling / Variational and generative sampling
Variational Walkback: Learning a Transition Operator as a Stochastic Recurrent Net
Anirudh Goyal ALIAS PARTH GOYAL, Nan Rosemary Ke, Surya Ganguli, Yoshua Bengio
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 |
|---|---|---|---|
| Increase MNIST walkback training from 5 to 15 steps (ablation) | VW · MNIST | log-likelihood lower bound: 510 -> 1018 | Improved |
| Increase MNIST walkback training from 5 to 10 steps (ablation) | VW · MNIST | log-likelihood lower bound: 510 -> 801 | Improved |
| Use importance sampling rather than the lower-bound estimate (ablation) | VW · MNIST | test log-likelihood: 1018 -> 1243 | Improved |
| Use VW instead of NET | generative modeling · CIFAR-10 test | bits/dim: 5.40 -> 4.40 | Improved |
| Use VW instead of VAE-10 | generative modeling · MNIST | test log-likelihood importance sampling estimate: 705 ± 7.4 -> 712 ± 3.4 | Improved |
| Use VW instead of VAE-50 | generative modeling · MNIST | test log-likelihood importance sampling estimate: 991 ± 6.5 -> 712 ± 3.4 | Hurt |
| Use layer normalization in the MNIST bottleneck instead of batch normalization (ablation) | VW encoder-decoder · MNIST | sample quality | Improved |
| Add dropout (ablation) | VW encoder-decoder · MNIST | sample quality | Improved |
| Increase walkback training trajectory length (ablation) | VW · MNIST | log-likelihood lower bound: 702 -> 890 -> 1148 | Improved |
Closest papers by meaning
- Sequence Modeling with Spectral Mean Flows (NeurIPS 2025 · similarity 0.86)
- End-to-End Identifiable and Consistent Recurrent Switching Dynamical Systems (NeurIPS 2026 · similarity 0.86)
- Inferring stochastic low-rank recurrent neural networks from neural data (NeurIPS 2024 · similarity 0.86)
- SING: SDE Inference via Natural Gradients (NeurIPS 2025 · similarity 0.86)
- Identifying latent state transitions in non-linear dynamical systems (ICLR 2025 · similarity 0.86)
- Offline Transition Modeling via Contrastive Energy Learning (ICML 2024 · similarity 0.85)
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.
