Research map / Probabilistic and scientific modeling / Variational and generative sampling
Automatic variational inference with cascading flows
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 |
|---|---|---|---|
| Use cascading flows instead of the strongest reported baseline | structured timeseries inference · LZ-r, LZ-c, PD-r, PD-c, RNN-r, RNN-c | predictive or latent log-likelihood: LZ-r Pred: -2.89 ± 0.17 vs -4.48 ± 0.60; LZ-c Pred: 5.10 ± 0.52 vs 0.86 ± 0.15; PD-r Latent: -2.32 ± 0.19 vs -3.14 ± 0.1 | Helped |
| Use cascading flows instead of ASVI | Brownian-motion timeseries inference · BR-r and BR-c | predictive and latent log-likelihood: BR-r Pred: -2.27 ± 0.26 vs -2.23 ± 0.21; BR-r Latent: -1.48 ± 0.19 vs -1.45 ± 0.14; BR-c Latent: -1.53 ± 0.21 vs -1.55 ± | Mixed |
| Use cascading flows instead of the strongest baseline | binary-tree collider inference · Linear-2 and Tanh-2 | latent log-likelihood: Linear-2: 1.24 ± 0.01 vs 0.97 ± 0.01; Tanh-2: 1.20 ± 0.01 vs 1.15 ± 0.01 | Helped |
| Use cascading flows instead of the strongest baseline | binary-tree collider inference · Linear-4 and Tanh-4 | latent log-likelihood: Linear-4: 5.42 ± 0.35 vs 6.56 ± 0.04; Tanh-4: 3.29 ± 3.17 vs 10.32 ± 0.05 | Hurt |
| Use cascading flows instead of global flows | amortized timeseries inference · PD and LZ | latent log-likelihood: PD: -2.8 ± 0.2 vs -5.1 ± 0.2; LZ: -2.7 ± 0.2 vs -9.1 ± 0.2 | Helped |
| Remove highway gates (CF non-res) (ablation) | cascading flows · timeseries tasks | predictive and latent log-likelihood: LZ-r Pred: -2.89 ± 0.17 vs -8.25 ± 0.27; LZ-c Pred: 5.10 ± 0.52 vs 0.88 ± 0.04; PD-c Latent: -2.77 ± 0.18 vs -8.41 ± 0.4 | Hurt |
Closest papers by meaning
- Amortized Variational Transdimensional Inference (NeurIPS 2025 · similarity 0.90)
- Variational Message Passing with Structured Inference Networks (ICLR 2018 · similarity 0.89)
- Model-Informed Flows for Bayesian Inference (NeurIPS 2025 · similarity 0.89)
- Embedded-model flows: Combining the inductive biases of model-free deep learning and explicit probabilistic modeling (ICLR 2022 · similarity 0.89)
- Multilayer Matrix Factorization via Dimension-Reducing Diffusion Variational Inference (ICML 2025 · similarity 0.86)
- VIKING: Deep variational inference with stochastic projections (NeurIPS 2025 · similarity 0.86)
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.