Research map / Probabilistic and scientific modeling / Variational and generative sampling
NAS-X: Neural Adaptive Smoothing via Twisting
Dieterich Lawson, Michael Y. Li, Scott Linderman
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 NAS-X instead of SIXO | HH inference, proposals trained with 4 particles · Squid giant axon HH | log-likelihood lower bound: NAS-X outperforms SIXO by roughly 34 nats at 4 particles | Improved |
| Use NAS-X instead of NASMC | HH inference · Squid giant axon HH | particle efficiency: 4 particles; 256 particles; 64-times increase | Improved |
| Use NAS-X instead of SIXO | HH inference · Squid giant axon HH | particle efficiency: 2x particle efficiency improvement | Improved |
| Increase the number of training particles | NAS-X · Squid giant axon HH | inference performance | No clear effect |
| Increase the number of training particles | VI-based methods · Squid giant axon HH | inference performance | Hurt |
| Use a bootstrap proposal instead of FIVO's parametric proposal | FIVO · Squid giant axon HH | log-likelihood lower bound | Improved |
| Use NAS-X instead of SIXO | Mouse pyramidal neuron model learning · Mouse pyramidal neuron | spike count error | Improved |
| Use NAS-X instead of SIXO | Mouse pyramidal neuron model learning · Mouse pyramidal neuron | test log-likelihood lower bound | Hurt |
| Use NAS-X instead of SIXO | Mouse pyramidal neuron model learning · Mouse pyramidal neuron | training-run failure rate | Improved |
| Use NAS-X instead of SIXO | HH model training with the reported particle counts · Squid giant axon HH | training bound: NAS-X: -664.3528; SIXO: -668.6865; NAS-X: -662.8712; SIXO: -653.6352; NAS-X: -662.0753; SIXO: -644.8764; NAS-X: -661.538 | Mixed |
| Use NAS-X instead of SIXO | HH bootstrap evaluation · Squid giant axon HH | BPF bound (256 particles): -660.7003 -> -636.2579 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- SIXO — SIXO: Smoothing Inference with Twisted Objectives (NeurIPS 2022) · better on 1 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Robust Neural Posterior Estimation and Statistical Model Criticism (NeurIPS 2022 · similarity 0.85)
- Preferential dynamic modeling with forward-backward smoothing (NeurIPS 2026 · similarity 0.85)
- Flexible statistical inference for mechanistic models of neural dynamics (NeurIPS 2017 · similarity 0.85)
- Probabilistic Decomposed Linear Dynamical Systems for Robust Discovery of Latent Neural Dynamics (NeurIPS 2024 · similarity 0.85)
- Mechanistic Interpretability of RNNs emulating Hidden Markov Models (NeurIPS 2025 · similarity 0.83)
- Closing the Approximation Gap in Simulation-free Latent SDEs (NeurIPS 2026 · 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.
