Research map / Probabilistic and scientific modeling / Variational and generative sampling
Natural gradient enables fast sampling in spiking neural networks
Paul Masset, Jacob A Zavatone-Veth, J. Patrick Connor, Venkatesh N Murthy, Cengiz Pehlevan
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 natural-gradient geometry (D = Sigma) instead of naive geometry (D = I) in a Langevin-sampling efficient balanced network (ablation) | EBN Langevin sampler · Equicorrelated Gaussian, increasing correlation rho | estimated variance and 2-Wasserstein distance (50 ms window) | Improved |
| Use natural geometry instead of naive geometry in the Langevin EBN as correlation rho grows (ablation) | EBN Langevin sampler · Equicorrelated Gaussian, varying rho | mean estimate after mean shift | Improved |
| Use natural geometry instead of naive geometry in the Langevin EBN as parameter dimension np grows from 2 to 64 (ablation) | EBN Langevin sampler · Equicorrelated Gaussian, varying np | mean 2-Wasserstein distance | Improved |
| Use natural geometry in the Langevin EBN to reduce discretization-induced variance overestimation (ablation) | EBN Langevin sampler · Equicorrelated Gaussian | variance overestimation | Improved |
| Use geometry-aware readout weights (Gamma Gamma^T ~ Sigma) instead of generic Gamma in the Metropolis-Hastings spiking network (ablation) | probabilistic spike-rule network · Equicorrelated Gaussian, strong rho (np=10, nn=100) | average firing rate | Improved |
| Use geometry-aware Gamma instead of naive Gamma in the probabilistic spiking network at strong correlation (ablation) | probabilistic spike-rule network · Equicorrelated Gaussian, strong rho | mean and variance estimates | Improved |
| Use geometry-aware Gamma instead of naive Gamma in the probabilistic spiking network as dimension np increases (nn = 10 np, rho = 0.75) (ablation) | probabilistic spike-rule network · Equicorrelated Gaussian, varying np | firing rate and 2-Wasserstein distance | Improved |
| Natural-space sampling versus naive Langevin dynamics in linear rate networks (analytical) (ablation) | linear rate network Langevin sampler · zero-mean Gaussian, analytical 2-Wasserstein | 2-Wasserstein distance to target over time | Improved |
Closest papers by meaning
- P-SPIKESSM: HARNESSING PROBABILISTIC SPIKING STATE SPACE MODELS FOR LONG-RANGE DEPENDENCY TASKS (ICLR 2025 · similarity 0.89)
- The Bayesian sampling in a canonical recurrent circuit with a diversity of inhibitory interneurons (NeurIPS 2024 · similarity 0.88)
- Latent Diffusion for Neural Spiking Data (NeurIPS 2024 · similarity 0.87)
- Brain-like Variational Inference (NeurIPS 2025 · similarity 0.87)
- A Scalable, Causal, and Energy Efficient Framework for Neural Decoding with Spiking Neural Networks (NeurIPS 2025 · similarity 0.86)
- The Dormant Spiking Neuron: A State-Driven Mechanism for Efficient Spiking Neural Networks (NeurIPS 2026 · 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.
