Research map / Probabilistic and scientific modeling / Dynamical systems
Capturing Membrane Dynamics and Spike Timing of Human Neurons at Scale using Neural Operators
Luca Ghafourpour, Cindy Zhao, Valentin Duruisseaux, Bahareh Tolooshams, Philip Wong, Costas Anastassiou, Animashree Anandkumar
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a cell-type-specific neural-operator framework to scalably emulate biophysically realistic human neuron membrane dynamics and spike timing across populations.
Problem
Biophysically realistic neuron models are hard to scale to large populations while preserving rich cellular dynamics and within-cell-type variability.
Approach
A cell-type-specific neural-operator framework that replaces training a separate surrogate model per neuron, instead modeling membrane dynamics and spike timing across cell types in a single scalable operator approach.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace two-feature (I_thr, s_thr) embedding with six selected electrophysiological features (ablation) | NOBLE · PVALB | inverse ISI error: 50.53% -> 16.97% | Improved |
| replace two-feature embedding with six features (ablation) | NOBLE · PVALB | missed firing rate: 6.47% -> 1.87% | Improved |
| add feature embedding (2 features) to a model with none (ablation) | NOBLE · PVALB | voltage error: 8.01% -> 2.21% | Improved |
| add a seventh embedding feature (ablation) | NOBLE · PVALB | electrophysiological metrics: 1.73% -> 1.79% voltage | No clear effect |
| use a learned invertible feature transform instead of fixed min-max scaling (ablation) | NOBLE · PVALB | voltage error: 1.73% -> 1.69% | No clear effect |
| add sigmoid-gated spike-weighted L2 loss in place of standardized L4 loss (ablation) | NOBLE · PVALB | inverse ISI error: 29.36% -> 16.97% | Improved |
| add sigmoid-gated spike-weighted loss (ablation) | NOBLE · PVALB | mean AP amplitude error: 15.71% -> 8.73% | Improved |
| share one operator across cells (cell-type-level) instead of cell-specific model (ablation) | NOBLE · PVALB single neuron | voltage error: 1.92% -> 2.30% | Hurt |
| evaluate on entire cells excluded from training | NOBLE · PVALB unseen cells | inverse ISI error: 16.97% -> 75.29% | Hurt |
| surrogate vs PDE simulator | NOBLE · PVALB/SST/VIP | inference time: 36,200 s per 10,000 predictions vs 3,640x speedup | Improved |
Closest papers by meaning
- NOBLE - Neural Operator with Biologically-informed Latent Embeddings to Capture Experimental Variability in Biological Neuron Models (NeurIPS 2025 · similarity 0.89)
- P-SPIKESSM: HARNESSING PROBABILISTIC SPIKING STATE SPACE MODELS FOR LONG-RANGE DEPENDENCY TASKS (ICLR 2025 · similarity 0.87)
- Disentangling the Roles of Distinct Cell Classes with Cell-Type Dynamical Systems (NeurIPS 2024 · similarity 0.87)
- Identifying multi-compartment Hodgkin-Huxley models with high-density extracellular voltage recordings (NeurIPS 2025 · similarity 0.87)
- Towards a "Universal Translator" for Neural Dynamics at Single-Cell, Single-Spike Resolution (NeurIPS 2024 · similarity 0.87)
- FiTS: Interpretable Spiking Neurons via Frequency Selectivity and Temporal Shaping (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.
