Pith. sign in

REVIEW 3 major objections 6 minor 1 cited by

Spike times alone can be mapped to whole degenerate populations of conductance-based neuron models via DIC intermediates, enabling fast inference of the many conductance sets behind one observed firing pattern.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review

2026-08-04 16:32 UTC pith:5S2QEPLV

load-bearing objection Genuinely useful methods paper: learned spike-to-DIC mapping plus iterative compensation works well in silico, but the fixed-threshold Vth assumption and lack of external validation keep it from being the final word. the 3 major comments →

arxiv 2509.12783 v2 pith:5S2QEPLV submitted 2025-09-16 q-bio.NC cs.LGmath.DSstat.ML

Fast reconstruction of degenerate populations of conductance-based neuron models from spike times

Julien Brandoit , Damien Ernst , Guillaume Drion , Arthur Fyon This is my paper
classification q-bio.NC cs.LGmath.DSstat.ML MSC 92C2068T07
keywords conductance-based modelsneuronal degeneracydynamic input conductancesspike train inferenceiterative compensationdeep learningSTG neuron modeldopaminergic neuron model
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper claims that the inverse problem of inferring ion-channel conductances from spike recordings can be solved in a fast, scalable way by splitting it into two steps joined by an interpretable low-dimensional representation. First, a lightweight deep network maps a variable-length spike train to three scalar Dynamic Input Conductance values at threshold; second, an iterative compensation algorithm maps those targets to many distinct conductance vectors that all reproduce the input activity. If correct, this turns spike-only recordings into millisecond-scale reconstructions of degenerate neuronal populations, capturing biological variability rather than a single best-fit model. The authors validate the pipeline on two conductance-based models and on noisy Poisson spike trains, and report a 15-fold reduction in residual DIC error from the iterative step alone.

Core claim

The central claim is that the inverse problem of mapping spike times to conductance-based model parameters becomes tractable when routed through Dynamic Input Conductances (DICs) evaluated at threshold. DICs reduce each high-dimensional conductance vector to three scalars — fast, slow, and ultra-slow feedback gains — through the identity g_DICs(Vth)=S(Vth;gbar)·gbar. The paper trains an attention-based encoder–decoder to predict the slow and ultra-slow DIC values directly from raw spike times, then uses an iterative compensation algorithm that recomputes the sensitivity matrix at each step to generate diverse conductance vectors satisfying those targets. After five iterations the residual DI

What carries the argument

Dynamic Input Conductances (DICs): three voltage-dependent curves — fast, slow, ultra-slow — that aggregate all ionic currents in a CBM into interpretable feedback components; their values at the threshold voltage Vth summarize a model in three scalars via the identity g_DICs(Vth)=S(Vth;gbar)·gbar, where S is the sensitivity matrix. The iterative compensation algorithm updates compensable conductances by repeatedly solving A(gbar_comp^(k))·gbar_comp^(k+1)=b(gbar_comp^(k)), recomputing S at each step, so that nonlinear models with calcium-dependent currents can satisfy DIC constraints.

Load-bearing premise

The pipeline treats the threshold voltage Vth as a fixed constant (−51 mV for STG, −55.5 mV for DA) across all conductance samples; if true threshold varies with conductances, DIC constraints are enforced at the wrong voltage and populations could deviate from intended firing patterns.

What would settle it

Sample a broad set of conductance vectors from the analysis distribution, compute each instance's true threshold from g_t(Vth)=0, and compare against the fixed a priori value; if the spread of true thresholds exceeds a few millivolts in a way that changes DIC-predicted firing class, the constant-threshold approximation fails. Alternatively, retrain the pipeline with per-instance thresholds and check whether generated population activity matches inputs more closely than the fixed-threshold version.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Spike recordings alone suffice to generate in silico populations of conductance-based models that reproduce the recorded firing statistics, removing the need for voltage traces in parameter inference.
  • Because the pipeline outputs many conductance vectors rather than one, it makes neuronal degeneracy explicit and lets experiments ask how variable channel densities can be while preserving activity.
  • The speed of the method (about 1.25e-5 seconds per forward pass on GPU, milliseconds per population) makes real-time or near-real-time inference feasible during experiments.
  • The pipeline transfers to a new neuron model with only about 34% additional parameters via low-rank adaptation, suggesting reuse across cell types with minimal retraining.
  • Robustness to Poisson-distributed spike trains indicates the method can handle irregular, physiologically realistic variability despite training on regular spontaneous activity.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A natural testable extension is to build datasets with DIC constraints at multiple voltages; the paper notes subthreshold features such as after-depolarizations are currently invisible, and multi-voltage DICs would recover them.
  • The one-dimensional manifold found for spiking DIC predictions implies that spike trains alone may not uniquely determine slow and ultra-slow DICs in the spiking regime; combining spike times with voltage snippets or stimuli may disambiguate them.
  • The same DIC-space sampling could be used to visualise how neuromodulators shift conductance distributions by comparing inferred populations before and after drug application.
  • The iterative compensation algorithm should apply to other nonlinear CBMs (the paper suggests models with calcium or SK dynamics); testing it on a model with a different nonlinearity would show whether the factor-15 improvement generalises.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes a two-stage pipeline for inferring degenerate populations of conductance-based neuron models from spike times alone. In the first stage, an attention-based encoder-decoder network maps variable-length spike-time sequences to slow and ultra-slow Dynamic Input Conductance (DIC) values evaluated at an a priori threshold voltage Vth. In the second stage, an iterative compensation algorithm converts the predicted DIC targets into populations of conductance vectors that approximately satisfy those targets, exploiting degeneracy. The method is validated on two conductance-based models (STG and DA) using synthetic datasets of 1.2 million instances, on Poisson-generated irregular spike trains, and with LoRA-based transfer from STG to DA. The authors report that five compensation iterations reduce the DIC residual norm by a factor of 15 relative to the previous linear method, that generated populations reproduce input spiking and bursting statistics with low bias across 500 test points, and that inference requires on the order of tens of microseconds per spike train.

Significance. If the claims hold, the paper is a significant methodological contribution: it provides an interpretable, low-dimensional intermediate (DICs) that makes spike-only reconstruction of degenerate populations tractable and fast, with open-source software. The engineering is careful in several respects: populations are kept together in train/validation/test splits, the dataset is balanced across spiking and bursting regimes, residual reduction is quantified, bias is analyzed across 500 populations, and robustness to Poisson inputs is explicitly tested. The LoRA transfer is a useful practical addition. However, the evaluation is entirely synthetic and generated by the same DIC-based compensation pipeline that is used at inference, so the quantitative agreement in Figs. 6-7 partly reflects internal consistency rather than external predictive power. The fixed-threshold approximation is also load-bearing and is not validated on the extreme DIC targets used for training. These issues are addressable and do not undermine the core idea, but they must be resolved before the experimental-applicability claims in the abstract and discussion are justified.

major comments (3)
  1. [Materials and Methods, 'Choice of a priori threshold voltage'; Eq. (10); Eq. (5); Eq. (14)] The entire pipeline evaluates DIC targets, residuals, and iterative updates at a fixed a priori Vth (-51 mV for STG, -55.5 mV for DA), while Eq. (10) defines Vth as a per-instance quantity: the first decreasing zero of gt = gf + gs + gu. The only support is an empirical distribution over 4,000 conductance vectors sampled from D_analysis, which does not cover the extreme targets used in the training dataset (e.g., gs ∈ [-20,20], gu ∈ [0,20] for STG). If the true threshold shifts by several mV for those extreme targets, the residual metric Eq. (5), the dataset labels, and the compensation update Eq. (14) all enforce constraints at a voltage that is not the threshold of the generated neuron. The rare outliers reported at extreme conditions in Fig. 7B are exactly where such a shift would appear. Please provide either a theoretical bound on the Vth shift over the sampled DIC box, an empirical
  2. [Results, 'Building a synthetic dataset...' and 'The full generative pipeline...'] The evaluation is a closed loop: the dataset is generated by sampling DIC targets and using the iterative compensation algorithm to produce populations (Fig. 3B), the network is trained to predict those same DIC targets, and the final validation (Figs. 6-7) compares activity of populations generated by the same algorithm from the predicted DICs against activity of populations generated by the same algorithm from the true DICs. This makes the reported agreement a test of self-consistency, not of predictive power on independently defined ground truth. I recommend adding an external validation in which conductance vectors are sampled directly from a biological prior (D_analysis or a held-out region), simulated without the DIC compensation step, and the resulting spike trains are passed through the pipeline; success should be measured by whether the generated populations reproduce the activi
  3. [Introduction; Results, 'Benchmarking...' and Discussion] The paper positions the method against existing inference approaches (refs. 13-22) and against the linear compensation method of [24], and claims speed and scalability. However, no quantitative comparison with any external inference method is provided. Fig. 2A compares iterative compensation only with its own linear predecessor, and the reported inference time (1.25e-5 s per forward pass) has no baseline. To make the 'fast' and 'scalable' claims meaningful, please include at least one comparison on a common task: e.g., recovering conductance populations or reproducing activity from a held-out set of spike trains, using an existing method such as a Bayesian sampler or the linear compensation of [24] as a baseline. This does not require exhaustive benchmarking, but it is necessary to locate the contribution relative to the state of the art.
minor comments (6)
  1. [General] The manuscript repeatedly defers essential details to S1 Appendix (model equations, DIC definitions, hyperparameters, threshold-distribution plots, DA dataset sizes). The appendix was not included with the submitted text; without it, reproducibility cannot be assessed. Please ensure the appendix is available to reviewers and that all referenced figures and tables are present.
  2. [Eq. (5)] The residual definition uses \(S\(V; [g_random; g_comp])\) and \S(V; g)\, but \hat{S}\ is not explicitly defined at that point. It would help to state that \hat{S}\ is the sensitivity matrix evaluated at a fixed default value of the compensated conductances.
  3. [Table 1] Table 1b reports MAE for gs and gu but does not report the standard deviation of these DIC values in the dataset, unlike Table 1a for activity descriptors. Adding DIC variability would help interpret the MAE magnitudes, especially for the spiking regime where degeneracy is invoked.
  4. [Fig. 2A] The caption states 1,633 target DICs, while the dataset section describes 75,000 sampled DIC couples. Please clarify the provenance of the 1,633 targets and why this subset is used for the residual comparison.
  5. [Fig. 7] The text says '16×16 neurons per point' and '500 random points', but Fig. 6 mentions 500 generated neurons per regime and Fig. 8 uses 16 generated instances per input. Please make the sample sizes consistent and explicit in each figure caption.
  6. [Results, robustness section] At high Poisson rates, the method sometimes classifies spiking inputs as bursting. The authors attribute this to the lack of a refractory period in Poisson processes. This is reasonable, but it would be useful to report how often this occurs as a function of λ, since it directly affects the practical operating range of the method.

Circularity Check

0 steps flagged

No significant circularity: the spike-to-DIC-to-conductance pipeline is a learned inverse of forward ODE simulation, not an input-output tautology.

full rationale

The pipeline is x → g_hat_DICs (neural network) → g_bar (iterative compensation) → simulated spike trains. DIC labels in the training set are the sampled (gs, gu) targets fed to the compensation algorithm; spike times are produced by full ODE simulation of the generated conductance vectors. Thus the network must genuinely invert a stochastic forward map. Held-out test populations and Poisson-based generalization inputs are not used to fit any parameter. The iterative compensation solves Eq (13)-(14) for conductance vectors matching target DICs; Eq (5) is a residual metric, not a fitted-output-as-prediction. The fixed a priori Vth is a stated modeling simplification validated on 4000 samples, and is not tuned to the test DIC targets. Self-citations to DIC theory [23,24] are backed by explicit equations (Eq 8-10) and by the paper's own dataset mapping, so they are not load-bearing in a circular way. The closed-loop synthetic evaluation tests internal consistency of the DIC intermediate, but the central claim of reconstructing degenerate populations from spike times retains independent empirical content.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The central pipeline rests on the prior DIC reduction, a fixed threshold approximation, a learned mapping trained on synthetic data, and several manually chosen constants (K=5, sampling bounds, network hyperparameters). No new physical entities are introduced. The ledger is moderate for a machine-learning method paper, but the lack of experimental validation means the domain assumptions are untested.

free parameters (5)
  • Fixed threshold voltage V_th = -51 mV (STG), -55.5 mV (DA)
    A constant threshold is used to evaluate DIC targets and sensitivities for all conductance instances; chosen as median of theoretical thresholds over 4,000 sampled conductance vectors (Materials and Methods, 'Choice of a priori threshold voltage').
  • Number of compensation iterations K = 5
    Chosen from Fig 2A as a trade-off between residual reduction and computational cost; residual reduced by factor 15 after 5 iterations.
  • DIC sampling bounds = [-20,20] x [0,20] (STG), [-10,15] x [0,20] (DA)
    Uniform sampling ranges for gs and gu in dataset generation, empirically derived from extensive model sampling (Materials and Methods, 'Synthetic dataset generation').
  • Neural network parameters = 115,627 trainable parameters
    Trained on the synthetic STG dataset via random hyperparameter search; the learned mapping is fitted to the generated data and is load-bearing for spike-to-DIC prediction.
  • LoRA rank r = 32
    Selected via grid search over {2,4,8,16,32,48,64} for transfer to the DA model; introduces 39,844 adapted parameters.
axioms (5)
  • domain assumption The CBM structure (equations, gating dynamics, reversal potentials) is fixed and known; only maximal conductances are unknown.
    Used throughout; inference is restricted to conductance values within a known model class (Results, 'General problem statement').
  • domain assumption DIC values at a single threshold voltage are sufficient to characterize spontaneous firing patterns.
    Relies on prior DIC theory [23,24]; justifies reducing the target to three scalars (Results, 'General problem statement'; Materials and Methods).
  • domain assumption The threshold voltage can be replaced by a fixed constant per model.
    Justified empirically by the narrow distribution of theoretical thresholds over 4,000 samples (Materials and Methods, 'Choice of a priori threshold voltage').
  • domain assumption Synthetic data generated by the same model and compensation algorithm is representative of the target application domain.
    Training and evaluation rely on simulated spike trains with I_ext=0; real recordings may include subthreshold dynamics, noise, and model mismatch (Results; Discussion limitations).
  • ad hoc to paper Iterative compensation converges to a small residual in K=5 iterations for the models considered.
    No convergence proof is given; supported only by empirical residual reduction on 1,633 targets (Results, Fig 2A).

reviewed 2026-08-04 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Fast reconstruction of degenerate populations of conductance-based neuron models from spike times." pith.science (2026). https://pith.science/paper/5S2QEPLV

@misc{pith2026250912783,
  author       = {Pith},
  title        = {Pith review of: Fast reconstruction of degenerate populations of conductance-based neuron models from spike times},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5S2QEPLV}},
  note         = {Machine review of arXiv:2509.12783}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Inferring the biophysical parameters of conductance-based models (CBMs) from experimentally accessible recordings remains a central challenge in computational neuroscience. Spike times are the most widely available data, yet they reveal little about which combinations of ion channel conductances generate the observed activity. This inverse problem is further complicated by neuronal degeneracy, where multiple distinct conductance sets yield similar spiking patterns. We introduce a method that addresses this challenge by combining deep learning with Dynamic Input Conductances (DICs), a theoretical framework that reduces complex CBMs to three interpretable feedback components governing excitability and firing patterns. Our approach first maps spike times to DIC densities at threshold using a neural network that learns a low-dimensional representation of neuronal activity. The predicted DIC values are then used to generate degenerate CBM populations via an iterative compensation algorithm, ensuring compatibility with the intermediate target DICs, and thereby reproducing the corresponding firing patterns, even in high-dimensional models. Applied to two models, this algorithmic pipeline reconstructs spiking and bursting regimes with high accuracy and robustness to variability, including spike trains generated under noisy current injection mimicking physiological stochasticity. It produces diverse degenerate populations within milliseconds on standard hardware, enabling scalable and efficient inference from spike recordings alone. Together, this work positions DICs as a practical and interpretable link between experimentally observed activity and mechanistic models. By enabling fast and scalable reconstruction of degenerate populations directly from spike times, our approach provides a powerful way to investigate how neurons exploit conductance variability to achieve reliable computation.

Figures

Figures reproduced from arXiv: 2509.12783 by Arthur Fyon, Damien Ernst, Guillaume Drion, Julien Brandoit.

Figure 1
Figure 1. Figure 1: Our proposed approach. Spike time sequences are processed by a deep learning model that predicts dynamic input conductances (DICs), a compact representation of the high-dimensional conductance space. These predicted DIC values serve as targets for an iterative compensation algorithm, which explores the degenerate solution space to generate multiple conductance configurations ¯g that reproduce the input spi… view at source ↗
Figure 2
Figure 2. Figure 2: Iterative compensation improves constraint satisfaction and preserves degeneracy in CBMs with [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The synthetic dataset generation process from samling in the DICs space. (A) [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Activity descriptors vary smoothly across the DIC space. [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Architecture predictions and test set distributions in the DIC space [PITH_FULL_IMAGE:figures/full_fig_p021_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Backbone pipeline output for the stomatogastric ganglion (STG) [PITH_FULL_IMAGE:figures/full_fig_p024_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Quantitative comparison of input and generated populations. (A) [PITH_FULL_IMAGE:figures/full_fig_p025_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Comparison of input Poisson spike trains with generated spike train [PITH_FULL_IMAGE:figures/full_fig_p028_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Performance of the adapted pipeline on the DA neuron model. (A) [PITH_FULL_IMAGE:figures/full_fig_p031_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: The deep learning architecture. (A) The input to the model consists of spike time sequences, from which ISIs and delta ISIs are extracted. These features are then stacked and fed into the encoder. (B) The encoder processes the input through three main components: the embedder, the interaction core, and the pooler. The embedder transforms the input sequence into a normalized, higher-dimensional representat… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Neuromodulation supports robust rhythmic pattern transitions in degenerate central pattern generators with fixed connectivity

    math.DS 2026-04 unverdicted novelty 7.0

    An adaptive neuromodulation controller using equivariant bifurcation theory enables robust gait transitions in degenerate central pattern generators with fixed connectivity.

Reference graph

Works this paper leans on

50 extracted references · 15 canonical work pages · cited by 1 Pith paper

  1. [1]

    Regulation of Spike Timing in Visual Cortical Circuits

    Tiesinga P, Fellous JM, Sejnowski TJ. Regulation of Spike Timing in Visual Cortical Circuits. Nature Reviews Neuroscience. 2008;9(2):97-107. arXiv:18200026. doi:10.1038/nrn2315

    doi:10.1038/nrn2315 2008

  2. [2]

    Reliability of Spike Timing in Neocortical Neurons

    Mainen ZF, Sejnowski TJ. Reliability of Spike Timing in Neocortical Neurons. Science (New York, NY). 1995;268(5216):1503-6. arXiv:7770778. doi:10.1126/science.7770778

    doi:10.1126/science.7770778 1995

  3. [3]

    Discovering Precise Temporal Patterns in Large-Scale Neural Recordings through Robust and Interpretable Time Warping

    Williams AH, Poole B, Maheswaranathan N, Dhawale AK, Fisher T, Wilson CD, et al. Discovering Precise Temporal Patterns in Large-Scale Neural Recordings through Robust and Interpretable Time Warping. Neuron. 2020;105(2):246-59.e8. arXiv:31786013. doi:10.1016/j.neuron.2019.10.020

    work page doi:10.1016/j.neuron.2019.10.020 2020

  4. [4]

    Neuromodulation of Neuronal Circuits: Back to the Future

    Marder E. Neuromodulation of Neuronal Circuits: Back to the Future. Neuron. 2012;76(1):1-11. doi:10.1016/j.neuron.2012.09.010

    doi:10.1016/j.neuron.2012.09.010 2012

  5. [5]

    From the Connectome to Brain Function

    Bargmann CI, Marder E. From the Connectome to Brain Function. Nature Methods. 2013 Jun;10(6):483-90. doi:10.1038/nmeth.2451. September 17, 2025 49/54

    doi:10.1038/nmeth.2451 2013

  6. [6]

    Neuromodulation of Brain State and Behavior

    McCormick DA, Nestvogel DB, He BJ. Neuromodulation of Brain State and Behavior. Annual Review of Neuroscience. 2020 Jul;43(Volume 43, 2020):391-415. doi:10.1146/annurev-neuro-100219-105424

    work page doi:10.1146/annurev-neuro-100219-105424 2020

  7. [7]

    Ion Channel Degeneracy, Variability, and Covariation in Neuron and Circuit Resilience

    Goaillard JM, Marder E. Ion Channel Degeneracy, Variability, and Covariation in Neuron and Circuit Resilience. Annual Review of Neuroscience. 2021;44:335-57. doi:10.1146/annurev-neuro-092920-121538

    work page doi:10.1146/annurev-neuro-092920-121538 2021

  8. [8]

    Global Structure, Robustness, and Modulation of Neuronal Models

    Goldman MS, Golowasch J, Marder E, Abbott LF. Global Structure, Robustness, and Modulation of Neuronal Models. Journal of Neuroscience. 2001;21(14):5229-38. arXiv:11438598. doi:10.1523/JNEUROSCI.21-14-05229.2001

    arXiv 2001

  9. [9]

    A Model Neuron with Activity-Dependent Conductances Regulated by Multiple Calcium Sensors

    Liu Z, Golowasch J, Marder E, Abbott LF. A Model Neuron with Activity-Dependent Conductances Regulated by Multiple Calcium Sensors. Journal of Neuroscience. 1998;18(7):2309-20. arXiv:9502792. doi:10.1523/JNEUROSCI.18-07-02309.1998

    arXiv 1998

  10. [10]

    Activity-Dependent Regulation of Potassium Currents in an Identified Neuron of the Stomatogastric Ganglion of the Crab Cancer Borealis

    Golowasch J, Abbott LF, Marder E. Activity-Dependent Regulation of Potassium Currents in an Identified Neuron of the Stomatogastric Ganglion of the Crab Cancer Borealis. Journal of Neuroscience. 1999;19(20):RC33-3. arXiv:10516335. doi:10.1523/JNEUROSCI.19-20-j0004.1999

    arXiv 1999

  11. [11]

    How Multiple Conductances Determine Electrophysiological Properties in a Multicompartment Model

    Taylor AL, Goaillard JM, Marder E. How Multiple Conductances Determine Electrophysiological Properties in a Multicompartment Model. Journal of Neuroscience. 2009;29(17):5573-86. arXiv:19403824. doi:10.1523/JNEUROSCI.4438-08.2009

    work page doi:10.1523/jneurosci.4438-08.2009 2009

  12. [12]

    Yu N, Canavier CC. A Mathematical Model of a Midbrain Dopamine Neuron Identifies Two Slow Variables Likely Responsible for Bursts Evoked by SK Channel Antagonists and Terminated by Depolarization Block. The Journal of Mathematical Neuroscience (JMN). 2015;5(1):5. doi:10.1186/s13408-015-0017-6

    work page doi:10.1186/s13408-015-0017-6 2015

  13. [13]

    A Novel Multiple Objective Optimization Framework for Constraining Conductance-Based Neuron Models by Experimental Data

    Druckmann S, Banitt Y, Gidon AA, Sch¨ urmann F, Markram H, Segev I. A Novel Multiple Objective Optimization Framework for Constraining Conductance-Based Neuron Models by Experimental Data. Frontiers in Neuroscience. 2007 Oct;1. doi:10.3389/neuro.01.1.1.001.2007. September 17, 2025 50/54

    doi:10.3389/neuro.01.1.1.001.2007 2007

  14. [14]

    Alternative to Hand-Tuning Conductance-Based Models: Construction and Analysis of Databases of Model Neurons

    Prinz AA, Billimoria CP, Marder E. Alternative to Hand-Tuning Conductance-Based Models: Construction and Analysis of Databases of Model Neurons. Journal of Neurophysiology. 2003;90(6):3998-4015. doi:10.1152/jn.00641.2003

    arXiv 2003

  15. [15]

    Algorithms for Inverting Hodgkin-Huxley Type Neuron Models [Ph.D

    Shepardson D. Algorithms for Inverting Hodgkin-Huxley Type Neuron Models [Ph.D. Thesis]. Georgia Institute of Technology; 2009. Available from: http://hdl.handle.net/1853/31686

    2009

  16. [16]

    Efficient Estimation of Detailed Single-Neuron Models

    Huys QJM, Ahrens MB, Paninski L. Efficient Estimation of Detailed Single-Neuron Models. Journal of Neurophysiology. 2006 Aug;96(2):872-90. doi:10.1152/jn.00079.2006

    arXiv 2006

  17. [17]

    Feedback Identification of Conductance-Based Models

    Burghi TB, Schoukens M, Sepulchre R. Feedback Identification of Conductance-Based Models. Automatica. 2021 Jan;123:109297. doi:10.1016/j.automatica.2020.109297

    arXiv 2021

  18. [18]

    Distributed Online Estimation of Biophysical Neural Networks

    Burghi TB, O’Leary T, Sepulchre R. Distributed Online Estimation of Biophysical Neural Networks. In: 2022 IEEE 61st Conference on Decision and Control (CDC); 2022. p. 628-34. doi:10.1109/CDC51059.2022.9993018

    arXiv 2022

  19. [19]

    A Sequential Monte Carlo Approach to Estimate Biophysical Neural Models from Spikes

    Meng L, Kramer MA, Eden UT. A Sequential Monte Carlo Approach to Estimate Biophysical Neural Models from Spikes. Journal of Neural Engineering. 2011;8(6):065006. arXiv:22058277. doi:10.1088/1741-2560/8/6/065006

    work page doi:10.1088/1741-2560/8/6/065006 2011

  20. [20]

    A Unified Approach to Linking Experimental, Statistical and Computational Analysis of Spike Train Data

    Meng L, Kramer MA, Middleton SJ, Whittington MA, Eden UT. A Unified Approach to Linking Experimental, Statistical and Computational Analysis of Spike Train Data. PLOS ONE. 2014;9(1):e85269. doi:10.1371/journal.pone.0085269

    work page doi:10.1371/journal.pone.0085269 2014

  21. [21]

    Estimation in the Partially Observed Stochastic Morris–Lecar Neuronal Model with Particle Filter and Stochastic Approximation Methods

    Ditlevsen S, Samson A. Estimation in the Partially Observed Stochastic Morris–Lecar Neuronal Model with Particle Filter and Stochastic Approximation Methods. The Annals of Applied Statistics. 2014 Jun;8(2):674-702. doi:10.1214/14-AOAS729. September 17, 2025 51/54

    work page doi:10.1214/14-aoas729 2014

  22. [22]

    Training Deep Neural Density Estimators to Identify Mechanistic Models of Neural Dynamics

    Gon¸ calves PJ, Lueckmann JM, Deistler M, Nonnenmacher M, ¨Ocal K, Bassetto G, et al. Training Deep Neural Density Estimators to Identify Mechanistic Models of Neural Dynamics. eLife. 2020;9:e56261. doi:10.7554/eLife.56261

    doi:10.7554/elife.56261 2020

  23. [23]

    Dynamic Input Conductances Shape Neuronal Spiking

    Drion G, Franci A, Dethier J, Sepulchre R. Dynamic Input Conductances Shape Neuronal Spiking. eNeuro. 2015;2(1). doi:10.1523/ENEURO.0031-14.2015

    work page doi:10.1523/eneuro.0031-14.2015 2015

  24. [24]

    Dimensionality Reduction of Neuronal Degeneracy Reveals Two Interfering Physiological Mechanisms

    Fyon A, Franci A, Sacr´ e P, Drion G. Dimensionality Reduction of Neuronal Degeneracy Reveals Two Interfering Physiological Mechanisms. PNAS Nexus. 2024 Oct;3(10):pgae415. doi:10.1093/pnasnexus/pgae415

    doi:10.1093/pnasnexus/pgae415 2024

  25. [25]

    Spike2Pop: Bridging Experimental Neuroscience and Computational Modeling [Software]

    Brandoit J. Spike2Pop: Bridging Experimental Neuroscience and Computational Modeling [Software]. GitHub

  26. [26]

    Multiple Models to Capture the Variability in Biological Neurons and Networks

    Marder E, Taylor AL. Multiple Models to Capture the Variability in Biological Neurons and Networks. Nature Neuroscience. 2011;14(2):133-8. doi:10.1038/nn.2735

    doi:10.1038/nn.2735 2011

  27. [27]

    Failure of Averaging in the Construction of a Conductance-Based Neuron Model

    Golowasch J, Goldman MS, Abbott LF, Marder E. Failure of Averaging in the Construction of a Conductance-Based Neuron Model. Journal of Neurophysiology. 2002;87(2):1129-31. doi:10.1152/jn.00412.2001

    arXiv 2002

  28. [28]

    Reliable Neuromodulation from Adaptive Control of Ion Channel Expression

    Fyon A, Sacr´ e P, Franci A, Drion G. Reliable Neuromodulation from Adaptive Control of Ion Channel Expression. IF AC-PapersOnLine. 2023;56(2):458-63. doi:10.1016/j.ifacol.2023.10.1610

    work page doi:10.1016/j.ifacol.2023.10.1610 2023

  29. [29]

    Degenerate Conductance-Based Models Populations Datasets: STG and DA models

    Brandoit J, Ernst D, Drion G, Fyon A. Degenerate Conductance-Based Models Populations Datasets: STG and DA models. Zenodo; 2025. doi:10.5281/zenodo.16912161

    work page doi:10.5281/zenodo.16912161 2025

  30. [30]

    Neuromodulation of Neurons and Synapses

    Nadim F, Bucher D. Neuromodulation of Neurons and Synapses. Current Opinion in Neurobiology. 2014;29:48-56. doi:10.1016/j.conb.2014.05.003

    work page doi:10.1016/j.conb.2014.05.003 2014

  31. [31]

    Input Synchrony and the Irregular Firing of Cortical Neurons

    Stevens CF, Zador AM. Input Synchrony and the Irregular Firing of Cortical Neurons. Nature Neuroscience. 1998;1(3):210-7. doi:10.1038/659. September 17, 2025 52/54

    work page doi:10.1038/659 1998

  32. [32]

    The Highly Irregular Firing of Cortical Cells Is Inconsistent with Temporal Integration of Random EPSPs

    Softky WR, Koch C. The Highly Irregular Firing of Cortical Cells Is Inconsistent with Temporal Integration of Random EPSPs. The Journal of Neuroscience: The Official Journal of the Society for Neuroscience. 1993;13(1):334-50. doi:10.1523/JNEUROSCI.13-01-00334.1993

    arXiv 1993

  33. [33]

    Stimulus Onset Quenches Neural Variability: A Widespread Cortical Phenomenon

    Churchland MM, Yu BM, Cunningham JP, Sugrue LP, Cohen MR, Corrado GS, et al. Stimulus Onset Quenches Neural Variability: A Widespread Cortical Phenomenon. Nature Neuroscience. 2010;13(3):369-78. doi:10.1038/nn.2501

    doi:10.1038/nn.2501 2010

  34. [34]

    Spike Count Reliability and the Poisson Hypothesis

    Amarasingham A, Chen TL, Geman S, Harrison MT, Sheinberg DL. Spike Count Reliability and the Poisson Hypothesis. The Journal of Neuroscience: The Official Journal of the Society for Neuroscience. 2006;26(3):801-9. arXiv:16421300. doi:10.1523/JNEUROSCI.2948-05.2006

    doi:10.1523/jneurosci.2948-05.2006 2006

  35. [35]

    LoRA: Low-Rank Adaptation of Large Language Models

    Hu EJ, Shen Y, Wallis P, Allen-Zhu Z, Li Y, Wang S, et al.. LoRA: Low-Rank Adaptation of Large Language Models. arXiv; 2021. arXiv:2106.09685

    Pith/arXiv arXiv 2021

  36. [36]

    How Modeling Can Reconcile Apparently Discrepant Experimental Results: The Case of Pacemaking in Dopaminergic Neurons

    Drion G, Massotte L, Sepulchre R, Seutin V. How Modeling Can Reconcile Apparently Discrepant Experimental Results: The Case of Pacemaking in Dopaminergic Neurons. PLOS Computational Biology. 2011;7(5):e1002050. doi:10.1371/journal.pcbi.1002050

    work page doi:10.1371/journal.pcbi.1002050 2011

  37. [37]

    Similar Network Activity from Disparate Circuit Parameters

    Prinz AA, Bucher D, Marder E. Similar Network Activity from Disparate Circuit Parameters. Nature Neuroscience. 2004;7(12):1345-52. doi:10.1038/nn1352

    doi:10.1038/nn1352 2004

  38. [38]

    Creating the Feedback Loop: Closed-Loop Neurostimulation

    Hebb AO, Zhang JJ, Mahoor MH, Tsiokos C, Matlack C, Chizeck HJ, et al. Creating the Feedback Loop: Closed-Loop Neurostimulation. Neurosurgery Clinics of North America. 2014;25(1):187-204. arXiv:24262909. doi:10.1016/j.nec.2013.08.006

    work page doi:10.1016/j.nec.2013.08.006 2014

  39. [39]

    Designing, Tuning and Building Ultra-low-power Neuromodulable Mixed-feedback Silicon Neurons

    Mendolia L, Franci A. Designing, Tuning and Building Ultra-low-power Neuromodulable Mixed-feedback Silicon Neurons. In: 44th Benelux Meeting on Systems and Control; 2025

    2025

  40. [40]

    Mathematical Analysis of Depolarization Block Mediated by Slow Inactivation of Fast Sodium Channels in September 17, 2025 53/54 Midbrain Dopamine Neurons

    Qian K, Yu N, Tucker KR, Levitan ES, Canavier CC. Mathematical Analysis of Depolarization Block Mediated by Slow Inactivation of Fast Sodium Channels in September 17, 2025 53/54 Midbrain Dopamine Neurons. Journal of Neurophysiology. 2014;112(11):2779-90. doi:10.1152/jn.00578.2014

    arXiv 2025

  41. [41]

    SciPy 1.0: Fundamental Algorithms for Scientific Computing in Python

    Virtanen P, Gommers R, Oliphant TE, Haberland M, Reddy T, Cournapeau D, et al. SciPy 1.0: Fundamental Algorithms for Scientific Computing in Python. Nature Methods. 2020;17:261-72. doi:10.1038/s41592-019-0686-2

    doi:10.1038/s41592-019-0686-2 2020

  42. [42]

    The MATLAB ODE Suite

    Shampine LF, Reichelt MW. The MATLAB ODE Suite. SIAM Journal on Scientific Computing. 1997;18(1):1-22. doi:10.1137/S1064827594276424

    doi:10.1137/s1064827594276424 1997

  43. [43]

    Attention is all you need

    Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez AN, et al. Attention is all you need. In: Proceedings of the 31st International Conference on Neural Information Processing Systems. NIPS’17. Red Hook, NY, USA: Curran Associates Inc.; 2017. p. 6000–6010

    2017

  44. [44]

    Layer Normalization; 2016

    Ba JL, Kiros JR, Hinton GE. Layer Normalization; 2016. arXiv:1607.06450

    Pith/arXiv arXiv 2016

  45. [45]

    Improving Neural Networks by Preventing Co-adaptation of Feature Detectors

    Hinton GE, Srivastava N, Krizhevsky A, Sutskever I, Salakhutdinov RR. Improving Neural Networks by Preventing Co-adaptation of Feature Detectors

  46. [46]

    Gaussian Error Linear Units (GELUs)

    Hendrycks D, Gimpel K. Gaussian Error Linear Units (GELUs). arXiv; 2023. arXiv:1606.08415

    Pith/arXiv arXiv 2023

  47. [47]

    Robust Estimation of a Location Parameter

    Huber PJ. Robust Estimation of a Location Parameter. The Annals of Mathematical Statistics. 1964;35(1):73-101. doi:10.1214/aoms/1177703732

    arXiv 1964

  48. [48]

    Random Search for Hyper-Parameter Optimization

    Bergstra J, Bengio Y. Random Search for Hyper-Parameter Optimization. J Mach Learn Res. 2012 Feb;13(null):281-305

    2012

  49. [49]

    Decoupled Weight Decay Regularization

    Loshchilov I, Hutter F. Decoupled Weight Decay Regularization. arXiv; 2019. doi:10.48550/arXiv.1711.05101. September 17, 2025 54/54

    Pith/arXiv arXiv doi:10.48550/arxiv.1711.05101 2019

  50. [2025]

    Software available athttps://github.com/julienbrandoit/ Spike2Pop---Bridging-Experimental-Neuroscience-and-Computational-Modeling

    Version 1.1.1. Software available athttps://github.com/julienbrandoit/ Spike2Pop---Bridging-Experimental-Neuroscience-and-Computational-Modeling

This paper was first reviewed by deepseek-v4-flash on August 4, 2026.