A Practical Guide to Tuning Spiking Neuronal Dynamics

Alexander G. Ororbia; Clare Thiem; Jack Lombardi; Nathan McDonald; William Gebhardt

REVIEW 4 major objections 6 minor 24 references

A Practical Guide to Tuning Spiking Neuronal Dynamics

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read The paper claims that tuning a spiking neural network must start by getting the randomly initialized LIF and RAF dynamics into a healthy regime, and offers concrete rules for encodings, time constants, and excitatory-inhibitory wiring to…

desk verdict A clearly framed teaching guide with correct standard equations, but its practical advice rests on under-reported simulations that need parameter tables and quantitative metrics before the guide can be reliably used. read the letter →

arxiv 2506.08138 v1 pith:NIBFPUXF submitted 2025-06-09 cs.NE q-bio.NC

William Gebhardt , Alexander G. Ororbia , Nathan McDonald , Clare Thiem , Jack Lombardi This is my paper

classification cs.NEq-bio.NC

keywords spikingneuralnetworksleakyintegrate-and-fireresonate-and-fireinputencodingexcitatory-inhibitorybalanceneurondynamicstuningneuromorphiccomputingpracticalguide

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that the usual way of building spiking neural networks—picking a neuron model, wiring it up, and then training or evolving it—skips a necessary step: first getting the raw dynamics of the randomly initialized network into a 'healthy' regime. Drawing on two standard neuron models, the leaky integrate-and-fire (LIF) and resonate-and-fire (RAF) cells, it shows how input encodings (Bernoulli, Poisson, phasor, latency), membrane time constants, decay and input coefficients, and excitatory-inhibitory wiring patterns change spike behavior. The authors' practical claim is that if the initial system is degenerate—for instance, emitting a constant stream of spikes from two input pulses—then no later learning or evolution can be expected to recover healthy behavior, so tuning the baseline dynamics is a prerequisite. The payoff, if correct, is a single orientation for newcomers who would otherwise have to piece together tuning rules from dozens of papers.

What carries the argument

The load-bearing objects are the coupled ODEs of the two neuron models. The LIF is described by voltage and current equations with time constants $\tau_v$ and $\tau_j$, leak coefficients $\gamma_v$ and $\gamma_j$, an adaptive threshold $\theta$, and reset dynamics; the RAF is a two-variable damped oscillator with voltage $v$ and angular driving variable $c$, whose eigenfrequency $\omega$ sets the resonant spike-timing window. Around these sit the input encoders—Bernoulli trials, Poisson-scaled Bernoulli, phasors, and latency/RC-circuit codes—and the lateral connectivity matrices between excitatory and inhibitory populations, which the paper treats as the main tuning knobs that determine whether a raster plot shows healthy integration, winner-take-all competition, denoising, or collapse.

What would settle it

Reproduce Figures 2, 3, and 6 from the equations in Section 2.2 using the parameter values in the supplement, and sweep $\tau_v$, $\gamma_v$, $R$, and the E-I ratio around the recommended settings. If the 'healthy' rasters become degenerate under small perturbations—or if a different observer labels the same raster differently—then the paper's central practical guidance is not a stable rule.

Watch

Extended reading notes

Core claim

The central discovery is a working prescription for setting up LIF and RAF circuits so that they show expected, 'healthy' spike dynamics before any training. For LIF neurons, the membrane time constant $\tau_v$, leak coefficient $\gamma_v$, resistance $R$, and adaptive-threshold parameters jointly control whether inputs are integrated cleanly or produce pathological output streams; the paper derives how to set $\tau$ and $\gamma$ so that a neuron decays to a chosen fraction of its voltage in a chosen time step. For RAF neurons, a spike doublet fires the cell only when the inter-spike interval matches the eigenperiod $T_p = 2\pi/\omega$, giving a resonance-detection behavior. In excitatory-inhibitory assemblies, the paper finds that a one-to-one/one-to-many wiring pattern produces winner-take-all dynamics in LIF populations and denoising in RAF populations, while swapping the pattern swaps the effects, and that the biological 80-20 excitatory-to-inhibitory split acts as a global threshold that dampens later excitatory activity.

Load-bearing premise

The load-bearing premise is that 'healthy' versus 'degenerate' dynamics can be recognized visually from raster plots, with no quantitative metric or fully reported parameter setting tying those labels to reproducible conditions; if that classification is subjective or settings-dependent, the recommended tuning rules (raise inhibitory R and tau, use an 80-20 split) may not transfer to other SNN configurations.

Editorial extensions

If this is right

Newcomers can assemble a working randomly initialized SNN by following the encoding, time-constant, and wiring rules, without surveying the literature.
A degenerate initial system should be treated as a red flag: tuning hyperparameters before learning is a prerequisite for later evolution or training to produce expected behavior.
In one-to-one/one-to-many excitatory-inhibitory wiring, LIF populations produce winner-take-all behavior and RAF populations perform denoising; reversing the wiring pattern swaps the two effects.
The 80-20 excitatory-to-inhibitory split functions as an activity threshold, meaning the first wave of excitatory spiking is weighted more heavily than later activity.
RAF neurons respond to spike doublets only when the inter-spike interval matches the eigenperiod $T_p$, so tuning the eigenfrequency selects which temporal patterns the circuit detects.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The healthy/degenerate distinction could be made quantitative—for instance, by measuring the coefficient of variation of inter-spike intervals or the stability of firing rates across time windows—which would let the recommended settings be tested across parameter grids.
The 80-20 threshold behavior suggests a general homeostatic control principle that might also emerge in other network ratios, making the split a candidate mechanism rather than a mere biological convention.
The guide's prescriptions could be tested by swapping in other neuron models, such as Izhikevich or adaptive-exponential LIF, to see whether the healthy/degenerate boundary shifts.
Because the paper does not cover learning, an obvious extension is to verify that healthy baseline dynamics improve STDP or other learning outcomes compared to degenerate baselines.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

First thing to know: this is an openly declared teaching document, not a research contribution. The authors say so in the introduction, and the content matches that framing. The standard equations—Stein's LIF and Izhikevich's RAF—are reproduced accurately, and the ODE algebra in Section 3.1 (time constant / decay coefficient relation) is correct. For a newcomer, the sections on Bernoulli, Poisson, phasor, and latency encoding are a fair, compact orientation, and the E/I wiring patterns are drawn from established references. Credit where due: the honest framing and the correct math are the paper's real assets.

The soft spot is exactly what the stress-test flags: the practical advice—increase inhibitory R and tau relative to excitatory, use the 80-20 wiring as a global threshold, expect WTA or denoising from a given pattern—rests almost entirely on a handful of raster plots (Figures 2, 3, 6) with no reported parameter values, no quantitative metric for 'healthy' vs 'degenerate,' and no repetitions or error bars. That matters for a 'practical guide' because the recommendations are regime-dependent. A newcomer cannot tell whether their network is in the same regime without knowing, say, tau_v, gamma_v, thresholds, input magnitudes, and population sizes. The GitHub supplement is mentioned without a commit hash or manifest, so even the exact code is not pinned. The motivation claim in Section 1—that an unhealthy initial system makes healthy evolved behavior unreasonable—is asserted without evidence; it is plausible but not established.

I would not call these fatal flaws. The paper is a teaching tool, and the verbal descriptions of the dynamics are mostly sound. But the empirical demonstrations are under-reported relative to their load-bearing role. A revision that adds a parameter table for every simulation, a quantitative definition of collapse (e.g., firing rate saturation, CV of inter-spike intervals), and a pinned code release would make this a genuinely useful reference. As it stands, the advice is trustworthy mainly for someone willing to re-derive the regimes themselves.

Who is this for? People starting out in SNN simulation and instructors looking for a compact overview. Not for researchers needing new results. It deserves a serious referee, but only with a request for the missing empirical details.

Referee Report

4 major / 6 minor

Summary. The manuscript presents itself as a practical guide for tuning spiking neural network components: it surveys several input encoding schemes (Bernoulli, Poisson, phasor, latency), derives the standard LIF and RAF ODEs, discusses the effects of membrane time constants and leak/input coefficients, and reports simulations of excitatory–inhibitory (E/I) populations with different connectivity patterns (one-to-one/one-to-many, one-to-many/one-to-one, and an 80-20 split). The central claim is that the paper provides a much-needed orientation for newcomers to build well-working SNN systems without surveying a large body of literature. The empirical sections assert that certain parameter regimes produce 'healthy' or 'degenerate' dynamics, and that tuning prescriptions such as increasing inhibitory resistance and time constant, or using an 80-20 excitatory-to-inhibitory ratio, produce winner-take-all, denoising, or thresholding effects.

Significance. The paper's conceptual scope is genuinely useful: it collects in one place the key equations and qualitative behaviors of LIF and RAF neurons, and the ODE algebra in Section 3.1 is correct. The encoding overview is a reasonable elementary introduction. However, the empirical core that drives the practical advice is not yet at the standard required to support the guidance. The claims about WTA, denoising, and the 80-20 threshold rest on single unparameterized raster plots with no quantitative validation, so a newcomer cannot currently determine the regime in which the advice applies. If the authors add parameter tables, reproducible code with a pinned version, and quantitative metrics with repetitions or sweeps, the paper could become a genuinely useful teaching resource.

major comments (4)

[§3.2, Figures 6e–6h] The central tuning prescriptions—increase inhibitory R and τ relative to excitatory, expect WTA or denoising from a given wiring pattern, and treat the 80-20 connectivity as a threshold—are supported only by single raster plots. The manuscript does not report any parameter values for these experiments: membrane time constants, resistances, thresholds, input magnitudes, simulation length, population sizes, or the random connectivity instances are all absent. This makes the experiments impossible to reproduce and, more importantly for a practical guide, leaves the reader unable to determine whether the advice applies to their own settings. Please provide a complete table of simulation parameters, and add either explicit parameter sweeps or a statement delimiting the regime in which each qualitative behavior was observed.
[§3.2.1] The claims that the one-to-one/one-to-many connectivity pattern produces a winner-take-all effect in LIF populations (Figure 6c) and a denoising effect in RAF populations (Figure 6f), and that the one-to-many/one-to-one pattern swaps these effects (Figures 6d and 6g), are based on a visual comparison of one realization per condition. No quantitative metric (e.g., population firing rate, Fano factor, coefficient of variation, or a sparsity measure) is reported, and no repetitions across random seeds or input sequences are shown. Because the paper's practical value depends on these classifications being robust, the authors should operationalize 'WTA' and 'denoising' and demonstrate that the effects hold across several independent runs.
[§3.2.2] The characterization of the 80-20 connectivity as 'effectively functions as a threshold applied to the total activity from excitatory neurons' is a specific mechanistic claim, yet the evidence consists solely of the two raster plots in Figures 6e and 6h. The text does not quantify excitatory activity before and after inhibitory feedback, nor does it compare the 80-20 pattern to a matched control with a different dropout rate. Without a quantitative comparison, the 'threshold' interpretation is one of several possible explanations for the observed spiking pattern.
[§3.1, Figure 2, and §1] The 'healthy' vs. 'degenerate' dichotomy is the load-bearing criterion for the guide, but it is defined only in qualitative terms: Section 1 calls collapsed systems those 'missing favorable properties of dynamics that might be expected,' and Figure 2 labels one case 'degenerate' because it produces an 'unexpected, constant stream of output spikes emitted from just two input pulses.' No formal definition, threshold, or parameter values are given for either regime. Furthermore, the motivating statement in Section 1—that a non-functioning initial system makes it unreasonable to expect healthy evolved behavior—is asserted without evidence. Please either provide a quantitative definition of healthy/degenerate dynamics (e.g., in terms of firing-rate stability or response selectivity) and the simulation parameters used, or reframe the guide's advice as conditional on a user-defined target behavior.

minor comments (6)

[§2.1.2, Figure 5] There are typographical errors: 'Benoulli' should be 'Bernoulli' in Section 2.1.2, and 'inhbitory' should be 'inhibitory' in the caption of Figure 5.
[Eq. (8)] Equation (8) is missing the velocity update step; as written, it only shows the wrap-around comparison and subtraction, but the text says the angle is updated by adding v_n. Please make the update rule explicit in the equation.
[Eq. (2)] Equation (2) uses x for both the Bernoulli probability and the outcome variable; using p for the probability would avoid confusion.
[Figure 6] The y-axes of the raster plots are not labeled; presumably they show neuron index, but this should be stated in the caption or axis labels. Also, the individual panels are not referenced in the body text with their letters, making it harder for the reader to locate the relevant plot.
[Title page vs. second page] The author list is inconsistent: the first page lists 'Clare Thiem' while the second page lists 'Clare Theim'.
[§3.2.2] The phrase 'the connection matrices are always densely connected with a roughly 20% dropout rate' is ambiguous—does this mean 80% of connections are present, and why is this 'always' the case? Please clarify the generative rule for the random connectivity and whether 'always' refers to the paper's simulations or a general property.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's claims are direct observations of its own defined dynamical models, and its self-citations are background references that do not enter the derivation.

full rationale

The paper makes no fitted predictions and derives no result from the claim it is trying to establish. Its central content is a description of standard encoding schemes and of LIF/RAF neuron models given by explicit ODEs (Eqs. 11-18), followed by simulations whose behavior is read directly off raster plots (Figures 2, 3, 6). There is no equation or parameter that is defined in terms of the output it is supposed to explain, and no fitted parameter is relabeled as a prediction. The few self-citations ([9], [15], [18], [19]) are cited as examples of SNN efforts or for the decision to exclude learning/evolution; none is load-bearing for the tuning advice. The 80-20 ratio is justified by external biological references ([16], [7], [1], [21]), not by a self-citation. The 'healthy' versus 'degenerate' classification is qualitative and the experimental parameters are unreported, but that is a reproducibility and rigor concern, not a circularity: the classifications are observations of the simulations, not consequences of the definitions. Consequently, no circular step can be exhibited, and the derivation chain is self-contained as far as circularity is concerned.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The guide contributes no new free parameters or entities; it relies on standard neuron models and encoding schemes. The hidden dependency is that all empirical conclusions depend on unreported simulation settings and subjective visual classification, which limits validation.

free parameters (4)

Membrane time constant(s) for LIF healthy/degenerate comparison = not reported
Figure 2 and Section 3.1 compare settings that produce 'healthy' and 'degenerate' behavior; the actual tau_v values are absent, but the comparison depends on them.
Resistance (R) and time constant (tau) of inhibitory population = not reported
Section 3.2.1.1 recommends increasing R and tau for the inhibitory structure relative to excitatory, without giving values or ranges.
RAF eigenfrequency omega and dampening factor b = not reported
Figure 3's resonance demonstration requires specific omega and b; these are not listed, so the period T_p = 2*pi/omega is unknown.
Population sizes and dropout rate for 80-20 connectivity = not reported
Section 3.2.2 states connectivity is dense with roughly 20% dropout, but the number of neurons and exact dropout are unspecified.

assumptions (4)

domain assumption LIF and RAF ODEs from Stein (1965) and Izhikevich (2001) accurately model spiking neuron dynamics.
Section 2.2 adopts these models as foundational without re-deriving or comparing against alternative neuron models.
domain assumption Poisson-distributed spiking is a desirable target for input encodings.
Section 2.1 motivates encodings via neurobiological support from one study in area MT; the paper treats Poisson behavior as the design goal.
domain assumption The 80/20 excitatory-inhibitory ratio is a relevant biological baseline.
Section 3.2.2 cites neocortex and mouse brain proportions; the paper assumes this ratio is the appropriate thing to test in artificial SNNs.
ad hoc to paper Visual inspection of raster plots reliably distinguishes healthy from collapsed dynamics.
All behavioral conclusions in Section 3 are drawn from raster plots without quantitative definitions of 'healthy' or 'degenerate'.

how reviews work

0 comments

Cite this review

Pith. "Pith review of A Practical Guide to Tuning Spiking Neuronal Dynamics." pith.science (2026). https://pith.science/paper/NIBFPUXF

@misc{pith2026250608138,
  author       = {Pith},
  title        = {Pith review of: A Practical Guide to Tuning Spiking Neuronal Dynamics},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/NIBFPUXF}},
  note         = {Machine review of arXiv:2506.08138}
}

read the original abstract

In this work, we examine fundamental elements of spiking neural networks (SNNs) as well as how to tune them. Concretely, we focus on two different foundational neuronal units utilized in SNNs -- the leaky integrate-and-fire (LIF) and the resonate-and-fire (RAF) neuron. We explore key equations and how hyperparameter values affect behavior. Beyond hyperparameters, we discuss other important design elements of SNNs -- the choice of input encoding and the setup for excitatory-inhibitory populations -- and how these impact LIF and RAF dynamics.

Figures

Figures reproduced from arXiv: 2506.08138 by the authors.

**Figure 1.** A latency-encoding spike train; probability values for ten input neurons were incrementally generated (through steps of 0.1) within the range [0.1,1.]. As the number of neurons in the layer increase, their individual expected firing rates will converge to be a Poisson distribution. Using each phasor’s actual firing rate, the phasor’s angular velocity can then be calculated in the following manner: vn = 2πfn = 2π∗ ∆t… view at source ↗

**Figure 2.** Latency-LIF coupled dynamics. Above is a visual of the dynamics of a four-cell system – 3 latencycoding input cells synaptically wired into 1 LIF cell – under two different time constant settings (left shows a “healthy” setting; right shows a “degenerate” setting). Both cases show the scenario where two initial latencycoded spikes occur near one another with the last one lagging behind. Notice, in the “healthy” se… view at source ↗

**Figure 3.** Resonate-and-Fire dynamics. Here, we visualize the dynamics of a resonate-and-fire (RAF) neuron under two important conditions: 1) a pair of input spikes, also known as a spike doublet, that are not resonant with the RAF unit (the interspike interval between the two pulses is not equal to the period of the RAF neuron, i.e., T = 2π/ω) will not push the neuron beyond its threshold, and 2) a doublet that is resonant wi… view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: Comparison of different time constants (τ) and their effect on decay rate for exponentially decaying values. The Membrane Time Constant: To gather an intuition behind membrane time constants – and time constants in general – in relation to the ODEs commonly found in co…

**Figure 5.** Figure 5: A visualization of various synaptic connectivity wiring patterns between a population of excitatory and inhbitory LIF cells: (Left) a one-to-one, one-to-many structure, (Middle) a one-to-many, one-to-one structure, and (Right) a sparsely-connected, random structure (wi…

**Figure 6.** Figure 6: The effect that lateral synaptic connectivity – one-to-one/one-to-many, one-to-many/one-to-one, and sparse 80:20 split connectivity (with 80:20 ratio of excitatory-to-inhibitory units) – has on dynamics of two populations (one excitatory & one inhibitory) of LIFs and R…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

24 extracted references · 22 canonical work pages

[1]

ALREJA, A., NEMENMAN, I.,ANDROZELL, C. J. Constrained brain volume in an efficient coding model explains the fraction of excitatory and inhibitory neurons in sensory cortices.PLOS Computational Biology 18, 1 (2022), e1009642

work page 2022
[2]

Power spectrum analysis of bursting cells in area mt in the behaving monkey.Journal of Neuroscience 14, 5 (1994), 2870–2892

BAIR, W., KOCH, C., NEWSOME, W.,ANDBRITTEN, K. Power spectrum analysis of bursting cells in area mt in the behaving monkey.Journal of Neuroscience 14, 5 (1994), 2870–2892

work page 1994
[3]

BI, G.-Q.,ANDPOO, M.-M. Synaptic modifications in cultured hippocampal neurons: depen- dence on spike timing, synaptic strength, and postsynaptic cell type.Journal of neuroscience 18, 24 (1998), 10464–10472. 2The more general principle is that there are more excitatory neurons in relation to inhibitory ones where some ratios that have also been examined i...

work page 1998
[4]

BURKITT, A. N. A review of the integrate-and-fire neuron model: I. homogeneous synaptic input. Biological cybernetics 95(2006), 1–19

work page 2006
[5]

M.,ANDYOUNG, E

CHASE, S. M.,ANDYOUNG, E. D. First-spike latency information in single neurons increases when referenced to population onset.Proceedings of the National Academy of Sciences 104, 12 (2007), 5175–5180

work page 2007
[6]

Spike timing-dependent plasticity of neural circuits.Neuron 44, 1 (2004), 23–30

DAN, Y.,ANDPOO, M.-M. Spike timing-dependent plasticity of neural circuits.Neuron 44, 1 (2004), 23–30

work page 2004
[7]

L., MANTINI, D.,ANDCORBETTA, M

DECO, G., PONCE-ALVAREZ, A., HAGMANN, P ., ROMANI, G. L., MANTINI, D.,ANDCORBETTA, M. How local excitation–inhibition ratio impacts the whole brain dynamics.Journal of Neuroscience 34, 23 (2014), 7886–7898

work page 2014
[8]

U.,ANDCOOK, M

DIEHL, P . U.,ANDCOOK, M. Unsupervised learning of digit recognition using spike-timing- dependent plasticity.Frontiers in computational neuroscience 9(2015), 99

work page 2015

Show all 24 references

[9]

GEBHARDT, W.,ANDORORBIA, A. G. Time-integrated spike-timing-dependent-plasticity.arXiv preprint arXiv:2407.10028(2024)

2024 arXiv
[10]

Spiking neurons.Pulsed neural networks(1998), 3–53

GERSTNER, W. Spiking neurons.Pulsed neural networks(1998), 3–53

1998
[11]

IZHIKEVICH, E. M. Resonate-and-fire neurons.Neural networks 14, 6-7 (2001), 883–894

2001
[12]

IZHIKEVICH, E. M. Polychronization: computation with spikes.Neural computation 18, 2 (2006), 245–282

2006
[13]

Computing with spiking neurons.Pulsed neural networks 2(1999), 55–85

MAASS, W. Computing with spiking neurons.Pulsed neural networks 2(1999), 55–85

1999
[14]

The performance (and limits) of simple neuron models: general- izations of the leaky integrate-and-fire model

NAUD, R.,ANDGERSTNER, W. The performance (and limits) of simple neuron models: general- izations of the leaky integrate-and-fire model. InComputational systems neurobiology. Springer, 2012, pp. 163–192

2012
[15]

W., GEBHARDT, W., TEULIÈRE, C., ZELDENRUST, F ., RAO, R

N’DRI, A. W., GEBHARDT, W., TEULIÈRE, C., ZELDENRUST, F ., RAO, R. P ., TRIESCH, J.,ANDOROR- BIA, A. Predictive coding with spiking neural networks: a survey.arXiv preprint arXiv:2409.05386 (2024)

2024 arXiv
[16]

R., RUGGIERO, D

NOBACK, C. R., RUGGIERO, D. A., STROMINGER, N. L.,ANDDEMAREST, R. J.The human nervous system: structure and function. Springer Science & Business Media, 2005

2005
[17]

Y., KIM, J.,ANDLEE, J.-H

OH, S., KWON, D., YEOM, G., KANG, W.-M., LEE, S., WOO, S. Y., KIM, J.,ANDLEE, J.-H. Neuron circuits for low-power spiking neural networks using time-to-first-spike encoding.IEEE Access 10 (2022), 24444–24455

2022
[18]

Spiking neural predictive coding for continually learning from data streams.Neuro- computing 544(2023), 126292

ORORBIA, A. Spiking neural predictive coding for continually learning from data streams.Neuro- computing 544(2023), 126292

2023
[19]

ORORBIA, A. G. Brain-inspired machine intelligence: A survey of neurobiologically-plausible credit assignment.arXiv preprint arXiv:2312.09257(2023). Page 17 of 18 NAC TR 2020-00X A Practical Guide to Tuning Spiking Neuronal Dynamics

2023 arXiv
[20]

K.,ANDUNNIKRISHNAN, N

PARHI, K. K.,ANDUNNIKRISHNAN, N. K. Brain-inspired computing: Models and architectures. IEEE Open Journal of Circuits and Systems 1(2020), 185–204

2020
[21]

A method to estimate the cellular composition of the mouse brain from heterogeneous datasets.PLOS Computational Biology 18, 12 (2022), e1010739

RODARIE, D., VERASZTÓ, C., ROUSSEL, Y., REIMANN, M., KELLER, D., RAMASWAMY, S., MARKRAM, H.,ANDGEWALTIG, M.-O. A method to estimate the cellular composition of the mouse brain from heterogeneous datasets.PLOS Computational Biology 18, 12 (2022), e1010739

2022
[22]

STEIN, R. B. A theoretical analysis of neuronal variability.Biophysical journal 5, 2 (1965), 173–194

1965
[23]

Spiking neural networks and their applications: A review.Brain sciences 12, 7 (2022), 863

YAMAZAKI, K., VO-HO, V.-K., BULSARA, D.,ANDLE, N. Spiking neural networks and their applications: A review.Brain sciences 12, 7 (2022), 863

2022
[24]

Synaptic ei balance underlies efficient neural coding.Frontiers in neuro- science 12(2018), 46

ZHOU, S.,ANDYU, Y. Synaptic ei balance underlies efficient neural coding.Frontiers in neuro- science 12(2018), 46. Page 18 of 18

2018

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

ALREJA, A., NEMENMAN, I.,ANDROZELL, C. J. Constrained brain volume in an efficient coding model explains the fraction of excitatory and inhibitory neurons in sensory cortices.PLOS Computational Biology 18, 1 (2022), e1009642

work page 2022

[2] [2]

Power spectrum analysis of bursting cells in area mt in the behaving monkey.Journal of Neuroscience 14, 5 (1994), 2870–2892

BAIR, W., KOCH, C., NEWSOME, W.,ANDBRITTEN, K. Power spectrum analysis of bursting cells in area mt in the behaving monkey.Journal of Neuroscience 14, 5 (1994), 2870–2892

work page 1994

[3] [3]

BI, G.-Q.,ANDPOO, M.-M. Synaptic modifications in cultured hippocampal neurons: depen- dence on spike timing, synaptic strength, and postsynaptic cell type.Journal of neuroscience 18, 24 (1998), 10464–10472. 2The more general principle is that there are more excitatory neurons in relation to inhibitory ones where some ratios that have also been examined i...

work page 1998

[4] [4]

BURKITT, A. N. A review of the integrate-and-fire neuron model: I. homogeneous synaptic input. Biological cybernetics 95(2006), 1–19

work page 2006

[5] [5]

M.,ANDYOUNG, E

CHASE, S. M.,ANDYOUNG, E. D. First-spike latency information in single neurons increases when referenced to population onset.Proceedings of the National Academy of Sciences 104, 12 (2007), 5175–5180

work page 2007

[6] [6]

Spike timing-dependent plasticity of neural circuits.Neuron 44, 1 (2004), 23–30

DAN, Y.,ANDPOO, M.-M. Spike timing-dependent plasticity of neural circuits.Neuron 44, 1 (2004), 23–30

work page 2004

[7] [7]

L., MANTINI, D.,ANDCORBETTA, M

DECO, G., PONCE-ALVAREZ, A., HAGMANN, P ., ROMANI, G. L., MANTINI, D.,ANDCORBETTA, M. How local excitation–inhibition ratio impacts the whole brain dynamics.Journal of Neuroscience 34, 23 (2014), 7886–7898

work page 2014

[8] [8]

U.,ANDCOOK, M

DIEHL, P . U.,ANDCOOK, M. Unsupervised learning of digit recognition using spike-timing- dependent plasticity.Frontiers in computational neuroscience 9(2015), 99

work page 2015