Pith. sign in

REVIEW 3 major objections 6 minor 44 references

Noise Adaptor: Enhancing Low-Latency Spiking Neural Networks through Noise-Injected Low-Bit ANN Conversion

T0 review · 3 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read Adding a uniform random jitter to the ANN activation quantizer before clip-and-round, called Noise Adaptor, makes converted low-latency spiking networks more accurate, reaching 96.72% on CIFAR-10 and 77.07% on ImageNet without runtime…

desk verdict Noise-injected quantization empirically improves low-latency SNN conversion, but the claimed mechanism is unsupported; the controlled comparison justifies peer review. read the letter →

arxiv 2411.17431 v1 pith:U3OECARN submitted 2024-11-26 cs.NE

Chen Li , Bipin.Rajendran This is my paper · ORCID
classification cs.NE
keywords spikingneuralnetworksANN-to-SNNconversionlow-latencySNNactivationquantizationnoiseinjectionstochasticroundingquantization-awaretrainingResNet
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

The paper sets out to show that a one-line change to how low-bit artificial networks are trained can fix the main accuracy loss in ANN-to-SNN conversion: adding a uniform random value to the signal just before it is rounded during training. This makes the quantizer stochastic, and the paper argues this mirrors the occasional spike-noise that appears when the trained ANN is simulated as a spiking network. The claimed result is that converted spiking networks run accurately at short simulation times without runtime noise correction, and that the method scales to activation-quantized ResNet-101 and ResNet-152. On CIFAR-10 the converted ResNet-18 reaches 96.72% at T=16; on ImageNet the converted ResNet-50 reaches 77.07% at T=256.

What carries the argument

The central object is the noise-injected activation quantizer, called Noise Adaptor: instead of $\hat{v}^l = s_l \lfloor \mathrm{clip}(v^l/s_l, 0, p)\rceil$, the forward pass uses $\hat{v}^l = s_l \lfloor \mathrm{clip}(v^l/s_l + \epsilon, 0, p)\rceil$ with $\epsilon \sim U(-0.5, 0.5)$, and gradients are obtained with a straight-through estimator plus a trainable scale factor $s_l$. The mechanism does two jobs at once: the noise creates a Bernoulli rounding rule whose up-probability equals the fractional distance to the next quantization level, which is the paper's model of how occasional noise perturbs a spiking neuron's spike count, and the expectation of the quantizer equals the clipped-ReLU response curve of a spiking neuron at long simulation times, so the ANN is implicitly trained toward the SNN's target response.

What would settle it

Run a converted SNN at a fixed time step, record each neuron's spike-count deviation from the value the ANN quantizer predicts for its pre-activation, and condition those deviations on the fractional distance to the next quantization level. The paper's model predicts that the probability of a one-count deviation equals that fractional distance; if the measured probabilities deviate systematically, or if deviations do not depend on fractional distance at all, the noise-injection premise is wrong.

Watch

Extended reading notes

Core claim

The central claim is that training an activation-quantized ANN with $\epsilon \sim U(-0.5, 0.5)$ added to $v^l/s_l$ before the clip-and-round step, as in Eqs. (5)-(6), produces a network whose conversion to a spiking network is more accurate at low latency than conversion of the same architecture trained by ordinary deterministic quantization. The injected noise makes the rounding decision probabilistic: a pre-activation whose fractional distance to the next quantization level is $w$ rounds up with probability $w$, so the expected quantizer output is exactly the clipped linear function $\mathrm{clip}(v^l,0,s_l p)$ (Eq. 10, proven in the supplement). The paper interprets this as training the ANN to be robust to spike-count changes of one, matching the behavior of occasional noise in spiking neurons, and reports consistent accuracy improvements over the no-noise baseline across CIFAR-10 and ImageNet at every tested time step.

Load-bearing premise

The method's success rests on the assumption that the random spike-count flutter observed in a low-latency spiking network is faithfully represented by adding uniform noise to the value being rounded during ANN training; if the real flutter follows a different rule, the injected noise tunes the network for the wrong perturbation.

Editorial extensions

If this is right

  • A network trained with Noise Adaptor reaches higher SNN accuracy at every tested time step than the same architecture trained by deterministic activation quantization, so low-latency SNN inference no longer depends on runtime noise correction for the reported settings.
  • Because runtime correction such as negative spikes or offset spikes can be omitted, the converted SNN keeps the standard integrate-and-fire neuron and a single-pass control flow, simplifying deployment on neuromorphic hardware.
  • The method extends conversion to activation-quantized ResNet-101 and ResNet-152; these very deep models require more time steps to approach ANN accuracy, while ResNet-34 and ResNet-50 already give strong accuracy at T=32 and T=64 respectively.
  • Max-pooling pre-trained ANNs can be used for SNN conversion by swapping in average-pooling during noise-aware quantization training, so existing full-precision checkpoints remain usable.
  • ANN accuracy slightly drops when Noise Adaptor is applied, yet SNN accuracy rises, indicating that raw ANN accuracy is not the right proxy for conversion quality.

Reading between the lines

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

  • If the benefit comes from making the quantizer unbiased, then the specific choice of U(-0.5,0.5) may be replaceable by any zero-mean noise of the same support; a direct comparison against standard stochastic rounding would separate the paper's spike-dynamics explanation from the simpler unbiased-expectation effect.
  • The paper's analysis suggests the advantage should be largest when the simulation time T is larger than the quantization level count p but not yet so large that the SNN response curve is smooth; plotting the accuracy gap against T/p would test the implicit-optimization story.
  • The compatibility experiments show that runtime corrections still help at very early time steps, so a hybrid recipe of noise-injected training plus cheap correction only for very small T could deliver both ultra-low latency and high accuracy, a combination the paper does not emphasize.
  • Because the method changes only the training-time quantizer and is pooling-agnostic, it should transfer to other conversion-based tasks such as object detection or transformer-based SNNs, though the paper does not demonstrate this.
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.

Referee Report

3 major / 6 minor

Summary. The paper proposes Noise Adaptor, a training-time modification for activation-quantized ANNs used in ANN-to-SNN conversion. Uniform noise epsilon ~ U(-0.5,0.5) is added to the pre-activation before clipping and rounding (Eqs. 5-6), while keeping the learnable quantization scale. The authors report that this improves the accuracy of converted low-latency SNNs on CIFAR-10 and ImageNet, avoids runtime noise-correction techniques, and scales to ResNet-101/152. A supplementary proof establishes that the noisy quantizer's expected output equals clip(v,0,sp), the rectified-ReLU response in the limit of large simulation time.

Significance. If the empirical results are reproducible, the method offers a simple and effective drop-in modification to quant-ANN-to-SNN conversion pipelines, backed by a clean controlled evaluation in Table 1 (with/without noise injection under the same pipeline). The supplement's proof of Eq. (10) is a correct self-contained derivation of the stochastic-rounding unbiasedness property, and the experiments cover networks up to ResNet-152, addressing a scalability gap in the literature. The main weakness is that the theoretical analysis only supports the large-T limit, while the method's practical gains appear at small T, so the mechanistic explanation of the low-latency benefit remains unverified.

major comments (3)
  1. [§4.4, Eq. (10), Figure 2] The theoretical support does not cover the operating regime where the method's gains are largest. Eq. (10) shows that the expected value of the noise-injected quantizer equals clip(v,0,sp), which is the response curve of the SNN only as T grows large. At small T, the SNN response is a staircase (Figure 2), so the unbiasedness theorem does not imply that the ANN quantizer matches the actual SNN response. Moreover, because the training loss is nonlinear in activations, unbiased activations do not imply an unbiased loss or optimal conversion. Since the largest accuracy improvements are reported at low latency (CIFAR-10 T=2: +1.61%; ImageNet T=8: +3.14% in Table 1), the paper's central claim that noise injection models occasional noise during low-latency conversion is not supported by the provided theory. The authors should either provide a theory or measurements covering finite T, or explicitly restrict the theoretical claim to large T and present the low-T gains as empirical.
  2. [§4.2, Eqs. (5)-(6)] The noise model's central assumption is unmeasured. The paper assumes that the probability of a spike-count transition in the SNN depends on the fractional distance to the next quantization level, and models this by adding uniform noise before rounding. No experiment in the paper verifies that the actual occasional-noise statistics in the converted SNN match this model. Without such validation, the reported gains could equally be explained by stochastic rounding acting as a regularizer. I recommend adding a diagnostic experiment that compares the distribution of predicted state transitions (from Eq. (5)) with measured spike-count fluctuations during SNN inference, or tempering the causal language throughout Section 4.
  3. [§5.3, Table 2] The cross-method comparison mixes architectures. The claim that Noise Adaptor 'consistently achieves the highest SNN accuracy across all time steps' is supported by comparing Noise Adaptor on ResNet-50/101/152 with QCFS and SlipReLU on ResNet-34. The only matched architecture for all three methods is ResNet-34, where Noise Adaptor starts from a lower ANN accuracy (71.26%) yet still attains good SNN accuracy; this should be highlighted and the cross-architecture comparison should be clearly qualified. In addition, Table 1 omits the without-NA baseline at T=256 for ImageNet, so the headline 77.07% figure is not directly compared to the same pipeline without noise injection at that time step.
minor comments (6)
  1. [Section 2] The word 'catagory' should be 'category'.
  2. [Table 2] The entry '36,31%' for Noise Adaptor ResNet-34 at T=16 should be '36.31%'.
  3. [Section 4.2] The sentence 'like doubling it to ϵ ∼ U(−0.5, 0.5)' is confusing; if the base noise is U(−0.5, 0.5), doubling the amplitude would yield U(−1, 1). Please correct the example.
  4. [Algorithm 1] The expression for the scale gradient in line 14 is hard to read; the intended division by sqrt(len(x_clip) * p) (or similar) should be written with explicit parentheses.
  5. [Section 4.3] The heading 'Updations on the Conversion Process' should be 'Updates to the Conversion Process'.
  6. [Table 1] The column headers 'CIFAR-10 ANN T T T' and 'ImageNet ANN T T T' are ambiguous; please list the T values explicitly, e.g., 'ANN, T=1, T=2, T=4, T=8, T=16'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the unbiasedness theorem is a self-contained stochastic-rounding identity, and the reported accuracy gains are empirically evaluated against internal and external baselines.

full rationale

The derivation chain in Section 4.4 and the Supplementary Material is self-contained: Theorem 1 proves E[sl floor(clip(vl/sl + epsilon, 0, p))] = clip(vl, 0, slp) for epsilon ~ U(-0.5, 0.5). This is a standard stochastic-rounding identity and does not assume that Noise Adaptor improves SNN accuracy. The claimed accuracy gains are supported empirically by Table 1, an internal controlled comparison with and without Noise Adaptor under otherwise identical quantization-aware training, and by external comparisons in Table 2. The choice of epsilon ~ U(-0.5, 0.5) is a stated modeling assumption rather than a fitted parameter renamed as a prediction; the trainable scale factors are optimized for classification loss, and the reported SNN accuracies are measured outcomes, not predictions derived from those fitted scales. Self-citations, such as [22] for runtime negative-spike correction and [23] for fast SNNs, are background references and are not load-bearing: no central premise or uniqueness claim is justified solely by the authors' prior work. The concern that the Section 4.2 noise rationale is not directly verified in the low-latency regime is a correctness or evidence limitation, not a circularity of the derivation. Therefore no circular step can be exhibited, and the paper is not circular.

Assumptions & free parameters 2 free parameters · 5 assumptions · 0 invented entities

The method introduces no new physical entities. It relies on two tuned hyperparameters (noise amplitude and quantization bound p) and several domain assumptions about how ANN training noise transfers to spiking neuron behavior. The main mathematical claim is a standard expected-value property.

free parameters (2)
  • Noise amplitude (uniform range) = U(-0.5,0.5)
    Chosen by hand to model +/-1 spike-count jumps. The paper notes that larger amplitudes destabilize ImageNet training, so this is a tuned hyperparameter.
  • Quantization upper bound p = p=2 for CIFAR-10, p=3 for ImageNet
    Selected per dataset; it controls the number of quantization states and directly affects the latency-accuracy tradeoff shown in Figure 3.
assumptions (5)
  • standard math Uniform additive noise before clipping and rounding makes the expected quantizer output equal clip(v,0,sp)
    Proved in the supplementary; this is a property of stochastic rounding.
  • domain assumption Occasional noise in IF SNNs manifests as +/-1 spike-count deviations with probability increasing as the input nears a firing level
    Assumed in Section 4.2 to justify the noise model; no measurement or proof is provided.
  • domain assumption Training with a stochastic quantizer whose expectation matches the SNN response curve implicitly optimizes SNN accuracy
    Section 4.4 asserts this; the theorem only establishes unbiasedness of the quantizer, not equivalence of the training objective.
  • domain assumption Replacing max-pooling with average-pooling before quantization-aware training does not harm convergence
    Empirical claim in Section 4.3 with no controlled comparison given.
  • domain assumption The IF neuron with reset-by-subtraction and 0.5th membrane pre-charge compensates for rounding versus flooring mismatch
    Adopted from cited prior work (QCFS and related methods) in Section 3.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Noise Adaptor: Enhancing Low-Latency Spiking Neural Networks through Noise-Injected Low-Bit ANN Conversion." pith.science (2026). https://pith.science/paper/U3OECARN

@misc{pith2026241117431,
  author       = {Pith},
  title        = {Pith review of: Noise Adaptor: Enhancing Low-Latency Spiking Neural Networks through Noise-Injected Low-Bit ANN Conversion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/U3OECARN}},
  note         = {Machine review of arXiv:2411.17431}
}
read the original abstract

We present Noise Adaptor, a novel method for constructing competitive low-latency spiking neural networks (SNNs) by converting noise-injected, low-bit artificial neural networks (ANNs). This approach builds on existing ANN-to-SNN conversion techniques but offers several key improvements: (1) By injecting noise during quantized ANN training, Noise Adaptor better accounts for the dynamic differences between ANNs and SNNs, significantly enhancing SNN accuracy. (2) Unlike previous methods, Noise Adaptor does not require the application of run-time noise correction techniques in SNNs, thereby avoiding modifications to the spiking neuron model and control flow during inference. (3) Our method extends the capability of handling deeper architectures, achieving successful conversions of activation-quantized ResNet-101 and ResNet-152 to SNNs. We demonstrate the effectiveness of our method on CIFAR-10 and ImageNet, achieving competitive performance. The code will be made available as open-source.

Figures

Figures reproduced from arXiv: 2411.17431 by the authors.

Figure 1
Figure 1. a. In the standard quant-ANN-to-SNN conversion pro￾cess, ANNs are trained with activation quantization to model the discrete spikes characteristic of SNNs. b. Our method goes further by introducing controllable noise into the activation quantization, optimizing the training process for both quantization accuracy and noise resilience. SNN compared to those trained without such constraints [3, 22, 29]. A notable chall… view at source ↗
Figure 2
Figure 2. The response curve of spiking neurons for time steps [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. SNN accuracy over time steps for ResNet-50 with quan [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 34 canonical work pages

  1. [1]

    Estimating or propagating gradients through stochastic neurons for conditional computation

    Yoshua Bengio, Nicholas L ´eonard, and Aaron Courville. Estimating or propagating gradients through stochastic neurons for conditional computation. arXiv preprint arXiv:1308.3432, 2013. 4

    arXiv 2013

  2. [2]

    Towards a bio-inspired real-time neuromorphic cerebellum

    Petrut ¸ A Bogdan, Beatrice Marcinn `o, Claudia Casellato, Stefano Casali, Andrew GD Rowley, Michael Hopkins, Francesco Leporati, Egidio D’Angelo, and Oliver Rhodes. Towards a bio-inspired real-time neuromorphic cerebellum. Frontiers in cellular neuroscience, 15:622870, 2021. 3

    work page 2021

  3. [3]

    Optimal ann-snn conversion for high- accuracy and ultra-low-latency spiking neural networks

    Tong Bu, Wei Fang, Jianhao Ding, PengLin Dai, Zhaofei Yu, and Tiejun Huang. Optimal ann-snn conversion for high- accuracy and ultra-low-latency spiking neural networks. In International Conference on Learning Representations ,

    work page

  4. [4]

    Neuromorphic computing using non-volatile memory

    Geoffrey W Burr, Robert M Shelby, Abu Sebastian, Sang- bum Kim, Seyoung Kim, Severin Sidler, Kumar Virwani, Masatoshi Ishii, Pritish Narayanan, Alessandro Fumarola, et al. Neuromorphic computing using non-volatile memory. Advances in Physics: X, 2(1):89–124, 2017. 3

    work page 2017

  5. [5]

    Training and deploying spiking nn applications to the mixed-signal neuromorphic chip dynap-se2 with rockpool

    U ˘gurcan C ¸ akal, Ilkay Ulusoy, and Dylan R Muir. Training and deploying spiking nn applications to the mixed-signal neuromorphic chip dynap-se2 with rockpool. arXiv preprint arXiv:2303.12167, 2023. 1

    arXiv 2023

  6. [6]

    Nanoscale room-temperature multilayer skyrmionic synapse for deep spiking neural networks

    Runze Chen, Chen Li, Yu Li, James J Miles, Giacomo Indiveri, Steve Furber, Vasilis F Pavlidis, and Christo- foros Moutafis. Nanoscale room-temperature multilayer skyrmionic synapse for deep spiking neural networks. Phys- ical Review Applied, 14(1):014096, 2020. 3

    work page 2020

  7. [7]

    Imagenet: A large-scale hierarchical image database

    Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009. 6

    2009

  8. [8]

    Optimal conversion of conven- tional artificial neural networks to spiking neural networks

    Shikuang Deng and Shi Gu. Optimal conversion of conven- tional artificial neural networks to spiking neural networks. arXiv preprint arXiv:2103.00476, 2021. 1, 2

    arXiv 2021

Show all 44 references
  1. [9]

    Fast-classifying, high- accuracy spiking deep networks through weight and thresh- old balancing

    Peter U Diehl, Daniel Neil, Jonathan Binas, Matthew Cook, Shih-Chii Liu, and Michael Pfeiffer. Fast-classifying, high- accuracy spiking deep networks through weight and thresh- old balancing. In 2015 International Joint Conference on Neural Networks (IJCNN), pages 1–8. ieee, 2...

    2015

  2. [10]

    Optimal ann-snn conversion for fast and accurate inference in deep spiking neural networks

    Jianhao Ding, Zhaofei Yu, Yonghong Tian, and Tiejun Huang. Optimal ann-snn conversion for fast and accurate inference in deep spiking neural networks. arXiv preprint arXiv:2105.11654, 2021. 6

    2021 arXiv

  3. [11]

    Noise in the nervous system

    A Aldo Faisal, Luc PJ Selen, and Daniel M Wolpert. Noise in the nervous system. Nature reviews neuroscience, 9(4): 292–303, 2008. 2

    2008

  4. [12]

    Deep residual learning in spiking neural networks

    Wei Fang, Zhaofei Yu, Yanqi Chen, Tiejun Huang, Timoth´ee Masquelier, and Yonghong Tian. Deep residual learning in spiking neural networks. arXiv preprint arXiv:2102.04159,

    arXiv

  5. [13]

    Stochastic resonance

    Luca Gammaitoni, Peter H ¨anggi, Peter Jung, and Fabio Marchesoni. Stochastic resonance. Reviews of modern physics, 70(1):223, 1998. 3

    1998

  6. [14]

    Rmp-snn: Residual membrane potential neuron for enabling deeper high-accuracy and low-latency spiking neural net- work

    Bing Han, Gopalakrishnan Srinivasan, and Kaushik Roy. Rmp-snn: Residual membrane potential neuron for enabling deeper high-accuracy and low-latency spiking neural net- work. In Proceedings of the IEEE/CVF conference on com- puter vision and pattern recognition , pages 13558–13567,

  7. [15]

    Reducing ann-snn conversion error through residual membrane potential

    Zecheng Hao, Tong Bu, Jianhao Ding, Tiejun Huang, and Zhaofei Yu. Reducing ann-snn conversion error through residual membrane potential. arXiv preprint arXiv:2302.02091, 2023. 2, 3, 4, 5

    2023 arXiv

  8. [16]

    Bridging the gap between anns and snns by calibrating offset spikes

    Zecheng Hao, Jianhao Ding, Tong Bu, Tiejun Huang, and Zhaofei Yu. Bridging the gap between anns and snns by calibrating offset spikes. arXiv preprint arXiv:2302.10685,

    arXiv

  9. [17]

    Fast-snn: Fast spiking neural network by converting quan- tized ann

    Yangfan Hu, Qian Zheng, Xudong Jiang, and Gang Pan. Fast-snn: Fast spiking neural network by converting quan- tized ann. arXiv preprint arXiv:2305.19868, 2023. 2, 3

    2023 arXiv

  10. [18]

    Low-latency spiking neural networks using pre-charged membrane potential and delayed evalua- tion

    Sungmin Hwang, Jeesoo Chang, Min-Hye Oh, Kyung Kyu Min, Taejin Jang, Kyungchul Park, Junsu Yu, Jong-Ho Lee, and Byung-Gook Park. Low-latency spiking neural networks using pre-charged membrane potential and delayed evalua- tion. Frontiers in Neuroscience, 15:135, 2021. 2

    2021

  11. [19]

    A unified optimization framework of ann-snn conversion: Towards optimal mapping from activa- tion values to firing rates

    Haiyan Jiang, Srinivas Anumasa, Giulia De Masi, Huan Xiong, and Bin Gu. A unified optimization framework of ann-snn conversion: Towards optimal mapping from activa- tion values to firing rates. 2023. 6, 7

    2023

  12. [20]

    Extracting oscillations: Neuronal co- incidence detection with noisy periodic spike input

    Richard Kempter, Wulfram Gerstner, J Leo Van Hemmen, and Hermann Wagner. Extracting oscillations: Neuronal co- incidence detection with noisy periodic spike input. Neural computation, 10(8):1987–2017, 1998. 2

    1987

  13. [21]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009. 6

    2009

  14. [22]

    Quantization frame- work for fast spiking neural networks

    Chen Li, Lei Ma, and Steve B Furber. Quantization frame- work for fast spiking neural networks. Frontiers in Neuro- science, page 1055, 2022. 1, 2, 3, 4, 7

    2022

  15. [23]

    Unleashing the potential of spiking neural networks by dynamic confidence

    Chen Li, Edward Jones, and Steve Furber. Unleashing the potential of spiking neural networks by dynamic confidence. arXiv preprint arXiv:2303.10276, 2023. 1

    2023 arXiv

  16. [24]

    A free lunch from ann: Towards efficient, ac- curate spiking neural networks calibration

    Yuhang Li, Shikuang Deng, Xin Dong, Ruihao Gong, and Shi Gu. A free lunch from ann: Towards efficient, ac- curate spiking neural networks calibration. arXiv preprint arXiv:2106.06984, 2021. 1

    2021 arXiv

  17. [25]

    Differentiable spike: Rethinking gradient-descent for training spiking neural networks

    Yuhang Li, Yufei Guo, Shanghang Zhang, Shikuang Deng, Yongqing Hai, and Shi Gu. Differentiable spike: Rethinking gradient-descent for training spiking neural networks. Ad- vances in Neural Information Processing Systems, 34, 2021. 1

    2021

  18. [26]

    Exploiting noise as a resource for computation and learning in spiking neural networks

    Gehua Ma, Rui Yan, and Huajin Tang. Exploiting noise as a resource for computation and learning in spiking neural networks. arXiv preprint arXiv:2305.16044, 2023. 2

    2023 arXiv

  19. [27]

    Networks of spiking neurons: the third generation of neural network models

    Wolfgang Maass. Networks of spiking neurons: the third generation of neural network models. Neural networks, 10 (9):1659–1671, 1997. 1

    1997

  20. [28]

    Spinnaker 2: A 10 million core processor system for brain simulation and machine learning

    Christian Mayr, Sebastian Hoeppner, and Steve Furber. Spinnaker 2: A 10 million core processor system for brain simulation and machine learning. arXiv preprint arXiv:1911.02385, 2019. 1

    1911 arXiv

  21. [29]

    Training much deeper spiking neural networks with a small number of time-steps

    Qingyan Meng, Shen Yan, Mingqing Xiao, Yisen Wang, Zhouchen Lin, and Zhi-Quan Luo. Training much deeper spiking neural networks with a small number of time-steps. Neural Networks, 153:254–268, 2022. 1, 2, 3, 6, 7

    2022

  22. [30]

    Surrogate gradient learning in spiking neural networks: Bringing the power of gradient-based optimization to spik- ing neural networks

    Emre O Neftci, Hesham Mostafa, and Friedemann Zenke. Surrogate gradient learning in spiking neural networks: Bringing the power of gradient-based optimization to spik- ing neural networks. IEEE Signal Processing Magazine, 36 (6):51–63, 2019. 1, 2

    2019

  23. [31]

    Efficient neuromorphic signal processing with loihi 2

    Garrick Orchard, E Paxon Frady, Daniel Ben Dayan Rubin, Sophia Sanborn, Sumit Bam Shrestha, Friedrich T Sommer, and Mike Davies. Efficient neuromorphic signal processing with loihi 2. In 2021 IEEE Workshop on Signal Processing Systems (SiPS), pages 254–259. IEEE, 2021. 1

    2021

  24. [32]

    The brainscales-2 accelerated neuromorphic sys- tem with hybrid plasticity

    Christian Pehle, Sebastian Billaudelle, Benjamin Cramer, Jakob Kaiser, Korbinian Schreiber, Yannik Stradmann, Jo- hannes Weis, Aron Leibfried, Eric M ¨uller, and Johannes Schemmel. The brainscales-2 accelerated neuromorphic sys- tem with hybrid plasticity. Frontiers in Neurosc...

  25. [33]

    Towards spike-based machine intelligence with neuromor- phic computing

    Kaushik Roy, Akhilesh Jaiswal, and Priyadarshini Panda. Towards spike-based machine intelligence with neuromor- phic computing. Nature, 575(7784):607–617, 2019. 1

    2019

  26. [34]

    Conversion of continuous-valued deep networks to efficient event-driven networks for image classification

    Bodo Rueckauer, Iulia-Alexandra Lungu, Yuhuang Hu, Michael Pfeiffer, and Shih-Chii Liu. Conversion of continuous-valued deep networks to efficient event-driven networks for image classification. Frontiers in neuroscience, 11:682, 2017. 1, 2, 3, 6

    2017

  27. [35]

    A wafer-scale neuromorphic hardware system for large-scale neural modeling

    Johannes Schemmel, Daniel Br ¨uderle, Andreas Gr ¨ubl, Matthias Hock, Karlheinz Meier, and Sebastian Millner. A wafer-scale neuromorphic hardware system for large-scale neural modeling. In 2010 IEEE International Symposium on Circuits and Systems (ISCAS) , pages 1947–1950. IEEE,

    2010

  28. [36]

    Going deeper in spiking neural networks: Vgg and residual architectures

    Abhronil Sengupta, Yuting Ye, Robert Wang, Chiao Liu, and Kaushik Roy. Going deeper in spiking neural networks: Vgg and residual architectures. Frontiers in neuroscience, 13:95,

  29. [37]

    Slayer: Spike layer error reassignment in time

    Sumit B Shrestha and Garrick Orchard. Slayer: Spike layer error reassignment in time. Advances in neural information processing systems, 31, 2018. 2

    2018

  30. [38]

    Neuronal variability: noise or part of the signal? Nature Reviews Neuroscience, 6(5):389–397, 2005

    Richard B Stein, E Roderich Gossen, and Kelvin E Jones. Neuronal variability: noise or part of the signal? Nature Reviews Neuroscience, 6(5):389–397, 2005. 2

    2005

  31. [39]

    Spatio-temporal backpropagation for training high- performance spiking neural networks

    Yujie Wu, Lei Deng, Guoqi Li, Jun Zhu, and Luping Shi. Spatio-temporal backpropagation for training high- performance spiking neural networks. Frontiers in neuro- science, 12:331, 2018. 1, 2

    2018

  32. [40]

    Spikezip-tf: Conversion is all you need for transformer-based snn

    Kang You, Zekai Xu, Chen Nie, Zhijie Deng, Qinghai Guo, Xiang Wang, and Zhezhi He. Spikezip-tf: Conversion is all you need for transformer-based snn. arXiv preprint arXiv:2406.03470, 2024. 2

    2024 arXiv

  33. [41]

    A plasticity-centric approach to train the non-differential spiking neural networks

    Tielin Zhang, Yi Zeng, Dongcheng Zhao, and Mengting Shi. A plasticity-centric approach to train the non-differential spiking neural networks. In Proceedings of the AAAI con- ference on artificial intelligence, 2018. 1 Noise Adaptor: Enhancing Low-Latency Spiking Neural Network...

    2018

  34. [42]

    Theorem 1

    Proof of Equation 10: Expected Mean of Ac- tivation Quantization Function in ANN In this section, we present a proof for Equation 10, demon- strating that the expected mean of ANN activation quanti- zation function, after incorporating a Noise Adaptor, aligns with the rectifie...

  35. [43]

    Pseudo Code for Implementing Noise Adap- tor Algorithm 1 ANN Activation Quantization Function with Noise Adaptor Input: Input tensor x, scale factor scale, quantization lower bound 0, quantization upper bound p Output: Quantized tensor ˆx 1: Forward Pass: 2: Compute x scale = ...

  36. [44]

    ⊕ (x clip ≥ p) 10: Compute gradient for activation: grad activation = grad output × internal f lag 11: Compute first part of scale gradient: grad one = x clip × internal f lag 12: Compute second part of scale gradient: grad two = round(x clip) 13: Compute element-wise scale gr...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.