pith. sign in

arxiv: 2606.31695 · v1 · pith:UEFNPICFnew · submitted 2026-06-30 · 💻 cs.CV

Intrinsically Stable Spiking Neural Networks: Overcoming the Performance Barrier in the Absence of Batch Normalization

Ruichen Ma , Xiaoyang Zhang , Jian Bai , Guanchao Qiao , Liwei Meng , Ning Ning , Yang Liu , Shaogang Hu This is my paper

Pith reviewed 2026-07-01 05:43 UTC · model grok-4.3

classification 💻 cs.CV
keywords spiking neural networksbatch normalizationweight standardizationresidual connectionsneuromorphic computingImageNethardware efficiencysignal homeostasis
0
0 comments X

The pith

Spiking neural networks match batch-norm accuracy without runtime multiplications by folding topology-aware weight standardization into static weights.

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

The paper identifies catastrophic firing-rate decay as the main driver of poor performance in spiking neural networks that lack normalization. It proposes the Intrinsically Stable SNN architecture that maintains signal homeostasis using topology-aware weight standardization together with modified residual connections. These standardization steps are computed once and folded into the fixed weights, eliminating any need for runtime statistics or multiplications during inference. Experiments across VGG, ResNet, and Transformer models show that the resulting networks reach accuracy levels competitive with dynamic batch-normalization methods, including 68.05 percent on ImageNet, while also cutting FPGA resource use sharply.

Core claim

The IS-SNN architecture removes activation-normalization layers by enforcing signal homeostasis through topology-aware weight standardization and modified residual connections. By folding the standardization operations into static weights offline, IS-SNN removes the runtime statistics tracking and multiplications introduced by activation normalization, restoring an accumulation-oriented inference datapath. Comprehensive experiments show that IS-SNN achieves performance competitive with or superior to computationally expensive dynamic BN techniques across VGG, ResNet, and Transformer-based models, notably a competitive accuracy of 68.05 percent on ImageNet, and overcomes the severe depth limi

What carries the argument

topology-aware weight standardization combined with modified residual connections that enforce signal homeostasis

If this is right

  • IS-SNN matches or exceeds dynamic BN performance on VGG, ResNet, and Transformer-based spiking models.
  • The method reaches 68.05 percent accuracy on ImageNet without any activation normalization.
  • It removes the depth limits that previously constrained BN-free spiking networks.
  • Neuron implementations consume 96.4 percent fewer FPGA lookup tables.

Where Pith is reading between the lines

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

  • The offline folding of standardization suggests the same homeostasis technique could be applied to other spike-based or event-driven models that must run on fixed-point hardware.
  • Because inference reduces to pure accumulation, the approach may directly translate to analog or mixed-signal neuromorphic chips that cannot afford digital multipliers.
  • The stability mechanism could be tested on additional datasets or neuron types to check whether the same firing-rate homeostasis holds outside the reported architectures.

Load-bearing premise

Catastrophic firing-rate decay is the primary cause of severe performance degradation in normalization-free SNNs and can be corrected by topology-aware weight standardization plus modified residual connections.

What would settle it

An ablation that removes the topology-aware standardization or the modified residuals and then measures whether firing-rate decay reappears and ImageNet accuracy falls below dynamic-BN baselines.

Figures

Figures reproduced from arXiv: 2606.31695 by Guanchao Qiao, Jian Bai, Liwei Meng, Ning Ning, Ruichen Ma, Shaogang Hu, Xiaoyang Zhang, Yang Liu.

Figure 1
Figure 1. Figure 1: Layer-wise firing rates in SNNs. (a-b) VGG-9 on CIFAR-10 before and after training. Without BN, the network suffers from rapid firing-rate decay and fails to recover. Both the BN baseline and the proposed IS-SNN prevent this collapse. (c-d) SEW-ResNet-152 on CIFAR-100. A similar signal-decay phenomenon is observed in the deep architecture, where IS-SNN maintains stable signal propagation without BN. of a C… view at source ↗
Figure 3
Figure 3. Figure 3: The (a) MRU of the residual IS￾SNN consists of one (b) transition block and several (c) residual blocks. The mod￾ified residual connection incorporates a scaling factor α to control variance growth. The calculation of σℓ depends on the network architecture. In plain non￾residual networks such as VGG, the output of layer ℓ serves directly as the input to layer ℓ + 1, as shown in [PITH_FULL_IMAGE:figures/fu… view at source ↗
Figure 4
Figure 4. Figure 4: Simulated output variance σ 2 out of a max-pooling layer processing binary spike inputs. activations. As the input firing rate µin increases from 0 to 1, the input variance σ 2 in first increases from 0 to a maximum of 0.25 before decreasing back to 0. This complex, non-monotonic relationship is detailed in [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Initial layer-wise firing rates for two modified SEW-ResNet-19 variants on CIFAR-100. As the number of transi￾tion blocks increases, the baseline exhibits stronger rate decay, while IS-SNN remains stable across these structural variants. compared methods, outperforming recent advanced methods such as TAB and TEBN. Additionally, IS-SNN can be applied to Spikformer, a Transformer-based architecture, to impro… view at source ↗
Figure 1
Figure 1. Figure 1: The surrogate gradient method in SNN training. The non-differentiable Heav￾iside step function Θ(x) is used for spike generation in the forward pass. It is ap￾proximated by a smooth surrogate function g(x), whose well-defined derivative g ′ (x) is used in the backward pass. A fundamental challenge in training SNNs is the non-differentiable na￾ture of the spike activation function. The neuronal firing proce… view at source ↗
Figure 2
Figure 2. Figure 2: Empirical output spike variance σ 2 g as a function of the pre-activation vari￾ance σ 2 pa for different neuron models. The pre-activations are drawn from a zero-mean Gaussian distribution. These plots show how different neuronal dynamics, including non-leaky and leaky integration, affect the statistics of the output spike train. The hyperparameter α controls the steepness of the surrogate function and mod… view at source ↗
Figure 3
Figure 3. Figure 3: Validation accuracy on CIFAR-10 and CIFAR-100. The naive w/o BN baseline fails to converge on VGG, while IS-SNN maintains stable training. flow through the neuron’s hard reset mechanism was detached during the back￾ward pass to improve optimization stability. The hyperparameters used for each dataset are detailed in Tab. 1. C.3 Data Augmentation (a) ImageNet. (b) DVS-Gesture [PITH_FULL_IMAGE:figures/full_… view at source ↗
Figure 4
Figure 4. Figure 4: Validation accuracy curves on Ima￾geNet and DVS-Gesture. IS-SNN prevents the training failures observed in the w/o BN baselines. Different data augmentation strate￾gies were applied according to the na￾ture of each dataset. CIFAR-10/100: Standard ran￾dom horizontal flips and random crops were used as the baseline aug￾mentation. For experiments denoted with strong augmentation (*), either Mixup or Cutmix (α… view at source ↗
read the original abstract

The performance of deep spiking neural networks (SNNs) often relies on batch normalization (BN). However, the advanced dynamic BN variants used in state-of-the-art models introduce runtime multiplications, which weaken the hardware-efficiency motivation of SNNs. To address this tension, we identify catastrophic firing-rate decay as a primary cause of severe performance degradation in normalization-free SNNs. Guided by this insight, this work proposes the Intrinsically Stable SNN (IS-SNN) architecture, which removes activation-normalization layers by enforcing signal homeostasis through topology-aware weight standardization and modified residual connections. By folding the standardization operations into static weights offline, IS-SNN removes the runtime statistics tracking and multiplications introduced by activation normalization, restoring an accumulation-oriented inference datapath. Comprehensive experiments show that IS-SNN achieves performance competitive with or superior to computationally expensive dynamic BN techniques across VGG, ResNet, and Transformer-based models. Notably, it achieves a competitive accuracy of 68.05\% on ImageNet and overcomes the severe depth limitations of prior BN-free attempts. Together with a 96.4\% reduction in FPGA lookup table resource consumption for neuron implementations, these results support IS-SNN as a practical framework for building accurate and hardware-friendly deep neuromorphic systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces the Intrinsically Stable Spiking Neural Network (IS-SNN) to overcome performance issues in deep SNNs without batch normalization. It identifies catastrophic firing-rate decay as the primary cause and proposes topology-aware weight standardization combined with modified residual connections to maintain signal homeostasis. These operations are folded into static weights offline, enabling an accumulation-oriented inference without runtime multiplications or statistics tracking. Experiments demonstrate competitive or superior performance to dynamic BN methods on VGG, ResNet, and Transformer architectures, including 68.05% accuracy on ImageNet, while achieving a 96.4% reduction in FPGA lookup table resources.

Significance. If the central claims hold, this work is significant for neuromorphic computing as it resolves the tension between accuracy in deep SNNs and hardware efficiency by eliminating dynamic normalization overhead. Strengths include the empirical validation across diverse model families and the substantial, quantified hardware benefit. The approach of offline folding to restore pure accumulation datapath is a practical contribution.

major comments (2)
  1. [Abstract] Abstract: the claim that IS-SNN achieves 'performance competitive with or superior to computationally expensive dynamic BN techniques' is presented without the specific baseline accuracies, variance, or number of runs; this is load-bearing for the central empirical result.
  2. [Insight paragraph] The paragraph beginning 'Guided by this insight': the assertion that topology-aware weight standardization plus modified residual connections enforce input-independent signal homeostasis that remains stable after offline folding into static weights lacks a dedicated stability analysis or pre/post-folding firing-rate comparison, which is central to the architecture claim.
minor comments (2)
  1. [Experiments] The experimental protocol (training details, hyperparameter choices, and ablation controls) is referenced but not summarized with sufficient granularity to allow reproduction of the ImageNet result.
  2. [Hardware evaluation] Figure or table presenting the 96.4% FPGA LUT reduction should include the exact neuron implementation baseline for clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their positive assessment of the significance of our work and for the constructive feedback. We address each major comment below and propose revisions where appropriate to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that IS-SNN achieves 'performance competitive with or superior to computationally expensive dynamic BN techniques' is presented without the specific baseline accuracies, variance, or number of runs; this is load-bearing for the central empirical result.

    Authors: We agree that the abstract would be strengthened by including specific baseline accuracies and details on experimental runs. The full manuscript provides comparisons showing IS-SNN matching or exceeding dynamic BN methods, with the ImageNet result of 68.05%. We will update the abstract to reference the specific baseline performance and indicate the number of runs used for the reported accuracies. revision: yes

  2. Referee: [Insight paragraph] The paragraph beginning 'Guided by this insight': the assertion that topology-aware weight standardization plus modified residual connections enforce input-independent signal homeostasis that remains stable after offline folding into static weights lacks a dedicated stability analysis or pre/post-folding firing-rate comparison, which is central to the architecture claim.

    Authors: The central claim is supported by the empirical results demonstrating that deep SNNs can be trained without BN using our method, with performance preserved after folding the operations into static weights. However, we acknowledge that a dedicated stability analysis would provide additional rigor. We will add pre- and post-folding firing-rate comparisons in a revised version to explicitly demonstrate the input-independent signal homeostasis. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper is an empirical architecture proposal that identifies firing-rate decay as a cause of degradation in BN-free SNNs and introduces topology-aware weight standardization plus modified residuals to enforce homeostasis, with the standardization folded offline into static weights. Claims are validated by reported accuracies (including 68.05% on ImageNet) and hardware metrics across VGG/ResNet/Transformer models rather than by any equations or self-citations that reduce performance metrics to quantities defined by the method itself. No load-bearing step reduces by construction to fitted inputs or prior self-citations; the derivation chain remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based solely on the abstract, the central claim rests on the premise that firing-rate decay can be prevented by topology-aware standardization without introducing new runtime operations; no explicit free parameters, axioms, or invented physical entities are described.

pith-pipeline@v0.9.1-grok · 5780 in / 1234 out tokens · 22878 ms · 2026-07-01T05:43:59.684130+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

50 extracted references · 1 canonical work pages

  1. [1]

    In: Proceedings of the IEEE conference on computer vision and pattern recognition

    Amir, A., Taba, B., Berg, D., Melano, T., McKinstry, J., Di Nolfo, C., Nayak, T., Andreopoulos, A., Garreau, G., Mendoza, M., et al.: A low power, fully event-based gesture recognition system. In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 7243–7252 (2017)

    2017

  2. [2]

    In: International Conference on Learning Representations (2021)

    Brock, A., De, S., Smith, S.L.: Characterizing signal propagation to close the per- formance gap in unnormalized resnets. In: International Conference on Learning Representations (2021)

    2021

  3. [3]

    In: International Conference on Machine Learning

    Brock, A., De, S., Smith, S.L., Simonyan, K.: High-performance large-scale im- age recognition without normalization. In: International Conference on Machine Learning. pp. 1059–1071. PMLR (2021)

    2021

  4. [4]

    bnn-bn=?

    Chen, T., Zhang, Z., Ouyang, X., Liu, Z., Shen, Z., Wang, Z.: " bnn-bn=?": Train- ing binary neural networks without batch normalization. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 4619–4629 (2021)

    2021

  5. [5]

    Ieee Micro38(1), 82–99 (2018)

    Davies, M., Srinivasa, N., Lin, T.H., Chinya, G., Cao, Y., Choday, S.H., Dimou, G., Joshi, P., Imam, N., Jain, S., et al.: Loihi: A neuromorphic manycore processor with on-chip learning. Ieee Micro38(1), 82–99 (2018)

    2018

  6. [6]

    In: International Conference on Learning Representations (2022)

    Deng, S., Li, Y., Zhang, S., Gu, S.: Temporal efficient training of spiking neu- ral network via gradient re-weighting. In: International Conference on Learning Representations (2022)

    2022

  7. [7]

    Advances in Neural Information Processing Systems35, 34377–34390 (2022)

    Duan, C., Ding, J., Chen, S., Yu, Z., Huang, T.: Temporal effective batch nor- malization in spiking neural networks. Advances in Neural Information Processing Systems35, 34377–34390 (2022)

    2022

  8. [8]

    Science Advances9(40), eadi1480 (2023)

    Fang, W., Chen, Y., Ding, J., Yu, Z., Masquelier, T., Chen, D., Huang, L., Zhou, H., Li, G., Tian, Y.: Spikingjelly: An open-source machine learning infrastructure platform for spike-based intelligence. Science Advances9(40), eadi1480 (2023)

    2023

  9. [9]

    Advances in Neural Information Processing Systems34, 21056–21069 (2021)

    Fang, W., Yu, Z., Chen, Y., Huang, T., Masquelier, T., Tian, Y.: Deep residual learning in spiking neural networks. Advances in Neural Information Processing Systems34, 21056–21069 (2021)

    2021

  10. [10]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Fang, W., Yu, Z., Chen, Y., Masquelier, T., Huang, T., Tian, Y.: Incorporating learnable membrane time constant to enhance learning of spiking neural networks. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 2661–2671 (2021)

    2021

  11. [11]

    Pro- ceedings of the IEEE102(5), 652–665 (2014)

    Furber, S.B., Galluppi, F., Temple, S., Plana, L.A.: The spinnaker project. Pro- ceedings of the IEEE102(5), 652–665 (2014)

    2014

  12. [12]

    In: Proceed- ings of the IEEE/CVF International Conference on Computer Vision

    Guo, Y., Zhang, Y., Chen, Y., Peng, W., Liu, X., Zhang, L., Huang, X., Ma, Z.: Membrane potential batch normalization for spiking neural networks. In: Proceed- ings of the IEEE/CVF International Conference on Computer Vision. pp. 19420– 19430 (2023)

    2023

  13. [13]

    In: Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition

    Han, B., Srinivasan, G., Roy, K.: Rmp-snn: Residual membrane potential neuron for enabling deeper high-accuracy and low-latency spiking neural network. In: Pro- ceedings of the IEEE/CVF conference on computer vision and pattern recognition. pp. 13558–13567 (2020)

    2020

  14. [14]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Hao, Z., Bu, T., Ding, J., Huang, T., Yu, Z.: Reducing ann-snn conversion error through residual membrane potential. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 37, pp. 11–21 (2023)

    2023

  15. [15]

    In: International Conference on Learning Representations (2023) 16 Ma et al

    Hao, Z., Ding, J., Bu, T., Huang, T., Yu, Z.: Bridging the gap between anns and snns by calibrating offset spikes. In: International Conference on Learning Representations (2023) 16 Ma et al

    2023

  16. [16]

    He,K.,Zhang,X.,Ren,S.,Sun,J.:Deepresiduallearningforimagerecognition.In: Proceedings of the IEEE conference on computer vision and pattern recognition. pp. 770–778 (2016)

    2016

  17. [17]

    Sensors22(8), 2876 (2022)

    Ikegawa,S.i.,Saiin,R.,Sawada,Y.,Natori,N.:Rethinkingtheroleofnormalization and residual blocks for spiking neural networks. Sensors22(8), 2876 (2022)

    2022

  18. [18]

    Nature Machine Intelligence2(3), 181–191 (2020)

    Imam, N., Cleland, T.A.: Rapid online learning and robust recall in a neuromorphic olfactory circuit. Nature Machine Intelligence2(3), 181–191 (2020)

    2020

  19. [19]

    In: International conference on machine learning

    Ioffe, S., Szegedy, C.: Batch normalization: Accelerating deep network training by reducing internal covariate shift. In: International conference on machine learning. pp. 448–456. pmlr (2015)

    2015

  20. [20]

    In: International Conference on Machine Learning

    Jiang, H., Anumasa, S., De Masi, G., Xiong, H., Gu, B.: A unified optimization framework of ann-snn conversion: towards optimal mapping from activation values to firing rates. In: International Conference on Machine Learning. pp. 14945–14974. PMLR (2023)

    2023

  21. [21]

    In: International Conference on Learning Representations (2024)

    Jiang, H., Zoonekynd, V., De Masi, G., Gu, B., Xiong, H.: Tab: Temporal accumu- lated batch normalization in spiking neural networks. In: International Conference on Learning Representations (2024)

    2024

  22. [22]

    Frontiers in neuroscience15, 773954 (2021)

    Kim, Y., Panda, P.: Revisiting batch normalization for training low-latency deep spiking neural networks from scratch. Frontiers in neuroscience15, 773954 (2021)

    2021

  23. [23]

    Master’s thesis, University of Tront (2009)

    Krizhevsky, A., Hinton, G., et al.: Learning multiple layers of features from tiny images. Master’s thesis, University of Tront (2009)

    2009

  24. [24]

    Frontiers in neuroscience11, 244131 (2017)

    Li, H., Liu, H., Ji, X., Li, G., Shi, L.: Cifar10-dvs: an event-stream dataset for object classification. Frontiers in neuroscience11, 244131 (2017)

    2017

  25. [25]

    Advances in Neural Information Processing Systems34, 23426–23439 (2021)

    Li, Y., Guo, Y., Zhang, S., Deng, S., Hai, Y., Gu, S.: Differentiable spike: Re- thinking gradient-descent for training spiking neural networks. Advances in Neural Information Processing Systems34, 23426–23439 (2021)

    2021

  26. [26]

    In: Proceedings of the AAAI Conference on Artificial Intelligence

    Ma, R., Meng, L., Qiao, G., Ning, N., Liu, Y., Hu, S.: I2e: Real-time image-to- event conversion for high-performance spiking neural networks. In: Proceedings of the AAAI Conference on Artificial Intelligence. vol. 40, pp. 1982–1990 (2026)

    1982

  27. [27]

    In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

    Ma, R., Qiao, G., Liu, Y., Meng, L., Ning, N., Liu, Y., Hu, S.: A&b bnn: Add&bit- operation-only hardware-friendly binary neural network. In: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. pp. 5704– 5713 (2024)

    2024

  28. [28]

    Science345(6197), 668–673 (2014)

    Merolla, P.A., Arthur, J.V., Alvarez-Icaza, R., Cassidy, A.S., Sawada, J., Akopyan, F., Jackson, B.L., Imam, N., Guo, C., Nakamura, Y., et al.: A million spiking- neuron integrated circuit with a scalable communication network and interface. Science345(6197), 668–673 (2014)

    2014

  29. [29]

    IEEE Signal Processing Magazine36(6), 51–63 (2019)

    Neftci, E.O., Mostafa, H., Zenke, F.: Surrogate gradient learning in spiking neural networks: Bringing the power of gradient-based optimization to spiking neural networks. IEEE Signal Processing Magazine36(6), 51–63 (2019)

    2019

  30. [30]

    Nature572(7767), 106–111 (2019)

    Pei, J., Deng, L., Song, S., Zhao, M., Zhang, Y., Wu, S., Wang, G., Zou, Z., Wu, Z., He, W., et al.: Towards artificial general intelligence with hybrid tianjic chip architecture. Nature572(7767), 106–111 (2019)

    2019

  31. [31]

    Neu- rocomputing544, 126234 (2023)

    Qiao, G., Ning, N., Zuo, Y., Zhou, P., Sun, M., Hu, S., Yu, Q., Liu, Y.: Batch normalization-free weight-binarized snn based on hardware-saving if neuron. Neu- rocomputing544, 126234 (2023)

    2023

  32. [32]

    Frontiers in neuroscience9, 141 (2015) IS-SNN 17

    Qiao, N., Mostafa, H., Corradi, F., Osswald, M., Stefanini, F., Sumislawska, D., Indiveri, G.: A reconfigurable on-line learning spiking neuromorphic processor com- prising 256 neurons and 128k synapses. Frontiers in neuroscience9, 141 (2015) IS-SNN 17

    2015

  33. [33]

    arXiv preprint arXiv:1903.10520 , year=

    Qiao, S., Wang, H., Liu, C., Shen, W., Yuille, A.: Micro-batch training with batch-channel normalization and weight standardization. arXiv preprint arXiv:1903.10520 (2019)

    work page arXiv 1903

  34. [34]

    Nature575(7784), 607–617 (2019)

    Roy, K., Jaiswal, A., Panda, P.: Towards spike-based machine intelligence with neuromorphic computing. Nature575(7784), 607–617 (2019)

    2019

  35. [35]

    International journal of computer vision115, 211–252 (2015)

    Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al.: Imagenet large scale visual recog- nition challenge. International journal of computer vision115, 211–252 (2015)

    2015

  36. [36]

    Frontiers in neuroscience13, 95 (2019)

    Sengupta, A., Ye, Y., Wang, R., Liu, C., Roy, K.: Going deeper in spiking neural networks: Vgg and residual architectures. Frontiers in neuroscience13, 95 (2019)

    2019

  37. [37]

    Ad- vances in neural information processing systems31(2018)

    Shrestha, S.B., Orchard, G.: Slayer: Spike layer error reassignment in time. Ad- vances in neural information processing systems31(2018)

    2018

  38. [38]

    Turrigiano, G.G.: Homeostatic plasticity in neuronal networks: the more things change,themoretheystaythesame.Trendsinneurosciences22(5),221–227(1999)

    1999

  39. [39]

    Nature reviews neuroscience5(2), 97–107 (2004)

    Turrigiano, G.G., Nelson, S.B.: Homeostatic plasticity in the developing nervous system. Nature reviews neuroscience5(2), 97–107 (2004)

    2004

  40. [40]

    In: IJCAI

    Wang, B., Cao, J., Chen, J., Feng, S., Wang, Y.: A new ann-snn conversion method with high accuracy, low latency and good robustness. In: IJCAI. pp. 3067–3075 (2023)

    2023

  41. [41]

    Frontiers in neuroscience12, 331 (2018)

    Wu, Y., Deng, L., Li, G., Zhu, J., Shi, L.: Spatio-temporal backpropagation for training high-performance spiking neural networks. Frontiers in neuroscience12, 331 (2018)

    2018

  42. [42]

    In: Proceedings of the AAAI conference on artificial intelligence

    Wu, Y., Deng, L., Li, G., Zhu, J., Xie, Y., Shi, L.: Direct training for spiking neural networks: Faster, larger, better. In: Proceedings of the AAAI conference on artificial intelligence. vol. 33, pp. 1311–1318 (2019)

    2019

  43. [43]

    Advances in neural information processing systems35, 20717–20730 (2022)

    Xiao, M., Meng, Q., Zhang, Z., He, D., Lin, Z.: Online training through time for spiking neural networks. Advances in neural information processing systems35, 20717–20730 (2022)

    2022

  44. [44]

    In: IJCAI

    Xu, Q., Qi, Y., Yu, H., Shen, J., Tang, H., Pan, G., et al.: Csnn: an augmented spiking based framework with perceptron-inception. In: IJCAI. vol. 1646 (2018)

    2018

  45. [45]

    In: Proceedings of the IEEE/CVF international conference on computer vision

    Yun, S., Han, D., Oh, S.J., Chun, S., Choe, J., Yoo, Y.: Cutmix: Regularization strategy to train strong classifiers with localizable features. In: Proceedings of the IEEE/CVF international conference on computer vision. pp. 6023–6032 (2019)

    2019

  46. [46]

    Neuron109(4), 571–575 (2021)

    Zenke,F.,Bohté,S.M.,Clopath,C.,Comşa,I.M.,Göltz,J.,Maass,W.,Masquelier, T., Naud, R., Neftci, E.O., Petrovici, M.A., et al.: Visualizing a joint future of neuroscience and neuromorphic engineering. Neuron109(4), 571–575 (2021)

    2021

  47. [47]

    In: International Confer- ence on Learning Representations (2018)

    Zhang, H.: mixup: Beyond empirical risk minimization. In: International Confer- ence on Learning Representations (2018)

    2018

  48. [48]

    In: Proceedings of the AAAI conference on artificial intelligence

    Zheng, H., Wu, Y., Deng, L., Hu, Y., Li, G.: Going deeper with directly-trained larger spiking neural networks. In: Proceedings of the AAAI conference on artificial intelligence. vol. 35, pp. 11062–11070 (2021)

    2021

  49. [49]

    In: Proceedings of the AAAI conference on artificial intelligence

    Zhou, S., Li, X., Chen, Y., Chandrasekaran, S.T., Sanyal, A.: Temporal-coded deep spiking neural network with easy training and robust performance. In: Proceedings of the AAAI conference on artificial intelligence. vol. 35, pp. 11143–11151 (2021)

    2021

  50. [50]

    Zhou, Z., Zhu, Y., He, C., Wang, Y., YAN, S., Tian, Y., Yuan, L.: Spikformer: When spiking neural network meets transformer. In: The Eleventh International Conference on Learning Representations (2023) Supplementary Material for Intrinsically Stable Spiking Neural Networks: Overcoming the Performance Barrier in the Absence of Batch Normalization Ruichen M...

    2023