Pith. sign in

REVIEW 4 major objections 6 minor 39 references

Effective and Efficient Intracortical Brain Signal Decoding with Spiking Neural Networks

T0 review · 4 major / 6 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read A spiking neural network with local synaptic stabilization and channel-wise attention surpasses ANN baselines on intracortical decoding accuracy while being estimated to use 15-44x less energy, with SpikeDrop augmentation adding further…

desk verdict Accuracy results on two monkey datasets are credible and the cross-session protocol is careful, but the 15-44x energy saving is an idealized FLOP-count ratio that likely won't survive system-level accounting. read the letter →

arxiv 2412.20714 v1 pith:B5VI5FD7 submitted 2024-12-30 cs.HC

Haotian Fu , Peng Zhang , Song Yang , Herui Zhang , Ziwei Wang , Dongrui Wu This is my paper · ORCID
classification cs.HC
keywords brain-computerinterfacespikingneuralnetworkintracorticaldecodinglocalsynapticstabilizationchannel-wiseattentionSpikeDropdataaugmentationenergyefficiency
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

Invasive brain-computer interfaces record spikes directly from neurons, and the neural activity arrives as binary spike trains that are naturally suited to spiking neural networks. This paper proposes a spiking network, LSS-CA-SNN, that adds a local synaptic stabilization mechanism to improve membrane-potential dynamics and a channel-wise attention module to suppress irrelevant channels; it also introduces SpikeDrop, a data augmentation that randomly masks time points, time segments, and channels. On data from two rhesus macaques, one performing a motor task and one a sensory task, the network reports the highest classification accuracies in both cross-session transfer scenarios, beating the best ANN baseline by 0.80-3.87 percentage points. The paper also reports an estimated energy reduction of 14.78-43.86 times relative to ANN baselines, based on counting multiply-accumulate versus accumulate operations. If these results hold, energy-efficient spiking decoders could make implantable and portable BCIs more practical without sacrificing accuracy.

What carries the argument

The central machinery is a parametric leaky integrate-and-fire (PLIF) neuron, a spiking neuron with a learnable membrane time constant, trained end-to-end by spatio-temporal backpropagation with surrogate gradients. Two additions carry the claimed gains: local synaptic stabilization adjusts the membrane-potential dynamics in the deep layers to improve learning, and a channel-wise attention module reweights the input channels before spiking layers, which suppresses irrelevant spikes and lowers the number of accumulate operations. SpikeDrop augments the binary spike tensor by randomly masking individual time points, contiguous time segments, or whole channels during training. The energy comparison uses a cost model in which a multiply-accumulate operation costs 4.6 pJ and a single accumulate operation costs 0.9 pJ under 45nm CMOS assumptions, so the attention module's reduction in spiking activity translates directly into lower estimated energy.

What would settle it

Measure end-to-end inference energy on a neuromorphic processor, or redo the same FLOP count with realistic memory-access and data-movement costs per layer; if the 14.78-43.86x gap drops below a few times, the paper's headline energy advantage is an artifact of the cost model rather than a property of the architecture.

Watch

Extended reading notes

Core claim

The paper's central claim is that a purpose-built spiking neural network, rather than a generic or converted artificial neural network, is the right model for decoding intracortical spiking data. LSS-CA-SNN consistently achieved the highest average accuracy across both paradigms and both transfer scenarios: it outperformed EEGNet by 3.87% and 2.96% in the motor paradigm and by 0.84% and 0.80% in the sensory paradigm, with statistical significance (p<0.01) against the ANN baselines. When the authors count FLOPs as accumulate operations for the spiking layers and multiply-accumulate operations for the small real-valued parts, LSS-CA-SNN's estimated inference energy is 1.35-1.36 uJ in the motor paradigm and 0.52-0.53 uJ in the sensory paradigm, giving 14.78-43.86x savings over EEGNet and larger savings over other ANNs. The ablation study attributes the gains to the combination of local synaptic stabilization and channel-wise attention, and SpikeDrop further raises accuracy (for example, from 86.35% to 87.35% in unsupervised motor transfer). The claim is that accuracy and energy efficiency can be improved together, not traded off.

Load-bearing premise

The energy claim stands on an assumed per-operation cost model (4.6 pJ per multiply-accumulate, 0.9 pJ per accumulate) that leaves out memory reads and data movement; if those costs are included, the reported 14-44x savings could shrink sharply.

Editorial extensions

If this is right

  • LSS-CA-SNN outperforms EEGNet, ShallowConvNet, DeepConvNet, and EEGConformer on both motor and sensory intracortical decoding under both unsupervised and supervised cross-session transfer.
  • The attention module reduces estimated energy by suppressing spiking activity: adding channel-wise attention lowers spiking counts and improves the reported energy ratio in the ablation study.
  • SpikeDrop provides a single augmentation method for binary spiking data that improves not only LSS-CA-SNN but also ANN baselines such as EEGNet and EEGConformer in several settings.
  • Because the first convolutional layer operates directly on binary spikes, the decoder is dominated by accumulate operations, which is the source of the large estimated energy gap relative to MAC-dominated ANNs.
  • With only 5-20% labeled target-session data, LSS-CA-SNN still outperforms EEGNet, suggesting the architecture adapts quickly to new sessions.

Reading between the lines

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

  • Extension: the energy ratios are computed from an idealized per-operation model; on real neuromorphic hardware the cost of memory access, control logic, and data movement would reduce the gap, so the 15-44x figure should be read as an upper bound until measured on a chip.
  • Extension: SpikeDrop's random spike masking is conceptually close to dropout for spike trains and likely regularizes against overfitting; one testable prediction is that its benefit grows as the labeled training set shrinks, which the paper's 5-20% target-data experiment partly supports.
  • Extension: the dataset comes from two macaques with different tasks, so the architecture's edge may be task- or subject-specific; applying the same pipeline to more subjects and to human intracortical recordings would show whether the accuracy gains generalize.
  • Extension: because the input is already thresholded binary spikes, SpikeDrop and LSS-CA-SNN could transfer to other event-based neural decoders or neuromorphic sensors where binary events are the native format, though the paper does not test that.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The manuscript proposes LSS-CA-SNN, a spiking neural network that combines local synaptic stabilization (LSS), channel-wise attention (CA), and a SpikeDrop data-augmentation method for decoding intracortical spike signals. It reports experiments on two rhesus macaque datasets (motor and sensory paradigms) under cross-session unsupervised and supervised transfer, comparing accuracy with EEGNet, DeepConvNet, ShallowConvNet, and EEGConformer, and comparing SpikeDrop with nine augmentation baselines. The paper's central claims are that LSS-CA-SNN achieves the highest average classification accuracy in all four settings, with 0.80-3.87% gains over EEGNet, and that it consumes 14.78-43.86 times less energy, estimated from MAC/AC operation counts. An ablation table reports that LSS and CA each improve accuracy and reduce spiking counts.

Significance. The cross-session held-out evaluation protocol with five random seeds is a strength, and the accuracy comparisons are largely credible as evidence that a directly trained SNN can match or exceed ANN decoders on these intracortical datasets. SpikeDrop appears to provide small but consistent gains across architectures and is a reasonable contribution to spiking data augmentation. The paper ships code, which aids reproducibility. However, the two load-bearing novelties are not yet supported as presented: LSS has no mathematical definition, and the energy-efficiency claim is a back-of-the-envelope operation-count estimate rather than a measured or simulated system-level result. The accuracy advantages are modest, so the practical significance of the paper currently depends on the credibility of the energy model.

major comments (4)
  1. [III.B.2] The local synaptic stabilization (LSS) mechanism is central to the paper's title and to the accuracy gains in Table VII, but it is never formally defined. The text only states that LSS is incorporated in the fourth layer and that it 'optimizes the membrane potential of deep neurons.' No equation, update rule, or description of what is being stabilized is provided. Please give a precise mathematical formulation and training-time implementation; without it the proposed architecture is not reproducible and the ablation 'LSS' cannot be interpreted.
  2. [IV.E, Eqs. (4)-(6), Table VI] The 14.78-43.86x energy saving is computed, not measured: Eqs. (4)-(6) count MAC/AC operations with assumed 45nm per-op costs (E_MAC=4.6 pJ, E_AC=0.9 pJ) and ignore memory access, control logic, and data movement, which often dominate real accelerator energy. The comparison is also asymmetric: ANN baselines are charged dense MACs while the SNN is credited with event-driven AC skips, even though the ANN inputs are binary spikes. Since the accuracy gain over EEGNet is only 0.80-3.87%, the energy ratio is a main pillar of the 'efficient' claim. Please validate on hardware or a cycle-accurate simulator, or re-scope the claim to 'estimated arithmetic energy' with a sensitivity analysis.
  3. [IV.C/IV.D] The comparison set contains only ANN decoders; no SNN baseline is included. References [24]-[26] describe SNN decoders for invasive BCIs, and the paper does not compare with them or with a vanilla PLIF-SNN without LSS/CA. Consequently the claims that LSS-CA-SNN is state-of-the-art among SNNs and that LSS/CA are responsible for the improvement over a standard SNN are unsupported.
  4. [Tables II/III and Fig. 6] The statement that LSS-CA-SNN 'consistently achieved the highest classification accuracies across both paradigms and both scenarios' is stronger than the data show. In Table II, motor unsupervised S8, LSS-CA-SNN scores 75.30% versus 77.97% for EEGNet and 78.22% for DeepConvNet; in Table III, sensory unsupervised S2 and S7, EEGNet beats it by 1.45 and 0.31 points. The asterisk denoting 'statistically significant difference (p<0.01)' is applied to every baseline row but no test statistic, pairing scheme, or multiple-comparison correction is described. Please restrict the claim to mean accuracy over sessions and report the statistical procedure.
minor comments (6)
  1. [III.B, Eq. (4)] The sentence after Eq. (4) refers to 'E_MAC and E_AC respectively,' but the formula contains only E_AC; the first convolution layer's MAC contribution should be written explicitly.
  2. [III.C] SpikeDrop's parameters (pt=0.1, ps=0.05, pc=0.1, M=N/2) are given without sensitivity analysis; please add an ablation or at least a sentence on how they were chosen.
  3. [IV.A] The compression from 12000 to 100 time bins by max pooling is described in one sentence; please justify this choice or compare with sum/any binning, since it determines the temporal information available to all methods.
  4. [Fig. 2 and III.B] The architecture description lacks filter counts, strides, pooling sizes, threshold and reset values, and the exact placement of the residual connection; please add a layer-by-layer specification.
  5. [Tables IV/V] The 'Event mix' column label and the bold/underline conventions should be defined in the captions, and the abbreviation 'Eventmix' should be used consistently.
  6. [Introduction and II.C] The paper calls SpikeDrop 'the first data augmentation approach specifically tailored for spiking data' while also citing NDA [32] and Eventmix [33], which operate on event streams; please clarify the distinction from those prior event-based augmentations.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: accuracy is measured on held-out sessions and the energy figure is an explicitly stated modeling estimate, not a fitted or definitional prediction.

full rationale

LSS-CA-SNN's central accuracy claim is established by held-out cross-session evaluation in Tables II and III: each session is used as the target session once, results are averaged over five random seeds, and all compared networks are trained under the same protocol. This is external, data-driven evidence and does not reduce to the method's own definitions. The energy-efficiency claim is not a fitted prediction: Eqs. (4)-(6) define an energy model using externally reported constants (E_MAC = 4.6 pJ, E_AC = 0.9 pJ, from reference [39]), and the reported 14.78-43.86x ratios are algebraic consequences of MAC/AC counts under those stated assumptions. This is an idealized estimate whose omission of memory-access, control, and data-movement overhead is a correctness or completeness limitation, not circularity: no parameter in the model is fit to the energy number being reported, and the SNN's fewer AC operations are counted, not assumed. SpikeDrop is evaluated by augmentation comparisons on held-out data. The self-citations in the manuscript (e.g., Wu et al. tutorial references and the [35] dataset) are background references or data provenance; none is invoked as a uniqueness theorem or as the sole justification for a core architectural choice. No ansatz is smuggled through self-citation, and no known result is merely renamed. Therefore no load-bearing step in the derivation chain is, by the paper's own equations, equivalent to its input.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The accuracy comparison rests on held-out sessions and is externally grounded, so the circularity burden is low for the main classification claim. However, the energy-efficiency claim rests on assumed per-operation costs and on the equation that energy equals spike count times those costs; CA is then credited with reducing energy precisely because it lowers the spike count, which is partly a definitional outcome. The main unstated assumptions are the surrogate-gradient approximation, the energy model, and the max-pooling preprocessing.

free parameters (5)
  • SpikeDrop masking probabilities = pt = 0.1, ps = 0.05, pc = 0.1
    Set by hand in Section III-C; no sensitivity analysis or search is reported, yet these probabilities control the amount of augmentation in every experiment.
  • SpikeDrop sample count M = N/2 in unsupervised transfer; target-session data in supervised transfer
    M is set to half the dataset in the unsupervised setting with no ablation showing the effect of this choice.
  • PLIF time constant initialization = tau = 2
    Given in Eq. (1); tau is learnable afterward, but this initialization can affect dynamics and results.
  • Surrogate gradient width = a = 4
    Set in Section III-A; no sensitivity analysis is provided.
  • Spike threshold and reset potential = uth, Vreset (values not reported)
    Threshold and reset govern firing behavior and are not numerically specified in the text, making replication harder.
assumptions (4)
  • domain assumption The surrogate-gradient approximation in Eq. (3) permits backpropagation through the Heaviside spike function.
    All training relies on this approximation; it is standard in SNN literature but not a formal guarantee of convergence or optimality.
  • domain assumption Energy comparison assumes 45nm CMOS with E_MAC = 4.6 pJ and E_AC = 0.9 pJ and ignores memory and control overhead.
    Assumed from Ref. [39] in Section IV-E; the 14.78-43.86x saving ratios are direct functions of these constants.
  • ad hoc to paper Reducing trial length from 12000 to 100 bins via max pooling preserves discriminative information in binary spike format.
    Stated in Section IV-A without validation; if timing structure is lost, all downstream accuracy results are affected.
  • domain assumption The PLIF neuron discretization in Eq. (2) with reset and threshold is an adequate model of the recorded spike trains.
    The paper uses this model as the computational substrate for all networks, but does not validate it against the raw recorded waveform statistics.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Effective and Efficient Intracortical Brain Signal Decoding with Spiking Neural Networks." pith.science (2026). https://pith.science/paper/B5VI5FD7

@misc{pith2026241220714,
  author       = {Pith},
  title        = {Pith review of: Effective and Efficient Intracortical Brain Signal Decoding with Spiking Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/B5VI5FD7}},
  note         = {Machine review of arXiv:2412.20714}
}
read the original abstract

A brain-computer interface (BCI) facilitates direct interaction between the brain and external devices. To concurrently achieve high decoding accuracy and low energy consumption in invasive BCIs, we propose a novel spiking neural network (SNN) framework incorporating local synaptic stabilization (LSS) and channel-wise attention (CA), termed LSS-CA-SNN. LSS optimizes neuronal membrane potential dynamics, boosting classification performance, while CA refines neuronal activation, effectively reducing energy consumption. Furthermore, we introduce SpikeDrop, a data augmentation strategy designed to expand the training dataset thus enhancing model generalizability. Experiments on invasive spiking datasets recorded from two rhesus macaques demonstrated that LSS-CA-SNN surpassed state-of-the-art artificial neural networks (ANNs) in both decoding accuracy and energy efficiency, achieving 0.80-3.87% performance gains and 14.78-43.86 times energy saving. This study highlights the potential of LSS-CA-SNN and SpikeDrop in advancing invasive BCI applications.

Figures

Figures reproduced from arXiv: 2412.20714 by the authors.

Figure 1
Figure 1. A closed-loop BCI system. H. Fu, S. Yang, H. Zhang, Z. Wang and D. Wu are with the Key Laboratory of the Ministry of Education for Image Processing and Intelligent Control, School of Artificial Intelligence and Automation, Huazhong University of Science and Technology, Wuhan 430074, China. P. Zhang is with Department of Biomedical Engineering, College of Life Science and Technology, Huazhong University of Science an… view at source ↗
Figure 2
Figure 2. Our proposed LSS-CA-SNN, which includes four differ [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. A convolution-based PLIF-SNN layer, including conv [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: SpikeDrop augmentation process, showcasing the app [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: The data collection device in (a) motor paradigm and ( [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 7
Figure 7. Figure 7: Performance of our proposed LSS-CA-SNN and EEGNet us [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

39 extracted references · 35 canonical work pages

  1. [24]

    A brain-machine interface operat ing with a real-time spiking neural network control algorithm,

    J. Dethier, P . Nuyujukian, C. Eliasmith, T. Stewart, S. Elasaad, K. V . Shenoy, and K. A. Boahen, “A brain-machine interface operat ing with a real-time spiking neural network control algorithm,” in Proc. Advances in Neural Information Processing Systems , Granada, Spain, Dec. 2011, pp. 2213–2221

    work page 2011

  2. [26]

    To wards neuromorphic brain-computer interfaces: Model and circui t Co-design of the spiking EEGNet,

    L. Feng, H. Shan, Z. Fan, Y . Zhang, L. Y ang, and Z. Zhu, “To wards neuromorphic brain-computer interfaces: Model and circui t Co-design of the spiking EEGNet,” Microelectronics Journal, vol. 137, p. 105808, 2023

    work page 2023

  3. [1]

    Brain–computer interface technologies in the coming deca des,

    B. J. Lance, S. E. Kerick, A. J. Ries, K. S. Oie, and K. McDow ell, “Brain–computer interface technologies in the coming deca des,” Proc. of the IEEE , vol. 100, no. Special Centennial Issue, pp. 1585–1599, 2012

    work page 2012

  4. [2]

    Re- habilitation with brain-computer interface systems,

    G. Pfurtscheller, G. R. M¨ uller-Putz, R. Scherer, and C. Neuper, “Re- habilitation with brain-computer interface systems,” Computer, vol. 41, no. 10, pp. 58–65, 2008. 10

    work page 2008

  5. [3]

    Affective brain-comp uter interfaces (aBCIs): A tutorial,

    D. Wu, B.-L. Lu, B. Hu, and Z. Zeng, “Affective brain-comp uter interfaces (aBCIs): A tutorial,” Proc. of the IEEE , vol. 11, no. 10, pp. 1314–1332, 2023

    2023

  6. [4]

    High-speed spelling with a noninvasive brain-computer in terface,

    X. Chen, Y . Wang, M. Nakanishi, X. Gao, T.-P . Jung, and S. G ao, “High-speed spelling with a noninvasive brain-computer in terface,” Proc. National Academy of Sciences, vol. 112, no. 44, pp. E6058–E6067, 2015

    work page 2015

  7. [5]

    Game s, gameplay, and BCI: the state of the art,

    D. Marshall, D. Coyle, S. Wilson, and M. Callaghan, “Game s, gameplay, and BCI: the state of the art,” IEEE Trans. on Computational Intelligence and AI in Games , vol. 5, no. 2, pp. 82–99, 2013

    work page 2013

  8. [6]

    EEG-based driver drowsiness est imation using feature weighted episodic training,

    Y . Cui, Y . Xu, and D. Wu, “EEG-based driver drowsiness est imation using feature weighted episodic training,” IEEE Trans. on Neural Systems and Rehabilitation Engineering , vol. 27, no. 11, pp. 2263–2273, 2019

    work page 2019

Show all 39 references
  1. [7]

    Transfer learning for EEG-bas ed brain– computer interfaces: A review of progress made since 2016,

    D. Wu, Y . Xu, and B.-L. Lu, “Transfer learning for EEG-bas ed brain– computer interfaces: A review of progress made since 2016,” IEEE Trans. on Cognitive and Developmental Systems , vol. 14, no. 1, pp. 4–19, 2020

    2016

  2. [8]

    Feature- selection- based transfer learning for intracortical brain–machine i nterface decod- ing,

    P . Zhang, W. Li, X. Ma, J. He, J. Huang, and Q. Li, “Feature- selection- based transfer learning for intracortical brain–machine i nterface decod- ing,” IEEE Trans. on Neural Systems and Rehabilitation Engineeri ng, vol. 29, pp. 60–73, 2020

    2020

  3. [9]

    Restoring cortical control of functional movement in a human with quadriplegia,

    C. E. Bouton, A. Shaikhouni, N. V . Annetta, M. A. Bockbrad er, D. A. Friedenberg, D. M. Nielson, G. Sharma, P . B. Sederberg, B. C. Glenn, W. J. Mysiw et al. , “Restoring cortical control of functional movement in a human with quadriplegia,” Nature, vol. 533, no. 7602, pp. 24...

    2016

  4. [10]

    Long-term stability of cortical population dynam ics underlying consistent behavior,

    J. A. Gallego, M. G. Perich, R. H. Chowdhury, S. A. Solla, and L. E. Miller, “Long-term stability of cortical population dynam ics underlying consistent behavior,” Nature Neuroscience, vol. 23, no. 2, pp. 260–270, 2020

    2020

  5. [11]

    Transfer learning for moto r imagery based brain–computer interfaces: A tutorial,

    D. Wu, X. Jiang, and R. Peng, “Transfer learning for moto r imagery based brain–computer interfaces: A tutorial,” Neural Networks, vol. 153, pp. 235–253, 2022

    2022

  6. [12]

    Lightweight source-free transfer f or privacy- preserving motor imagery classification,

    W. Zhang and D. Wu, “Lightweight source-free transfer f or privacy- preserving motor imagery classification,” IEEE Trans. on Cognitive and Developmental Systems , vol. 15, no. 2, pp. 938–949, 2022

    2022

  7. [13]

    Adversarial attacks and defenses in physiological c omputing: a systematic review,

    D. Wu, J. Xu, W. Fang, Y . Zhang, L. Y ang, X. Xu, H. Luo, and X. Y u, “Adversarial attacks and defenses in physiological c omputing: a systematic review,” National Science Open , vol. 2, no. 1, p. 20220023, 2023

    2023

  8. [14]

    Deep learning with convolutional neural networks for EEG d ecoding and visualization,

    R. T. Schirrmeister, J. T. Springenberg, L. D. J. Fieder er, M. Glasstetter, K. Eggensperger, M. Tangermann, F. Hutter, W. Burgard, and T . Ball, “Deep learning with convolutional neural networks for EEG d ecoding and visualization,” Human Brain Mapping , vol. 38, no. 11, pp. ...

    2017

  9. [15]

    EEGNet: a compact convolutional neural net work for EEG-based brain–computer interfaces,

    V . J. Lawhern, A. J. Solon, N. R. Waytowich, S. M. Gordon, C. P . Hung, and B. J. Lance, “EEGNet: a compact convolutional neural net work for EEG-based brain–computer interfaces,” Journal of Neural Engineering , vol. 15, no. 5, p. 056013, 2018

    2018

  10. [16]

    EEG conformer: Con volutional transformer for EEG decoding and visualization,

    Y . Song, Q. Zheng, B. Liu, and X. Gao, “EEG conformer: Con volutional transformer for EEG decoding and visualization,” IEEE Trans. on Neural Systems and Rehabilitation Engineering , vol. 31, pp. 710–719, 2022

    2022

  11. [17]

    A million spiking-neuron integrated circuit with a scalable communication network and interface,

    P . A. Merolla, J. V . Arthur, R. Alvarez-Icaza, A. S. Cass idy, J. Sawada, F. Akopyan, B. L. Jackson, N. Imam, C. Guo, Y . Nakamura et al. , “A million spiking-neuron integrated circuit with a scalable communication network and interface,” Science, vol. 345, no. 6197, pp. 668...

    2014

  12. [18]

    Loihi: A neuromorphic manycore processor with on-chip learning,

    M. Davies, N. Srinivasa, T.-H. Lin, G. Chinya, Y . Cao, S. H. Choday, G. Dimou, P . Joshi, N. Imam, S. Jain et al. , “Loihi: A neuromorphic manycore processor with on-chip learning,” IEEE Micro, vol. 38, no. 1, pp. 82–99, 2018

    2018

  13. [19]

    Towards artificial general intelligence with hybrid Tianjic chip architecture,

    J. Pei, L. Deng, S. Song, M. Zhao, Y . Zhang, S. Wu, G. Wang, Z. Zou, Z. Wu, W. He et al., “Towards artificial general intelligence with hybrid Tianjic chip architecture,” Nature, vol. 572, no. 7767, pp. 106–111, 2019

    2019

  14. [20]

    Fast-classifying, high-accuracy spiking deep networks t hrough weight and threshold balancing,

    P . U. Diehl, D. Neil, J. Binas, M. Cook, S.-C. Liu, and M. P feiffer, “Fast-classifying, high-accuracy spiking deep networks t hrough weight and threshold balancing,” in Proc. Int’l Joint Conf. on Neural Networks , Killarney, Ireland, Jul. 2015, pp. 1–8

    2015

  15. [21]

    Spatio-tempora l backpropa- gation for training high-performance spiking neural netwo rks,

    Y . Wu, L. Deng, G. Li, J. Zhu, and L. Shi, “Spatio-tempora l backpropa- gation for training high-performance spiking neural netwo rks,” Frontiers in Neuroscience, vol. 12, p. 331, 2018

    2018

  16. [22]

    Incorporating learnable membrane time constant to enhanc e learning of spiking neural networks,

    W. Fang, Z. Y u, Y . Chen, T. Masquelier, T. Huang, and Y . Ti an, “Incorporating learnable membrane time constant to enhanc e learning of spiking neural networks,” in Proc. IEEE/CVF Int’l Conf. on Computer Vision, Montreal, Canada, Oct. 2021, pp. 2661–2671

    2021

  17. [23]

    DIET-SNN: A low-latency spiking ne ural network with direct input encoding and leakage and threshold optimi zation,

    N. Rathi and K. Roy, “DIET-SNN: A low-latency spiking ne ural network with direct input encoding and leakage and threshold optimi zation,” IEEE Trans. on Neural Networks and Learning Systems , vol. 34, no. 6, pp. 3174–3182, 2021

    2021

  18. [25]

    An energy-efficient sp iking neural network for finger velocity decoding for implantable brain- machine interface,

    J. Liao, L. Widmer, X. Wang, A. Di Mauro, S. R. Nason-Toma szewski, C. A. Chestek, L. Benini, and T. Jang, “An energy-efficient sp iking neural network for finger velocity decoding for implantable brain- machine interface,” in Proc. Int’l Conf. on Artificial Intelligence Circuits...

    2022

  19. [27]

    Data augmentation for EEG-based emotion recognition with deep convolutional neural networks,

    F. Wang, S.-h. Zhong, J. Peng, J. Jiang, and Y . Liu, “Data augmentation for EEG-based emotion recognition with deep convolutional neural networks,” in Proc. Int. Conf. on Multimedia Modeling , Bangkok, Thailand, Feb. 2018, pp. 82–93

    2018

  20. [28]

    Contrastive repre- sentation learning for electroencephalogram classificati on,

    M. N. Mohsenvand, M. R. Izadi, and P . Maes, “Contrastive repre- sentation learning for electroencephalogram classificati on,” in Machine Learning for Health , Virtual, Dec. 2020, pp. 238–253

    2020

  21. [29]

    Addressing class imbalance in classification pr oblems of noisy signals by using fourier transform surrogates,

    J. T. Schwabedal, J. C. Snyder, A. Cakmak, S. Nemati, and G. D. Clifford, “Addressing class imbalance in classification pr oblems of noisy signals by using fourier transform surrogates,” arXiv preprint arXiv:1806.08675, 2018

    2018 arXiv

  22. [30]

    Channel reflecti on: Knowledge-driven data augmentation for EEG-based brain–c omputer interfaces,

    Z. Wang, S. Li, J. Luo, J. Liu, and D. Wu, “Channel reflecti on: Knowledge-driven data augmentation for EEG-based brain–c omputer interfaces,” Neural Networks , vol. 176, p. 106351, 2024

    2024

  23. [31]

    L earning from heterogeneous EEG signals with differentiable channel reo rdering,

    A. Saeed, D. Grangier, O. Pietquin, and N. Zeghidour, “L earning from heterogeneous EEG signals with differentiable channel reo rdering,” in Proc. Int’l Conf. on Acoustics, Speech and Signal Processin g, Virtual, Jun. 2021, pp. 1255–1259

    2021

  24. [32]

    Neuromor phic data augmentation for training spiking neural networks,

    Y . Li, Y . Kim, H. Park, T. Geller, and P . Panda, “Neuromor phic data augmentation for training spiking neural networks,” in Proc. European Conf. on Computer Vision , Tel Aviv, Israel, Oct. 2022, pp. 631–649

    2022

  25. [33]

    Eventmix: An efficient dat a augmenta- tion strategy for event-based learning,

    G. Shen, D. Zhao, and Y . Zeng, “Eventmix: An efficient dat a augmenta- tion strategy for event-based learning,” Information Sciences , vol. 644, p. 119170, 2023

    2023

  26. [34]

    A comp lex- valued convolutional fusion-type multi-stream spatiotem poral network for automatic modulation classification,

    Y . Wang, S. Fang, Y . Fan, M. Wang, Z. Xu, and S. Hou, “A comp lex- valued convolutional fusion-type multi-stream spatiotem poral network for automatic modulation classification,” Scientific Reports , vol. 14, no. 1, p. 22401, 2024

    2024

  27. [35]

    Decoder calibration with ultra small current sample set for intraco rtical brain– machine interface,

    P . Zhang, X. Ma, L. Chen, J. Zhou, C. Wang, W. Li, and J. He, “Decoder calibration with ultra small current sample set for intraco rtical brain– machine interface,” Journal of Neural Engineering , vol. 15, no. 2, p. 026019, 2018

    2018

  28. [36]

    Cutm ix: Reg- ularization strategy to train strong classifiers with local izable features,

    S. Y un, D. Han, S. J. Oh, S. Chun, J. Choe, and Y . Y oo, “Cutm ix: Reg- ularization strategy to train strong classifiers with local izable features,” in Proc. IEEE/CVF Int’l Conf. on Computer Vision , Seoul, South Korea, Oct. 2019, pp. 6023–6032

    2019

  29. [37]

    mix up: Beyond empirical risk minimization,

    H. Zhang, M. Cisse, Y . N. Dauphin, and D. Lopez-Paz, “mix up: Beyond empirical risk minimization,” in Proc. Int’l Conf. on Learning Representations, Toulon, France, Apr. 2017

    2017

  30. [38]

    IFNet: An interactive freq uency con- volutional neural network for enhancing motor imagery deco ding from EEG,

    J. Wang, L. Y ao, and Y . Wang, “IFNet: An interactive freq uency con- volutional neural network for enhancing motor imagery deco ding from EEG,” IEEE Trans. on Neural Systems and Rehabilitation Engineeri ng, vol. 31, pp. 1900–1911, 2023

    1900

  31. [39]

    Attention spiking neural networks,

    M. Y ao, G. Zhao, H. Zhang, Y . Hu, L. Deng, Y . Tian, B. Xu, an d G. Li, “Attention spiking neural networks,” IEEE Trans. on Pattern Analysis and Machine Intelligence , vol. 45, no. 8, pp. 9393–9410, 2023

    2023

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.