Pith. sign in

REVIEW 3 major objections 5 minor 60 references

TSkips: Efficiency Through Explicit Temporal Delay Connections in Spiking Neural Networks

T0 review · 3 major / 5 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 The paper proposes TSkips, forward and backward skip connections with explicit temporal delays, and reports that they cut optical-flow error by up to 18% and raise classification accuracy by up to 16% on four event-based benchmarks.

desk verdict Solid method paper with real gains, but Eq. (4) makes backward TSkips acausal as written—needs clarification before the central claim is fully trusted. read the letter →

arxiv 2411.16711 v2 pith:LGUWM3PS submitted 2024-11-22 cs.NE

Prajna G. Malettira , Shubham Negi , Wachirawit Ponghiran , Kaushik Roy This is my paper · ORCID
classification cs.NE
keywords spikingneuralnetworkstemporalskipconnectionsexplicitdelaysevent-basedvisionaudiotraining-freearchitecturesearchopticalflowestimationLIFneurons
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

TSkips proposes a small architectural change with a clear hypothesis: if a spiking network's skip connections carry an explicit temporal delay, the network can route information from an earlier time step directly to layers it would otherwise not reach, and this improves learning on event streams. The paper tests this on four event-based tasks—optical flow, hand-gesture recognition, and two speech benchmarks—by adding forward and backward delayed skips to spiking and hybrid backbones. It reports consistent gains with almost no added parameters: up to 18% lower average endpoint error on DSEC-flow, 8% higher accuracy on DVS128 Gesture, and up to 8% and 16% higher accuracy on SHD and SSC. Because the space of possible skip positions and delays grows exponentially, the authors use a training-free architecture-search score to pick configurations, and they report that the selected networks also converge faster and can often be run at inference without the delayed connection while keeping most of the gain. If correct, the result would make temporal wiring a practical and cheap design axis for time-sensitive spiking models.

What carries the argument

The central object is the TSkip connection: a forward or backward shortcut between layers $l$ and $l\pm k$ that carries the source layer's activation from time step $t-\Delta t$ instead of the current step, mixed with the current activation through a learnable coefficient $\alpha$. By unrolling the network in time, a TSkip creates a direct route across both space and time, so spikes and gradients can bypass many intermediate layers and time steps. The paper pairs this wiring with adaptive leaky integrate-and-fire (LIF) neurons, surrogate-gradient training, and a training-free neural architecture search that scores candidate configurations by the Sparsity-Aware Hamming Distance of their spike patterns at initialization, letting it pick skip positions, delays, and depth without training each candidate.

What would settle it

Run a controlled comparison on SHD: train the top-ranked TSkip configurations from the search score and an equal number of randomly sampled TSkip configurations under the same architecture budget and training protocol. If random configurations match or beat the selected ones in mean accuracy, or if the ranking correlation for TSkip networks falls to near zero on a larger hold-out sample, then the search is not carrying the reported improvement.

Watch

Extended reading notes

Core claim

The paper's central claim is that explicit temporal delays in forward and backward skip connections—TSkips—let SNNs and hybrid ANN-SNN models capture long-term spatio-temporal dependencies that ordinary skips (delay zero forward, delay one backward) miss. The layer update becomes $h^t_l = f_l(h^t_{l-1} \oplus W_s(\alpha h^t_{l\pm k} + (1-\alpha) h^{t-\Delta t}_{l\pm k}))$, where $\Delta t$ is chosen per connection and $\alpha$ is a learned mixing coefficient. Across DSEC-flow, DVS128 Gesture, SHD, and SSC, the TSkips-augmented networks beat their baselines by up to 18% in AEE reduction and 8–16 percentage points in accuracy, with smaller TSkip models matching larger baselines. The paper attributes these gains to better spike flow, reduced vanishing-spike effects, and finer control over spike timing, and it argues that the improvements reflect better learning rather than extra connectivity because removing the skip at inference retains most of the accuracy gain.

Load-bearing premise

The load-bearing premise is that the training-free architecture-search score reliably ranks TSkip configurations, so the reported gains come from a systematic property of TSkips rather than from a few lucky sampled architectures; the paper's own validation shows the ranking-correlation measure drops from 0.63 on baselines to 0.58 and 0.55 on TSkip networks.

Editorial extensions

If this is right

  • TSkips would make temporal wiring a first-class design choice: skipping a layer is no longer only a spatial shortcut but also a way to choose how many time steps of context feed into a computation.
  • Smaller TSkip-augmented models can match or beat larger baselines on optical flow and speech, so the mechanism could transfer to other latency- or energy-constrained event-based tasks.
  • The same delayed-skip mechanism improves fully spiking, hybrid ANN-SNN, and plain ANN models, which points to a general temporal-processing limitation rather than a quirk of one architecture.
  • Because trained TSkip networks retain most of their accuracy when the delayed connection is removed at inference, deployment could skip storing hidden temporal states, lowering memory and energy use on neuromorphic or edge hardware.

Reading between the lines

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

  • A testable extension is to let the search choose a different delay for every skip on long, non-stationary streams; the paper fixes each delay per connection, and it is plausible that adaptive per-window delays would help where event statistics drift.
  • The no-skip inference result hints that TSkips may function partly as a training-time regularizer that reshapes learned dynamics; if so, the same benefit might be obtainable from a gradient-shaping or initialization trick without the extra connection at all.
  • TSkips could be combined with per-weight or per-layer learnable delays: use the training-free search to set the coarse delay structure and let training fine-tune the fine-grained delays, covering a spectrum not explored in the paper.
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 / 5 minor

Summary. The paper proposes TSkips, skip connections with explicit temporal delays added between non-adjacent layers of spiking and hybrid ANN-SNN networks, and uses the training-free NAS method NASWOT-SAHD to select delay values, skip positions, and network depth. Experiments on DSEC-flow, DVS128 Gesture, SHD, SSC, and CIFAR10-DVS report accuracy and efficiency gains over baseline SNNs, with backward TSkips often performing best. The paper also includes ablations on delay, position, and depth, energy estimates, and a three-seed robustness study on DSEC-flow.

Significance. If the mechanism is correctly specified, TSkips are a simple, low-parameter architectural modification with consistent accuracy gains across four event-based benchmark tasks, and the use of training-free NAS avoids expensive architecture search. The paper contains useful ablations and energy estimates, and the three-seed DSEC-flow results in Appendix D.2 are a good reproducibility practice. However, the central mechanism is not fully specified for backward TSkips, the efficiency claim is stated ambiguously, and the main classification results are single-seed with a moderate NAS proxy correlation, so the strengths do not yet justify acceptance.

major comments (3)
  1. [Section 3.3, Eq. (4)] The backward TSkip as written contains a same-timestep term α h^t_{l+k} that is not causally computable in a feedforward SNN processed layer by layer within a time step: for k>0, h^t_{l+k} is not yet available when h^t_l is computed. The constraint 0<t−Δt<T in Section 3.3 restricts only the temporal index and does not resolve this layer dependency. Since backward TSkips are the best or near-best variant on SHD, SSC, and DVS128 Gesture (Tables 2–3), the reported gains depend on resolving this ambiguity. Please state explicitly whether α=0 for backward paths (in which case Eq. (4) misdescribes the model actually run), or specify the alternative intra-timestep scheduling or iterative scheme used, and provide pseudocode or code for the forward pass.
  2. [Section 4.3 / Appendix B.1] The energy-efficiency claim is stated in contradictory ways. At a fixed model scale, Table 5 shows that adding TSkips increases estimated inference energy (e.g., Base baseline 23.3 mJ vs. 29.4 mJ for forward and 27.6 mJ for backward; Pico baseline 1.90 mJ vs. 2.07/2.02 mJ), so the opening sentence of Appendix B.1, which says TSkips reduce AEE 'without increasing model complexity or energy consumption,' is not supported by the paper's own table. The energy savings in Fig. 1 come from comparing a smaller TSkip-augmented model with a larger baseline, which is a legitimate efficiency argument, but it must be stated as a cross-model comparison and the same-scale increase should be acknowledged. Please correct the wording and clarify the basis of the 40.75% energy-reduction figure.
  3. [Appendix D.1 / Section 3.4] The NAS proxy used to select TSkip architectures has Kendall's tau of 0.58 for forward and 0.55 for backward TSkip searches, below the 0.63 reported for the baseline, and the paper does not report the number of architectures sampled in the random search described in Section 3.4. Because Tables 2 and 3 report single-seed results, the main classification gains could in principle reflect favorable selected configurations rather than a systematic property of TSkips. Please provide multi-seed mean±std for the main classification results, or at least for one of the three classification datasets, and report the search budget; the three-seed DSEC-flow analysis in Appendix D.2 is a good model to follow.
minor comments (5)
  1. [Abstract / Table 1] The abstract and Section 1 claim 'up to 18%' AEE reduction on DSEC-flow, but the largest reduction in Table 1 is 17.0% (Base + Forward, from 1.35 to 1.12); please reconcile the number or state the comparison basis.
  2. [Figure 1] The energy labels 'E?52%, E?36%, E?56%, E?19%' are difficult to read and are not explained in the caption; please state what these percentages compare and which baseline is used.
  3. [Section 2.2] There are several typographical errors, including 'evalauted', 'tempral', and 'achievess'; please proofread the text.
  4. [Section 4.2 / Figure 6] The ablation study plots do not indicate whether the reported accuracies are single runs or averaged over seeds; please state the number of runs and, if available, error bars, since the paper argues that delay, position, and depth have a systematic effect.
  5. [Section 5] The discussion says trained TSkips models can perform well 'without explicitly using the temporal skip at inference,' but the Appendix B.2/C.2 experiments replace the skip input with a zero tensor, which is not the same as removing the stored historical state on hardware; please clarify which operation is meant.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the TSkips gains are measured on external benchmarks and the NAS ranking is an independent, training-free heuristic.

full rationale

The paper's central claim is an empirical architecture result: augmenting SNN/hybrid backbones with delayed forward/backward skips improves AEE and accuracy on DSEC-flow, DVS128 Gesture, SHD, SSC, and CIFAR10-DVS. These numbers are measured post-training against baselines trained under the same protocol, so no fitted parameter is renamed as a prediction. Equations (3)-(4) define the TSkips mechanism rather than deriving a target from an input; the learned alpha in Eq. (4) is a trained weight, not a fitted proxy for the reported accuracies. The NASWOT-SAHD search (Kim et al., 2022) is a training-free ranking heuristic; the paper reports its Kendall tau (0.58 forward, 0.55 backward) on TSkips-augmented networks, which is a validation of the proxy rather than a circular restatement of results. Self-citations to Kosta & Roy (2023), Negi et al. (2024), and Ponghiran et al. (2023) supply backbones, adaptive-LIF components, and evaluation splits, but the TSkips comparisons are new measurements and do not reduce to those citations; no uniqueness theorem is imported. The strongest manuscript-level concern is not circularity: Eq. (4)'s same-timestep deeper-layer term alpha h^t_{l+k} for backward skips creates a layer-wise causal dependency that the paper does not resolve, but this is a correctness/implementation ambiguity, not a derivation that equals its own input. Overall, the derivation chain is self-contained with respect to the claimed contributions.

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

The paper contributes an architectural mechanism plus NAS search; it relies on standard SNN components (LIF, surrogate gradients, adaptive LIF), a NAS proxy with measured correlation 0.55 to 0.58, and hyperparameters (delta, skip positions, architecture widths) selected on validation data. No new physical entities are invented.

free parameters (4)
  • Temporal delay Δt = per-dataset values (e.g., DSEC 3-4, DVS 5-8, SHD/SSC 8-24)
    Selected by NAS random search over a constrained range (Appendix D.1) using validation accuracy; no first-principles rule determines these values.
  • TSkip position (origin and destination layers) = varying per model, listed in Table 13 and Table 7
    Search parameter of NAS; final positions are data-dependent choices that affect results.
  • Architecture widths and depths = various, e.g., 700-124-288-144-20 for SHD Baseline-1
    Channel counts, kernel sizes, strides, and depth are searched by NAS on validation accuracy, so the reported networks are selected, not derived.
  • Learnable scaling factor α = trained, not reported in final values
    Introduced in Eq. 4 to weight current vs delayed input; trained by gradient descent and used in the 'no TSkips' inference analysis, but final values are not reported.
assumptions (4)
  • standard math LIF neuron dynamics with hard/soft reset (Eqs. 1-2) and ArcTangent surrogate gradient trainability
    Neuron model and training procedure used throughout; standard in the SNN literature.
  • domain assumption NASWOT-SAHD score correlates with accuracy for TSkips-augmented networks
    The whole search relies on this proxy; the paper measures Kendall tau 0.55 to 0.58 for TSkips networks, lower than the 0.63 baseline (Appendix D.1, Fig. 12).
  • domain assumption Explicit temporal delays do not feed future information, enforced by 0 < t - Δt < T
    Constraint stated in Section 3.3; backward skips access deeper layers at earlier time steps, assuming layer-wise unrolled computation order.
  • domain assumption Custom test split for DSEC-flow following Ponghiran et al. 2023 is an unbiased evaluation
    Evaluation protocol described in Section 4.1; DSEC-flow has no standard test split, so results depend on this custom split.

how reviews work

0 comments
Cite this review

Pith. "Pith review of TSkips: Efficiency Through Explicit Temporal Delay Connections in Spiking Neural Networks." pith.science (2026). https://pith.science/paper/LGUWM3PS

@misc{pith2026241116711,
  author       = {Pith},
  title        = {Pith review of: TSkips: Efficiency Through Explicit Temporal Delay Connections in Spiking Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LGUWM3PS}},
  note         = {Machine review of arXiv:2411.16711}
}
read the original abstract

Spiking Neural Networks (SNNs) with their bio-inspired Leaky Integrate-and-Fire (LIF) neurons inherently capture temporal information. This makes them well-suited for sequential tasks like processing event-based data from Dynamic Vision Sensors (DVS) and event-based speech tasks. Harnessing the temporal capabilities of SNNs requires mitigating vanishing spikes during training, capturing spatio-temporal patterns and enhancing precise spike timing. To address these challenges, we propose TSkips, augmenting SNN architectures with forward and backward skip connections that incorporate explicit temporal delays. These connections capture long-term spatio-temporal dependencies and facilitate better spike flow over long sequences. The introduction of TSkips creates a vast search space of possible configurations, encompassing skip positions and time delay values. To efficiently navigate this search space, this work leverages training-free Neural Architecture Search (NAS) to identify optimal network structures and corresponding delays. We demonstrate the effectiveness of our approach on four event-based datasets: DSEC-flow for optical flow estimation, DVS128 Gesture for hand gesture recognition and Spiking Heidelberg Digits (SHD) and Spiking Speech Commands (SSC) for speech recognition. Our method achieves significant improvements across these datasets: up to 18% reduction in Average Endpoint Error (AEE) on DSEC-flow, 8% increase in classification accuracy on DVS128 Gesture, and up to 8% and 16% higher classification accuracy on SHD and SSC, respectively.

Figures

Figures reproduced from arXiv: 2411.16711 by the authors.

Figure 1
Figure 1. Model size vs. average endpoint error (AEE) for [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. LIF Neuron When the membrane potential surpasses the volt￾age threshold, the neuron emits a spike. The first term in Eq. (1) denotes the leakage in the membrane potential, the second term computes the weighted summation of output spikes from layer (l − 1) and the third term denotes the reduction in membrane potential when an output spike is generated at layer l. The membrane potential is then reset, either to be zer… view at source ↗
Figure 3
Figure 3. Illustration of TSkips in an SNN. (a) A 3-layer SNN with forward and backward TSkips, annotated with their origin and destination layers and the delay (∆t). (b) Unrolled SNN with T = 4 with a forward TSkips and ∆t = 3 compared with a standard skip connection (no delay). (c) Unrolled SNN with T = 4 with a backward TSkips and ∆t = 3 compared with a standard backward connection (∆t = 1) (Rumelhart et al., 1986) Efficie… view at source ↗
Figures from the paper (9 more)
Figure 4
Figure 4. Figure 4: Illustration of the NAS Search Space. (a) Example of possible TSkips combinations in a 3-layer network with T = 6. (b) Top-ranked forward and backward TSkips with ∆t = 4 and ∆t = 3, respectively, found by (Kim et al., 2022). For visual clarity, the figure depicts the T…
Figure 5
Figure 5. Figure 5: Fully spiking EV-FlowNet (Zhu et al., 2018a) architecture with forward and backward [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]
Figure 6
Figure 6. Figure 6: Ablation study results on the SHD dataset demonstrating the impact of varying the (a) temporal [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Qualitative results on the DSEC-flow dataset. (a) Ground Truth (GT) mask, (b) Mini (3.4M) SNN [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Trained network’s response to temporal patterns at inference, visualized by average spiking activity. [PITH_FULL_IMAGE:figures/full_fig_p012_8.png]
Figure 9
Figure 9. Figure 9: Ablation study on the SHD dataset demonstrating the impact of varying the (a) temporal delay [PITH_FULL_IMAGE:figures/full_fig_p020_9.png]
Figure 10
Figure 10. Figure 10: Convergence analysis of fully spiking (Kosta & Roy, 2023) and hybrid (Negi et al., 2024) EV [PITH_FULL_IMAGE:figures/full_fig_p022_10.png]
Figure 11
Figure 11. Figure 11: Test accuracy convergence on the SHD dataset, comparing the best-performing models with [PITH_FULL_IMAGE:figures/full_fig_p024_11.png]
Figure 12
Figure 12. Figure 12: Correlation between accuracy and SAHD score for different net [PITH_FULL_IMAGE:figures/full_fig_p026_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

60 extracted references · 47 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION format.date year duplicate empty "emp...

  2. [2]

    Lapicque’s introduction of the integrate-and-fire model neuron (1907)

    Larry F Abbott. Lapicque’s introduction of the integrate-and-fire model neuron (1907). Brain research bulletin, 50 0 (5-6): 0 303--304, 1999

    work page 1907

  3. [3]

    Skip connections in spiking neural networks: An analysis of their effect on network training, 2023

    Hadjer Benmeziane, Amine Ziad Ounnoughene, Imane Hamzaoui, and Younes Bouhadjar. Skip connections in spiking neural networks: An analysis of their effect on network training, 2023

    work page 2023

  4. [4]

    A surrogate gradient spiking baseline for speech command recognition

    Alexandre Bittar and Philip N Garner. A surrogate gradient spiking baseline for speech command recognition. Frontiers in Neuroscience, 16: 0 865897, 2022

    work page 2022

  5. [5]

    A 240 180 130 db 3 s latency global shutter spatiotemporal vision sensor

    Christian Brandli, Raphael Berner, Minhao Yang, Shih-Chii Liu, and Tobi Delbruck. A 240 180 130 db 3 s latency global shutter spatiotemporal vision sensor. IEEE Journal of Solid-State Circuits, 49 0 (10): 0 2333--2341, 2014

    work page 2014

  6. [6]

    Once-for-all: Train one network and specialize it for efficient deployment, 2020

    Han Cai, Chuang Gan, Tianzhe Wang, Zhekai Zhang, and Song Han. Once-for-all: Train one network and specialize it for efficient deployment, 2020

    2020

  7. [7]

    Spiking deep convolutional neural networks for energy-efficient object recognition

    Yongqiang Cao, Yang Chen, and Deepak Khosla. Spiking deep convolutional neural networks for energy-efficient object recognition. International Journal of Computer Vision, 113: 0 54--66, 2015

    work page 2015

  8. [8]

    One timestep is all you need: Training spiking neural networks with ultra low latency, 2021

    Sayeed Shafayet Chowdhury, Nitin Rathi, and Kaushik Roy. One timestep is all you need: Training spiking neural networks with ultra low latency, 2021

    work page 2021

Show all 60 references
  1. [9]

    The heidelberg spiking data sets for the systematic evaluation of spiking neural networks

    Benjamin Cramer, Yannik Stradmann, Johannes Schemmel, and Friedemann Zenke. The heidelberg spiking data sets for the systematic evaluation of spiking neural networks. IEEE Transactions on Neural Networks and Learning Systems, 33 0 (7): 0 2744--2757, 2022

    2022

  2. [10]

    Investigating current-based and gating approaches for accurate and energy-efficient spiking recurrent neural networks

    Manon Dampfhoffer, Thomas Mesquida, Alexandre Valentian, and Lorena Anghel. Investigating current-based and gating approaches for accurate and energy-efficient spiking recurrent neural networks. In International Conference on Artificial Neural Networks, pages 359--370. Springer, 2022

    2022

  3. [11]

    Rethinking the performance comparison between snns and anns

    Lei Deng, Yujie Wu, Xing Hu, Ling Liang, Yufei Ding, Guoqi Li, Guangshe Zhao, Peng Li, and Yuan Xie. Rethinking the performance comparison between snns and anns. Neural networks, 121: 0 294--307, 2020

    2020

  4. [12]

    Incorporating learnable membrane time constant to enhance learning of spiking neural networks

    Wei Fang, Zhaofei Yu, Yanqi Chen, Timoth \'e e Masquelier, Tiejun Huang, and Yonghong Tian. Incorporating learnable membrane time constant to enhance learning of spiking neural networks. In Proceedings of the IEEE/CVF international conference on computer vision, pages 2661--2671, 2021

    2021

  5. [13]

    Davison, Jörg Conradt, Kostas Daniilidis, and Davide Scaramuzza

    Guillermo Gallego, Tobi Delbrück, Garrick Orchard, Chiara Bartolozzi, Brian Taba, Andrea Censi, Stefan Leutenegger, Andrew J. Davison, Jörg Conradt, Kostas Daniilidis, and Davide Scaramuzza. Event-based vision: A survey. IEEE Transactions on Pattern Analysis and Machine Intell...

    2022

  6. [14]

    Dsec: A stereo event camera dataset for driving scenarios

    Mathias Gehrig, Willem Aarents, Daniel Gehrig, and Davide Scaramuzza. Dsec: A stereo event camera dataset for driving scenarios. IEEE Robotics and Automation Letters, 2021 a

    2021

  7. [15]

    E-raft: Dense optical flow from event cameras

    Mathias Gehrig, Mario Millhäusler, Daniel Gehrig, and Davide Scaramuzza. E-raft: Dense optical flow from event cameras. In 2021 International Conference on 3D Vision (3DV), pages 197--206, 2021 b

    2021

  8. [16]

    Self-supervised learning of event-based optical flow with spiking neural networks, 2021

    Jesse Hagenaars, Federico Paredes-Vallés, and Guido de Croon. Self-supervised learning of event-based optical flow with spiking neural networks, 2021

    2021

  9. [17]

    Learning delays in spiking neural networks using dilated convolutions with learnable spacings, 2023

    Ilyass Hammouamri, Ismail Khalfaoui-Hassani, and Timothée Masquelier. Learning delays in spiking neural networks using dilated convolutions with learnable spacings, 2023

    2023

  10. [18]

    Deep residual learning for image recognition, 2015

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition, 2015

    2015

  11. [19]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770--778, 2016

    2016

  12. [20]

    Long short-term memory

    S Hochreiter. Long short-term memory. Neural Computation MIT-Press, 1997

    1997

  13. [21]

    1.1 computing's energy problem (and what we can do about it)

    Mark Horowitz. 1.1 computing's energy problem (and what we can do about it). In 2014 IEEE International Solid-State Circuits Conference Digest of Technical Papers (ISSCC), pages 10--14, 2014

    2014

  14. [22]

    Hand gesture recognition system using the dynamic vision sensor

    Yu Hu, Ziteng Li, Xinpeng Li, Jianfeng Li, Xintong Yu, Xiaofan Chen, and Lei Wang. Hand gesture recognition system using the dynamic vision sensor. In 2022 5th International Conference on Circuits, Systems and Simulation (ICCSS), pages 102--110, 2022

    2022

  15. [23]

    Densely connected convolutional networks

    Gao Huang, Zhuang Liu, Laurens Van Der Maaten, and Kilian Q Weinberger. Densely connected convolutional networks. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 4700--4708, 2017

    2017

  16. [24]

    Klif: An optimized spiking neuron unit for tuning surrogate gradient function

    Chunming Jiang and Yilei Zhang. Klif: An optimized spiking neuron unit for tuning surrogate gradient function. Neural Computation, pages 1--15, 2024

    2024

  17. [25]

    Synaptic plasticity dynamics for deep continuous local learning (decolle)

    Jacques Kaiser, Hesham Mostafa, and Emre Neftci. Synaptic plasticity dynamics for deep continuous local learning (decolle). Frontiers in Neuroscience, 14: 0 424, 2020

    2020

  18. [26]

    Dilated convolution with learnable spacings, 2023

    Ismail Khalfaoui-Hassani, Thomas Pellegrini, and Timothée Masquelier. Dilated convolution with learnable spacings, 2023

    2023

  19. [27]

    Revisiting batch normalization for training low-latency deep spiking neural networks from scratch

    Youngeun Kim and Priyadarshini Panda. Revisiting batch normalization for training low-latency deep spiking neural networks from scratch. Frontiers in neuroscience, 15: 0 773954, 2021

    2021

  20. [28]

    Neural architecture search for spiking neural networks, 2022

    Youngeun Kim, Yuhang Li, Hyoungseob Park, Yeshwanth Venkatesha, and Priyadarshini Panda. Neural architecture search for spiking neural networks, 2022

    2022

  21. [29]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014

    2014 arXiv

  22. [30]

    Adaptive-spikenet: Event-based optical flow estimation using spiking neural networks with learnable neuronal dynamics, 2023

    Adarsh Kumar Kosta and Kaushik Roy. Adaptive-spikenet: Event-based optical flow estimation using spiking neural networks with learnable neuronal dynamics, 2023

    2023

  23. [31]

    Fusion-flownet: Energy-efficient optical flow estimation using sensor fusion and deep fused spiking-analog network architectures, 2021

    Chankyu Lee, Adarsh Kumar Kosta, and Kaushik Roy. Fusion-flownet: Energy-efficient optical flow estimation using sensor fusion and deep fused spiking-analog network architectures, 2021

    2021

  24. [32]

    Spikeformer: A novel architecture for training high-performance low-latency spiking neural network, 2022

    Yudong Li, Yunlin Lei, and Xu Yang. Spikeformer: A novel architecture for training high-performance low-latency spiking neural network, 2022

    2022

  25. [33]

    Blinkflow: A dataset to push the limits of event-based optical flow estimation

    Yijin Li, Zhaoyang Huang, Shuo Chen, Xiaoyu Shi, Hongsheng Li, Hujun Bao, Zhaopeng Cui, and Guofeng Zhang. Blinkflow: A dataset to push the limits of event-based optical flow estimation. In 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages ...

    2023

  26. [34]

    A 128 128 120 db 15 s latency asynchronous temporal contrast vision sensor

    Patrick Lichtsteiner, Christoph Posch, and Tobi Delbruck. A 128 128 120 db 15 s latency asynchronous temporal contrast vision sensor. IEEE Journal of Solid-State Circuits, 43 0 (2): 0 566--576, 2008

    2008

  27. [35]

    Sgdr: Stochastic gradient descent with warm restarts, 2017

    Ilya Loshchilov and Frank Hutter. Sgdr: Stochastic gradient descent with warm restarts, 2017

    2017

  28. [36]

    Surrogate gradient learning in spiking neural networks: Bringing the power of gradient-based optimization to spiking 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 spiking neural networks. IEEE Signal Processing Magazine, 36 0 (6): 0 51--63, 2019

    2019

  29. [37]

    Best of both worlds: Hybrid snn-ann architecture for event-based optical flow estimation, 2024

    Shubham Negi, Deepika Sharma, Adarsh Kumar Kosta, and Kaushik Roy. Best of both worlds: Hybrid snn-ann architecture for event-based optical flow estimation, 2024

    2024

  30. [38]

    Toward scalable, efficient, and accurate deep spiking neural networks with backward residual connections, stochastic softmax, and hybridization

    Priyadarshini Panda, Sai Aparna Aketi, and Kaushik Roy. Toward scalable, efficient, and accurate deep spiking neural networks with backward residual connections, stochastic softmax, and hybridization. Frontiers in Neuroscience, 14: 0 535502, 2020

    2020

  31. [39]

    Back to event basics: Self-supervised learning of image reconstruction for event cameras via photometric constancy

    Federico Paredes-Vall \'e s and Guido CHE De Croon. Back to event basics: Self-supervised learning of image reconstruction for event cameras via photometric constancy. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 3446--3455, 2021

    2021

  32. [40]

    On the difficulty of training recurrent neural networks, 2013

    Razvan Pascanu, Tomas Mikolov, and Yoshua Bengio. On the difficulty of training recurrent neural networks, 2013

    2013

  33. [41]

    Hybrid analog-spiking long short-term memory for energy efficient computing on edge devices

    Wachirawit Ponghiran and Kaushik Roy. Hybrid analog-spiking long short-term memory for energy efficient computing on edge devices. In 2021 Design, Automation & Test in Europe Conference & Exhibition (DATE), pages 581--586. IEEE, 2021

    2021

  34. [42]

    Event-based temporally dense optical flow estimation with sequential learning, 2023

    Wachirawit Ponghiran, Chamika Mihiranga Liyanagedera, and Kaushik Roy. Event-based temporally dense optical flow estimation with sequential learning, 2023

    2023

  35. [43]

    Mamba-spike: Enhancing the mamba architecture with a spiking front-end for efficient temporal data processing, 2024

    Jiahao Qin and Feng Liu. Mamba-spike: Enhancing the mamba architecture with a spiking front-end for efficient temporal data processing, 2024

    2024

  36. [44]

    Diet-snn: Direct input encoding with leakage and threshold optimization in deep spiking neural networks, 2020

    Nitin Rathi and Kaushik Roy. Diet-snn: Direct input encoding with leakage and threshold optimization in deep spiking neural networks, 2020

    2020

  37. [45]

    Lite-snn: Leveraging inherent dynamics to train energy-efficient spiking neural networks for sequential learning

    Nitin Rathi and Kaushik Roy. Lite-snn: Leveraging inherent dynamics to train energy-efficient spiking neural networks for sequential learning. IEEE Transactions on Cognitive and Developmental Systems, pages 1--10, 2024

    2024

  38. [46]

    Exploring spike-based learning for neuromorphic computing: Prospects and perspectives

    Nitin Rathi, Amogh Agrawal, Chankyu Lee, Adarsh Kumar Kosta, and Kaushik Roy. Exploring spike-based learning for neuromorphic computing: Prospects and perspectives. In 2021 Design, Automation & Test in Europe Conference & Exhibition (DATE), pages 902--907. IEEE, 2021

    2021

  39. [47]

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

    B Rueckauer, IA Lungu, Y Hu, M Pfeiffer, and SC Liu. Conversion of continuous-valued deep networks to efficient event-driven networks for image classification. front. neurosci. 11, 682 (2017), 2017

    2017

  40. [48]

    Learning internal representations by error propagation, parallel distributed processing, explorations in the microstructure of cognition, ed

    David E Rumelhart, Geoffrey E Hinton, and Ronald J Williams. Learning internal representations by error propagation, parallel distributed processing, explorations in the microstructure of cognition, ed. de rumelhart and j. mcclelland. vol. 1. 1986. Biometrika, 71: 0 599--607, 1986

    1986

  41. [49]

    Convolutional spiking neural networks for spatio-temporal feature extraction

    Ali Samadzadeh, Fatemeh Sadat Tabatabaei Far, Ali Javadi, Ahmad Nickabadi, and Morteza Haghir Chehreghani. Convolutional spiking neural networks for spatio-temporal feature extraction. Neural Processing Letters, 55 0 (6): 0 6979--6995, 2023

    2023

  42. [50]

    Eventmix: An efficient data augmentation strategy for event-based learning

    Guobin Shen, Dongcheng Zhao, and Yi Zeng. Eventmix: An efficient data augmentation strategy for event-based learning. Information Sciences, 644: 0 119170, 2023

    2023

  43. [51]

    Slayer: Spike layer error reassignment in time, 2018

    Sumit Bam Shrestha and Garrick Orchard. Slayer: Spike layer error reassignment in time, 2018

    2018

  44. [52]

    Learnable axonal delay in spiking neural networks improves spoken word recognition

    Pengfei Sun, Yansong Chua, Paul Devos, and Dick Botteldooren. Learnable axonal delay in spiking neural networks improves spoken word recognition. Frontiers in Neuroscience, 17: 0 1275944, 2023 a

    2023

  45. [53]

    Adaptive axonal delays in feedforward spiking neural networks for accurate spoken word recognition

    Pengfei Sun, Ehsan Eqlimi, Yansong Chua, Paul Devos, and Dick Botteldooren. Adaptive axonal delays in feedforward spiking neural networks for accurate spoken word recognition. In ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASS...

    2023

  46. [54]

    P.J. Werbos. Backpropagation through time: what it does and how to do it. Proceedings of the IEEE, 78 0 (10): 0 1550--1560, 1990

    1990

  47. [55]

    A new spiking convolutional recurrent neural network (scrnn) with applications to event-based hand gesture recognition

    Yannan Xing, Gaetano Di Caterina, and John Soraghan. A new spiking convolutional recurrent neural network (scrnn) with applications to event-based hand gesture recognition. Frontiers in neuroscience, 14: 0 590164, 2020

    2020

  48. [56]

    Accurate and efficient time-domain classification with adaptive spiking recurrent neural networks

    Bojian Yin, Federico Corradi, and Sander M Boht \'e . Accurate and efficient time-domain classification with adaptive spiking recurrent neural networks. Nature Machine Intelligence, 3 0 (10): 0 905--913, 2021

    2021

  49. [57]

    Long short-term memory with two-compartment spiking neuron, 2023

    Shimin Zhang, Qu Yang, Chenxiang Ma, Jibin Wu, Haizhou Li, and Kay Chen Tan. Long short-term memory with two-compartment spiking neuron, 2023

    2023

  50. [58]

    Spikingformer: Spike-driven residual learning for transformer-based spiking neural network, 2023

    Chenlin Zhou, Liutao Yu, Zhaokun Zhou, Zhengyu Ma, Han Zhang, Huihui Zhou, and Yonghong Tian. Spikingformer: Spike-driven residual learning for transformer-based spiking neural network, 2023

    2023

  51. [59]

    Ev-flownet: Self-supervised optical flow estimation for event-based cameras

    Alex Zhu, Liangzhe Yuan, Kenneth Chaney, and Kostas Daniilidis. Ev-flownet: Self-supervised optical flow estimation for event-based cameras. In Robotics: Science and Systems XIV. Robotics: Science and Systems Foundation, 2018 a

    2018

  52. [60]

    Unsupervised event-based learning of optical flow, depth, and egomotion, 2018 b

    Alex Zihao Zhu, Liangzhe Yuan, Kenneth Chaney, and Kostas Daniilidis. Unsupervised event-based learning of optical flow, depth, and egomotion, 2018 b

    2018

Pith tools

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