pith. sign in

arxiv: 2606.17775 · v2 · pith:IHM3G3GGnew · submitted 2026-06-16 · 💻 cs.SD · cs.AI· cs.NE

A Neuromorphic Trigger for Efficient Audio Event Detection

Benjamin Hatton , Oliver Rhodes , Luca Peres This is my paper

Pith reviewed 2026-06-26 23:15 UTC · model grok-4.3

classification 💻 cs.SD cs.AIcs.NE
keywords spiking neural networksneuromorphic computingaudio event detectionsound event detectionanomalous sound detectionefficient inferencetrigger mechanismedge audio processing
0
0 comments X

The pith

A lightweight spiking neural network can gate audio streams to cut downstream computation by 42 times while improving detection bounds.

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

The paper sets out to show that a simple spiking neural network can serve as an always-on front-end filter that passes only the important parts of an audio stream to a heavier classifier. If this works, continuous monitoring becomes feasible on devices that cannot afford to run full models all the time. The trigger is built as a fully connected SNN followed by a close-open filter; it is tested first on a class-agnostic version of URBAN-SED for anomalous sound detection and then paired with an existing classifier on the DCASE 2017 Task 2 dataset. The reported outcomes are a 0.97 segment-based F1 score on the first task and a 42.6-fold drop in FLOPs together with a lower event-error bound on the second.

Core claim

A neuromorphic trigger implemented as a lightweight fully connected spiking neural network with close-open post-processing identifies salient audio segments and gates them to downstream models. On class-agnostic URBAN-SED it reaches a one-second segment F1 of 0.97 for anomalous sound detection. When combined with the Dang classifier on DCASE 2017 Task 2 it yields a potential 42.6 times reduction in FLOPs and lowers the lower bound on event-based error rate from 0.41 to 0.25.

What carries the argument

lightweight fully connected spiking neural network (SNN) with close-open filter post-processing that selectively gates input segments to a heavier downstream model

If this is right

  • The trigger can be inserted as a low-cost front-end before any computationally heavy audio classifier.
  • On the DCASE 2017 Task 2 benchmark the combination lowers the event-error lower bound while cutting FLOPs by a factor of 42.6.
  • The same architecture delivers 0.97 segment-based F1 on class-agnostic URBAN-SED for anomalous sound detection.
  • Selective gating makes real-time, resource-constrained audio event detection practical by processing only the identified salient segments.

Where Pith is reading between the lines

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

  • If the trigger generalizes, similar front-ends could be applied to other continuous sensor streams such as vibration or environmental monitoring.
  • The approach opens the possibility of running multiple specialized classifiers only on demand rather than in parallel.
  • Energy measurements on actual neuromorphic hardware would be needed to confirm that the FLOP reduction produces proportional power savings.

Load-bearing premise

The SNN trigger will catch nearly all relevant audio events across varied real-world conditions without missing too many, so that the reported FLOP savings translate into actual system-level gains rather than being offset by undetected events.

What would settle it

A deployment test on continuous noisy audio in which the fraction of missed events causes the combined system's overall error rate to exceed the error rate of the downstream classifier running without any trigger.

Figures

Figures reproduced from arXiv: 2606.17775 by Benjamin Hatton, Luca Peres, Oliver Rhodes.

Figure 1
Figure 1. Figure 1: Proposed pipeline for more efficient processing of data. The audio is converted into a Mel spectrogram that is then [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Comparison of the error rate (AEER) and the num [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: A diagram of the closing then the opening of a spike [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 1
Figure 1. Figure 1: The input of the trigger is a Mel spectrogram using 128 [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 4
Figure 4. Figure 4: A comparison of the theoretical FLOP count assum [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
read the original abstract

Efficient processing of continuous audio streams remains a key challenge for real-time and resource-constrained systems. This paper introduces a neuromorphic trigger for audio event detection, based on a spiking neural network (SNN) that selectively gates input to downstream models. The proposed neuromorphic trigger acts as a flexible low-cost front-end, identifying salient audio segments and enabling these to be processed by a more computationally intensive model for tasks such as classification. The trigger is implemented as a lightweight fully connected SNN using a close-open filter for postprocessing, and is evaluated on two representative tasks: Anomalous Sound Detection (ASD) and Sound Event Detection (SED). For ASD, the trigger achieves a one-second segment-based F1 score of 0.97 on a class-agnostic form of the URBAN-SED dataset, demonstrating high reliability in identifying relevant audio regions. For SED, the trigger is combined with the Dang classifier on the DCASE 2017 Challenge Task 2 dataset, showing a potential $42.6\times$ reduction in FLOPs while reducing the lower bound of the event-based error rate from 0.41 to 0.25. These results highlight the potential of neuromorphic triggers as real-time, energy-efficient front-end filters, enabling substantial reductions in computational cost.

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 / 0 minor

Summary. The paper introduces a neuromorphic trigger implemented as a lightweight fully connected spiking neural network (SNN) with a close-open filter postprocessor. This acts as a low-cost front-end to identify salient audio segments and gate them to downstream models for anomalous sound detection (ASD) and sound event detection (SED). On a class-agnostic URBAN-SED dataset the trigger reports a one-second segment-based F1 of 0.97; when paired with the Dang classifier on DCASE 2017 Task 2 it claims a 42.6× FLOPs reduction while lowering the event-based error-rate lower bound from 0.41 to 0.25.

Significance. If the empirical claims are substantiated with complete methodology and hardware validation, the work would demonstrate a practical neuromorphic front-end that materially reduces compute for continuous audio pipelines while preserving detection performance, addressing a recognized bottleneck in real-time, resource-constrained audio systems.

major comments (2)
  1. [Evaluation / ASD results] Evaluation / ASD results: the abstract states a segment-based F1 of 0.97 on class-agnostic URBAN-SED, yet no training procedure, validation splits, hyper-parameter search, or error bars are supplied; without these the headline reliability claim cannot be assessed and is load-bearing for the central efficiency argument.
  2. [SED combination results] SED combination results: the reported 42.6× FLOPs reduction and error-rate improvement (0.41 → 0.25) do not state whether trigger overhead is included in the count, nor do they provide neuromorphic-hardware energy measurements or stress tests under noise/domain shift; both omissions directly affect the practical-efficiency claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and outline the revisions that will be incorporated to improve clarity and completeness.

read point-by-point responses

  1. Referee: [Evaluation / ASD results] Evaluation / ASD results: the abstract states a segment-based F1 of 0.97 on class-agnostic URBAN-SED, yet no training procedure, validation splits, hyper-parameter search, or error bars are supplied; without these the headline reliability claim cannot be assessed and is load-bearing for the central efficiency argument.

    Authors: We agree that the current manuscript does not provide adequate detail on the training procedure, validation splits, hyper-parameter search, or error bars for the reported 0.97 segment-based F1 on class-agnostic URBAN-SED. These elements are required to substantiate the reliability of the result. In the revised manuscript we will add a dedicated experimental setup subsection that specifies the data partitioning, the hyper-parameter selection process, the optimization details, and error bars obtained across multiple random seeds. This addition will directly support assessment of the headline claim. revision: yes

  2. Referee: [SED combination results] SED combination results: the reported 42.6× FLOPs reduction and error-rate improvement (0.41 → 0.25) do not state whether trigger overhead is included in the count, nor do they provide neuromorphic-hardware energy measurements or stress tests under noise/domain shift; both omissions directly affect the practical-efficiency claim.

    Authors: We will revise the text to explicitly state that the 42.6× FLOPs reduction figure incorporates the computational overhead of the trigger, obtained by comparing total operations in the gated pipeline against the ungated baseline. However, the evaluation remains at the level of algorithmic simulation; no neuromorphic hardware energy measurements were performed. Likewise, while the URBAN-SED experiments include some acoustic variability, dedicated stress tests under additional noise or domain-shift conditions were not conducted. We will add a limitations paragraph in the discussion to acknowledge these points and thereby qualify the practical-efficiency claims. revision: partial

Circularity Check

0 steps flagged

No circularity: empirical results on public datasets with no self-referential derivations.

full rationale

The paper reports empirical F1 scores, error rates, and FLOPs reductions from evaluations on standard public datasets (URBAN-SED, DCASE 2017 Task 2). No equations, predictions, or uniqueness claims reduce by construction to fitted inputs or self-citations. The trigger is a lightweight FC-SNN with close-open filter; performance metrics are measured outcomes, not tautological. No load-bearing self-citation chains or ansatzes imported from prior author work appear in the provided text. This is a standard empirical systems paper whose central claims rest on external benchmarks rather than internal redefinitions.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on empirical results from the proposed SNN trigger architecture applied to two audio tasks; no free parameters are explicitly fitted in the abstract, and the main addition is the application rather than new theoretical elements.

axioms (1)
  • domain assumption A lightweight fully connected spiking neural network can serve as an effective front-end trigger for identifying salient audio segments.
    Invoked in the design and evaluation of the trigger without further justification provided in the abstract.
invented entities (1)
  • neuromorphic trigger no independent evidence
    purpose: To selectively gate audio input to downstream computationally intensive models
    Introduced as the core new component based on the SNN implementation.

pith-pipeline@v0.9.1-grok · 5758 in / 1458 out tokens · 43167 ms · 2026-06-26T23:15:04.256283+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

47 extracted references · 17 canonical work pages

  1. [1]

    L. F. Abbott. 1999. Lapicque’s Introduction of the Integrate-and-Fire Model Neuron (1907).Brain Research Bulletin50, 5–6 (1999), 303–304. doi:10.1016/S0361- 9230(99)00161-6

    work page doi:10.1016/s0361- 1999

  2. [2]

    Sharath Adavanne, Giambattista Parascandolo, Pasi Pertilä, Toni Heittola, and Tuomas Virtanen. 2017. Sound Event Detection in Multichannel Audio Using Spatial and Harmonic Features. arXiv:1706.02293 [cs.SD] https://arxiv.org/abs/ 1706.02293

    Pith/arXiv arXiv 2017

  3. [3]

    Christopher M. Bishop. 2006.Pattern Recognition and Machine Learning. Springer

    2006

  4. [4]

    Peter Blouw, Xuan Choo, Eric Hunsberger, and Chris Eliasmith. 2019. Benchmark- ing Keyword Spotting Efficiency on Neuromorphic Hardware. InProceedings of the 7th Annual Neuro-Inspired Computational Elements Workshop (NICE ’19). Association for Computing Machinery, Article 1. doi:10.1145/3320288.3320304

    work page doi:10.1145/3320288.3320304 2019

  5. [5]

    2017.Convolutional Recurrent Neural Networks for Rare Sound Event Detection

    Emre Cakir and Tuomas Virtanen. 2017.Convolutional Recurrent Neural Networks for Rare Sound Event Detection. Technical Report. DCASE2017 Challenge

    2017

  6. [6]

    Gianmarco Cerutti, Renzo Andri, Lukas Cavigelli, Elisabetta Farella, Michele Magno, and Luca Benini. 2020. Sound event detection with binary neural net- works on tightly power-constrained IoT devices. InProceedings of the ACM/IEEE International Symposium on Low Power Electronics and Design (ISLPED ’20). Asso- ciation for Computing Machinery, 19–24. doi:10...

    work page doi:10.1145/3370748.3406588 2020

  7. [7]

    Iulia-Maria Comşa, Luca Versari, Thomas Fischbacher, and Jyrki Alakuijala. 2021. Spiking Autoencoders With Temporal Coding.Frontiers in NeuroscienceVolume 15 - 2021 (2021). doi:10.3389/fnins.2021.712667

    work page doi:10.3389/fnins.2021.712667 2021

  8. [8]

    Benjamin Cramer, Yannik Stradmann, Johannes Schemmel, and Friedemann Zenke. 2022. The Heidelberg Spiking Data Sets for the Systematic Evaluation of Spiking Neural Networks.IEEE Transactions on Neural Networks and Learning Systems33, 7 (July 2022), 2744–2757. doi:10.1109/tnnls.2020.3044364

    work page doi:10.1109/tnnls.2020.3044364 2022

  9. [9]

    2017.Deep Learning for DCASE2017 Challenge

    An Dang, Toan Vu, and Jia-Ching Wang. 2017.Deep Learning for DCASE2017 Challenge. Technical Report. DCASE2017 Challenge

    2017

  10. [10]

    Toledano

    Diego De Benito-Gorrón, Daniel Ramos, and Doroteo T. Toledano. 2021. A Multi- Resolution CRNN-Based Approach for Semi-Supervised Sound Event Detection in DCASE 2020 Challenge.IEEE Access9 (2021), 89029–89042. doi:10.1109/ ACCESS.2021.3088949

    arXiv 2021

  11. [11]

    Jason K Eshraghian, Max Ward, Emre Neftci, Xinxin Wang, Gregor Lenz, Girish Dwivedi, Mohammed Bennamoun, Doo Seok Jeong, and Wei D Lu. 2023. Training spiking neural networks using lessons from deep learning.Proc. IEEE111, 9 (2023), 1016–1054

    2023

  12. [12]

    2016.Deep Learning

    Ian Goodfellow, Yoshua Bengio, and Aaron Courville. 2016.Deep Learning. MIT Press

    2016

  13. [13]

    Song Han, Jeff Pool, John Tran, and William J. Dally. 2015. Learning both weights and connections for efficient neural networks. InProceedings of the 29th Interna- tional Conference on Neural Information Processing Systems - Volume 1(Montreal, Canada)(NIPS’15). MIT Press, Cambridge, MA, USA, 1135–1143

    2015

  14. [14]

    Soroush Heydari and Qusay H. Mahmoud. 2025. Tiny Machine Learning and On- Device Inference: A Survey of Applications, Challenges, and Future Directions. Sensors25, 10 (2025). doi:10.3390/s25103191

    work page doi:10.3390/s25103191 2025

  15. [15]

    Yohei Kawaguchi and Takashi Endo. 2017. How can we detect anomalies from subsampled audio signals?. In2017 IEEE 27th International Workshop on Machine Learning for Signal Processing (MLSP). 1–6. doi:10.1109/MLSP.2017.8168164

    work page doi:10.1109/mlsp.2017.8168164 2017

  16. [16]

    Seijoon Kim, Seongsik Park, Byunggook Na, and Sungroh Yoon. 2020. Spiking- YOLO: Spiking Neural Network for Energy-Efficient Object Detection.Proceedings of the AAAI Conference on Artificial Intelligence34 (04 2020), 11270–11277. doi:10. 1609/aaai.v34i07.6787

    2020

  17. [17]

    Naoki Koga, Yoshiaki Bando, and Keisuke Imoto. 2024. LEAD Dataset: How Can Labels for Sound Event Detection Vary Depending on Annotators?. In2024 Asia Pacific Signal and Information Processing Association Annual Summit and Conference (APSIPA ASC). 1–6. doi:10.1109/APSIPAASC63619.2025.10848643

    work page doi:10.1109/apsipaasc63619.2025.10848643 2024

  18. [18]

    Edgar Lemaire, Loïc Cordone, Andrea Castagnetti, Pierre-Emmanuel Novac, Jonathan Courtois, and Benoît Miramond. 2022. An Analytical Estimation of Spiking Neural Networks Energy Efficiency. InNeural Information Processing. Springer, 574–587. doi:10.1007/978-3-031-30105-6_48

    work page doi:10.1007/978-3-031-30105-6_48 2022

  19. [19]

    2017.The SEIE-SCUT Systems for IEEE AASP Chal- lenge on DCASE 2017: Deep Learning Techniques for Audio Representation and Classification

    Yanxiong Li and Xianku Li. 2017.The SEIE-SCUT Systems for IEEE AASP Chal- lenge on DCASE 2017: Deep Learning Techniques for Audio Representation and Classification. Technical Report. DCASE2017 Challenge

    2017

  20. [20]

    2017.Rare Sound Event Detection Using 1D Convolutional Recurrent Neural Networks

    Hyungui Lim, Jeongsoo Park, and Yoonchang Han. 2017.Rare Sound Event Detection Using 1D Convolutional Recurrent Neural Networks. Technical Report. DCASE2017 Challenge

    2017

  21. [21]

    Rui Lu. 2017. BIDIRECTIONAL GRU FOR SOUND EVENT DETECTION. https: //api.semanticscholar.org/CorpusID:209452474

    2017

  22. [22]

    Iván López-Espejo, Zheng-Hua Tan, John H. L. Hansen, and Jesper Jensen. 2022. Deep Spoken Keyword Spotting: An Overview.IEEE Access10 (2022), 4169–4199. doi:10.1109/ACCESS.2021.3139508

    work page doi:10.1109/access.2021.3139508 2022

  23. [23]

    Wolfgang Maass. 1997. Networks of spiking neurons: The third generation of neural network models.Neural Networks10, 9 (1997), 1659–1671. doi:10.1016/ S0893-6080(97)00011-7

    1997

  24. [24]

    1975.Random Sets and Integral Geometry

    Georges Matheron. 1975.Random Sets and Integral Geometry. Wiley

    1975

  25. [25]

    Mesaros, A

    A. Mesaros, A. Diment, B. Elizalde, T. Heittola, E. Vincent, B. Raj, and T. Virtanen

  26. [26]

    doi:10.1109/TASLP.2019

    Sound event detection in the DCASE 2017 Challenge.IEEE/ACM Transac- tions on Audio, Speech, and Language Processing(2019). doi:10.1109/TASLP.2019. 2907016 In press

    work page doi:10.1109/taslp.2019 2017

  27. [27]

    Annamaria Mesaros, Toni Heittola, Aleksandr Diment, Benjamin Elizalde, Ankit Shah, Emmanuel Vincent, Bhiksha Raj, and Tuomas Virtanen. 2017. DCASE 2017 Challenge setup: Tasks, datasets and baseline system. InDCASE 2017 - Workshop on Detection and Classification of Acoustic Scenes and Events. Munich, Germany. https://inria.hal.science/hal-01627981

    2017

  28. [28]

    Annamaria Mesaros, Toni Heittola, and Tuomas Virtanen. 2016. Metrics for Polyphonic Sound Event Detection.Applied Sciences6, 6 (2016). doi:10.3390/ app6060162

    2016

  29. [29]

    Neftci, Hesham Mostafa, and Friedemann Zenke

    Emre O. Neftci, Hesham Mostafa, and Friedemann Zenke. 2019. Surrogate Gradi- ent Learning in Spiking Neural Networks: Bringing the Power of Gradient-Based Optimization to Spiking Neural Networks.IEEE Signal Processing Magazine36, 6 (2019), 51–63. doi:10.1109/MSP.2019.2931595

    work page doi:10.1109/msp.2019.2931595 2019

  30. [30]

    Huy Phan, Martin Krawczyk-Becker, Timo Gerkmann, and Alfred Mertins. 2017. DNN and CNN with Weighted and Multi-Task Loss Functions for Audio Event Detection. Technical Report. DCASE2017 Challenge

    2017

  31. [31]

    2017.Bosch Rare Sound Events Detection Systems for DCASE2017 Challenge

    Anravich Ravichandran and Samarjit Das. 2017.Bosch Rare Sound Events Detection Systems for DCASE2017 Challenge. Technical Report. DCASE2017 Challenge

    2017

  32. [32]

    Justin Salamon, Christopher Jacoby, and Juan Pablo Bello. 2014. A Dataset and Tax- onomy for Urban Sound Research. InProceedings of the 22nd ACM International Conference on Multimedia(Orlando, Florida, USA)(MM ’14). Association for Com- puting Machinery, New York, NY, USA, 1041–1044. doi:10.1145/2647868.2655045

    work page doi:10.1145/2647868.2655045 2014

  33. [33]

    2024.IMPROVING AUDIO SPECTROGRAM TRANSFORMERS FOR SOUND EVENT DETECTION THROUGH MULTI-STAGE TRAINING

    Florian Schmid, Paul Primus, Tobias Morocutti, Jonathan Greif, and Gerhard Wid- mer. 2024.IMPROVING AUDIO SPECTROGRAM TRANSFORMERS FOR SOUND EVENT DETECTION THROUGH MULTI-STAGE TRAINING. Technical Report. DCASE2024 Challenge

    2024

  34. [34]

    1982.Image Analysis and Mathematical Morphology

    Jean Serra. 1982.Image Analysis and Mathematical Morphology. Academic Press, London

    1982

  35. [35]

    Jaime Sevilla, Lennart Heim, Anson Ho, Tamay Besiroglu, Marius Hobbhahn, and Pablo Villalobos. 2022. Compute Trends Across Three Eras of Machine Learning. In2022 International Joint Conference on Neural Networks (IJCNN). 1–8. doi:10.1109/IJCNN55064.2022.9891914

    work page doi:10.1109/ijcnn55064.2022.9891914 2022

  36. [36]

    Plumbley

    Dan Stowell, Dimitrios Giannoulis, Emmanouil Benetos, Mathieu Lagrange, and Mark D. Plumbley. 2015. Detection and Classification of Acoustic Scenes and Events.IEEE Transactions on Multimedia17, 10 (2015), 1733–1746. doi:10.1109/ TMM.2015.2428998

    arXiv 2015

  37. [37]

    Guangzhi Tang, Kanishkan Vadivel, Yingfu Xu, Refik Bilgic, Kevin Shidqi, Paul Detterer, Stefano Traferro, Mario Konijnenburg, Manolis Sifalakis, Gert-Jan van Schaik, and Amirreza Yousefzadeh. 2023. SENECA: building a fully digital neuro- morphic processor, design trade-offs and challenges.Frontiers in Neuroscience Volume 17 (2023). doi:10.3389/fnins.2023.1187252

    work page doi:10.3389/fnins.2023.1187252 2023

  38. [38]

    M. C. W. Van Rossum. 2001. A Novel Spike Distance.Neural Comput.13, 4 (April 2001), 751–763

    2001

  39. [39]

    Satvik Venkatesh, David Moffat, and Eduardo Reck Miranda. 2022. You Only Hear Once: A YOLO-like Algorithm for Audio Segmentation and Sound Event Detection.Applied Sciences12, 7 (2022). doi:10.3390/app12073293

    work page doi:10.3390/app12073293 2022

  40. [40]

    2017.Transfer Learning Based DNN-HMM Hybrid System for Rare Sound Event Detection

    Jianfei Wang, Weiqiang Zhang, and Jia Liu. 2017.Transfer Learning Based DNN-HMM Hybrid System for Rare Sound Event Detection. Technical Report. DCASE2017 Challenge

    2017

  41. [41]

    2025.PRE-TRAINED MODEL ENHANCED ANOMALOUS SOUND DETECTION SYSTEM FOR DCASE2025 TASK2

    Lei Wang. 2025.PRE-TRAINED MODEL ENHANCED ANOMALOUS SOUND DETECTION SYSTEM FOR DCASE2025 TASK2. Technical Report. DCASE2025 Challenge

    2025

  42. [42]

    Yaoguang Wang, Yaohao Zheng, Yunxiang Zhang, Yongsheng Xie, Sen Xu, Ying Hu, and Liang He. 2021. Unsupervised Anomalous Sound Detection for Machine Condition Monitoring Using Classification-Based Methods.Applied Sciences11, 23 (2021). doi:10.3390/app112311128

    work page doi:10.3390/app112311128 2021

  43. [43]

    2023.calflops: a FLOPs and Params calculate tool for neural networks in pytorch framework

    xiaoju ye. 2023.calflops: a FLOPs and Params calculate tool for neural networks in pytorch framework. https://github.com/MrYxJ/calculate-flops.pytorch

    2023

  44. [44]

    2025.A TWO STAGE FUSION ANOMALY DETECTION APPROACH FOR TASK2

    Jie Yang. 2025.A TWO STAGE FUSION ANOMALY DETECTION APPROACH FOR TASK2. Technical Report. DCASE2025 Challenge

    2025

  45. [45]

    Yukun Yang, Wenrui Zhang, and Peng Li. 2021. Backpropagated Neighborhood Aggregation for Accurate Training of Spiking Neural Networks. InInternational Conference on Machine Learning. https://api.semanticscholar.org/CorpusID: 235826047

    2021

  46. [46]

    Zhirong Ye, Xiangdong Wang, Hong Liu, Yueliang Qian, Rui Tao, Long Yan, and Kazushige Ouchi. 2021. Sound Event Detection Transformer: An Event- based End-to-End Model for Sound Event Detection. arXiv:2110.02011 [cs.SD] https://arxiv.org/abs/2110.02011

    arXiv 2021

  47. [47]

    Wenrui Zhang and Peng Li. 2020. Temporal spike sequence learning via back- propagation for deep spiking neural networks(NIPS ’20). Curran Associates Inc., Red Hook, NY, USA, Article 1008. Received 31 March 2026; accepted 8 June 2026; revised 24 June 2026

    2020