Research map / Architectures and efficiency / Spiking networks
Training deep spiking networks
Trains spiking neural networks with implicit differentiation, distillation, calibration, and energy-efficient search.
141 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2018–2026) · an approach in Spiking networks, keyword cluster “snn · energy · ann”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 2 of 6 tracked benchmarks.
35 papers in 2023–24 and 66 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 47.9% to 42.6%.
Open questions the data points to
- ImageNet (accuracy): best went from 79.37 (up to 2024) to 85.83 in 2025 — Training High Performance Spiking Neural Network by Temporal Model Calibration (ICML 2025)
- DVS-CIFAR10 (accuracy): best went from 85.23 (up to 2024) to 87.63 in 2025 — Training High Performance Spiking Neural Network by Temporal Model Calibration (ICML 2025)
What looks settled
- CIFAR10 (accuracy): best up to 2024 98.7, best in 2025–26 96.84 from 6 papers, at most 0.2% better, if at all.
- CIFAR-100 (accuracy): best up to 2024 89.7, best in 2025–26 81.98 from 8 papers, at most 0.2% better, if at all.
- CIFAR-100 (top-1 accuracy): best up to 2024 82.43, best in 2025–26 81.9 from 4 papers, at most 0.2% better, if at all.
- ImageNet (top-1 accuracy): best up to 2024 83.82, best in 2025–26 71.76 from 3 papers, at most 0.2% better, if at all.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2018 · SLAYER: Spike Layer Error Reassignment in Time (NeurIPS 2018 · 467 citations)
- 2019 · Direct Training for Spiking Neural Networks: Faster, Larger, Better (AAAI 2019 · 653 citations)
- 2020 · RMP-SNN: Residual Membrane Potential Neuron for Enabling Deeper High-Accuracy and Low-Latency Spiking Neural Network (CVPR 2020 · 321 citations)
- 2021 · Going Deeper With Directly-Trained Larger Spiking Neural Networks (AAAI 2021 · 504 citations)
- 2022 · Training High-Performance Low-Latency Spiking Neural Networks by Differentiation on Spike Representation (CVPR 2022 · 120 citations)
- 2023 · Constructing Deep Spiking Neural Networks From Artificial Neural Networks With Knowledge Distillation (CVPR 2023 · 87 citations)
- 2024 · Gated Attention Coding for Training High-Performance and Efficient Spiking Neural Networks (AAAI 2024 · 44 citations)
- 2025 · FSTA-SNN:Frequency-Based Spatial-Temporal Attention Module for Spiking Neural Networks (AAAI 2025 · 14 citations)
- 2026 · Spikingformer: A Key Foundation Model for Spiking Neural Networks (AAAI 2026 · 7 citations)
Earliest papers here
- SLAYER: Spike Layer Error Reassignment in Time (NeurIPS 2018 · 467 citations)
- A Plasticity-Centric Approach to Train the Non-Differential Spiking Neural Networks (AAAI 2018 · 63 citations)
- Hybrid Macro/Micro Level Backpropagation for Training Deep Spiking Neural Networks (NeurIPS 2018 · 42 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
CIFAR10 · accuracy (higher is better)
29 papers, 2019–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 90.53 | Direct Training for Spiking Neural Networks: Faster, Larger, Better (AAAI 2019 · 653 citations) |
| 2020 | 91.41 | Temporal Spike Sequence Learning via Backpropagation for Deep Spiking Neural Networks (NeurIPS 2020 · 116 citations) |
| 2021 | 93.44 | Near Lossless Transfer Learning for Spiking Neural Networks (AAAI 2021 · 59 citations) |
| 2022 | 95.5 | Differentiable hierarchical and surrogate gradient search for spiking neural networks (NeurIPS 2022 · 7 citations) |
| 2024 | 98.7 | SpikeZIP-TF: Conversion is All You Need for Transformer-based SNN (ICML 2024 · 2 citations) |
CIFAR-100 · accuracy (higher is better)
27 papers, 2021–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 71.52 | Near Lossless Transfer Learning for Spiking Neural Networks (AAAI 2021 · 59 citations) |
| 2022 | 78.76 | Temporal Effective Batch Normalization in Spiking Neural Networks (NeurIPS 2022 · 10 citations) |
| 2023 | 81.7 | Surrogate Module Learning: Reduce the Gradient Error Accumulation in Training Spiking Neural Networks (ICML 2023) |
| 2024 | 89.7 | SpikeZIP-TF: Conversion is All You Need for Transformer-based SNN (ICML 2024 · 2 citations) |
ImageNet · accuracy (higher is better)
14 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 69.26 | Deep Residual Learning in Spiking Neural Networks (NeurIPS 2021 · 220 citations) |
| 2023 | 69.35 | Surrogate Module Learning: Reduce the Gradient Error Accumulation in Training Spiking Neural Networks (ICML 2023) |
| 2024 | 79.37 | Are Conventional SNNs Really Efficient? A Perspective from Network Quantization (CVPR 2024 · 10 citations) |
| 2025 | 85.83 | Training High Performance Spiking Neural Network by Temporal Model Calibration (ICML 2025) |
DVS-CIFAR10 · accuracy (higher is better)
13 papers, 2019–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2019 | 60.5 | Direct Training for Spiking Neural Networks: Faster, Larger, Better (AAAI 2019 · 653 citations) |
| 2022 | 84.9 | Temporal Effective Batch Normalization in Spiking Neural Networks (NeurIPS 2022 · 10 citations) |
| 2023 | 85.23 | Surrogate Module Learning: Reduce the Gradient Error Accumulation in Training Spiking Neural Networks (ICML 2023) |
| 2025 | 87.63 | Training High Performance Spiking Neural Network by Temporal Model Calibration (ICML 2025) |
CIFAR-100 · top-1 accuracy (higher is better)
11 papers, 2020–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 70.93 | RMP-SNN: Residual Membrane Potential Neuron for Enabling Deeper High-Accuracy and Low-Latency Spiking Neural Network (CVPR 2020 · 321 citations) |
| 2022 | 71.22 | SpikeConverter: An Efficient Conversion Framework Zipping the Gap between Artificial Neural Networks and Spiking Neural Networks (AAAI 2022 · 49 citations) |
| 2023 | 75.42 | Reducing ANN-SNN Conversion Error through Residual Membrane Potential (AAAI 2023 · 54 citations) |
| 2024 | 82.43 | EnOF-SNN: Training Accurate Spiking Neural Networks via Enhancing the Output Feature (NeurIPS 2024) |
ImageNet · top-1 accuracy (higher is better)
11 papers, 2020–2025 · flat since 2024
| Year | Best so far | Set by |
|---|---|---|
| 2020 | 73.09 | RMP-SNN: Residual Membrane Potential Neuron for Enabling Deeper High-Accuracy and Low-Latency Spiking Neural Network (CVPR 2020 · 321 citations) |
| 2024 | 83.82 | SpikeZIP-TF: Conversion is All You Need for Transformer-based SNN (ICML 2024 · 2 citations) |
What has been tried
94 papers here report 956 measured changes (a component added, removed or swapped, compared with a baseline): helped 71%, no clear effect 7%, hurt 19%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Increase spatial-channel attention kernel size | CIFAR100 · accuracy | Helped | Gated Attention Coding for Training High-Performance and Efficient Spiking Neural Networks (AAAI 2024 · 44 citations) | |
| replace binary spikes with ternary spikes | ImageNet, 4 timesteps · accuracy | 61.07\% -> 66.90\%; 63.82\% -> 70.12\% | Helped | Ternary Spike: Learning Ternary Spikes for Spiking Neural Networks (AAAI 2024 · 40 citations) |
| Replace learned DCLS delays with fixed random delays | SHD (fully connected) · Test accuracy | No clear effect | Learning Delays in Spiking Neural Networks using Dilated Convolutions with Learnable Spacings (ICLR 2024 · 18 citations) | |
| Use serial rather than parallel submodule combination | CIFAR-100, T=2 · accuracy | +1.32% | Helped | FSTA-SNN:Frequency-Based Spatial-Temporal Attention Module for Spiking Neural Networks (AAAI 2025 · 14 citations) |
| Increase network depth in the reversible model | Static and neuromorphic datasets · memory per image | up to 8.1\times; RevSFormer-16-384 saves 9.1\times | Helped | Memory-Efficient Reversible Spiking Neural Networks (AAAI 2024 · 13 citations) |
| Add distillation | CIFAR-100 (timestep=2) · accuracy | 68.27% -> 68.74% | Helped | Enhancing Representation of Spiking Neural Networks via Similarity-Sensitive Contrastive Learning (AAAI 2024 · 11 citations) |
| Allocate more bits to spike patterns and fewer to time steps for static-image SNNs | ImageNet, CIFAR10, CIFAR100 · accuracy | 96.50 vs 95.94; 80.71 vs 78.77 | Helped | Are Conventional SNNs Really Efficient? A Perspective from Network Quantization (CVPR 2024 · 10 citations) |
| Use scale 4 rather than no SMA | DVS128 Gesture · accuracy | 96.52 -> 97.91(±0.34) | Helped | Advancing Spiking Neural Networks Towards Multiscale Spatiotemporal Interaction Learning (AAAI 2025 · 6 citations) |
| add temporal separation to the CE loss | CIFAR100 · accuracy | 77.15% -> 77.41% | Helped | Temporal Separation with Entropy Regularization for Knowledge Distillation in Spiking Neural Networks (CVPR 2025 · 6 citations) |
| Apply LTS to SEW-ResNet50 | ImageNet1k · accuracy | 67.78 -> 71.24 | Helped | Rethinking the Membrane Dynamics and Optimization Objectives of Spiking Neural Networks (NeurIPS 2024 · 5 citations) |
| Use SR for the encoder instead of SA | nuScenes perception · Avg. IoU | 16.28 -> 34.06 | Helped | Autonomous Driving with Spiking Neural Networks (NeurIPS 2024 · 5 citations) |
| Use channel-wise threshold balancing instead of layer-wise threshold balancing | ImageNet · accuracy | Helped | Inference-Scale Complexity in ANN-SNN Conversion for High-Performance and Low-Power Applications (CVPR 2025 · 4 citations) |
What reviewers will expect
Datasets and baselines named in the 36 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- CIFAR-10 · 21 of 36 papers
- CIFAR-100 · 19 of 36 papers
- ImageNet · 14 of 36 papers
- CIFAR10-DVS · 11 of 36 papers
- ImageNet-1K · 7 of 36 papers
- DVS-Gesture · 7 of 36 papers
- DVS-CIFAR10 · 6 of 36 papers
- N-Caltech101 · 4 of 36 papers
Baselines
- TET · 10 of 36 papers
- SEW-ResNet · 8 of 36 papers
- GLIF · 8 of 36 papers
- Dspike · 7 of 36 papers
- Spikformer · 6 of 36 papers
- QCFS · 6 of 36 papers
- DSR · 4 of 36 papers
- KDSNN · 4 of 36 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 101 papers; from OpenAlex and the papers' affiliations).
- Zhejiang University · 13 papers
- Peking University · 13 papers
- University of Electronic Science and Technology of China · 7 papers
- Mohamed bin Zayed University of Artificial Intelligence · 6 papers
- Technology Innovation Institute · 5 papers
- Intelligent Science & Technology Academy of CASIC · 5 papers
- Dalian University of Technology · 4 papers
- Harbin Institute of Technology · 4 papers
- Chinese Academy of Sciences · 4 papers
- China Aerospace Science and Industry Corporation Limited · 4 papers
Use it
Download BibTeX of 40 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Spiking transformers (87 papers)
- Event-camera spiking networks (13 papers)
Elsewhere in the topic
- Sparse efficient spiking sequence models (26 papers)
- Bio-inspired spiking neuron models (22 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).