Classification on SHD
96.3AccuracyPfA SNN
Evaluation Results
| Method | Links | |
|---|---|---|
| PfA SNNRecurrent=false, Hardware=false, #Params=0.2M2026.02 | 96.3 | |
| Event-SSMRecurrent=true, Hardware=false, #Params=0.4M2026.02 | 95.9 | |
| SE-adLIFRecurrent=true, Hardware=false, #Params=450k2026.02 | 95.8 | |
| DCLS-DelaysRecurrent=false, Hardware=false, #Params=200k2026.02 | 95.1 | |
| Bullet TrainsModel Architecture=2F512HD, Exact Gradients=true, Continuous Spike Times=true, Parallelized=true, Compute Depth=O(C log K), Memory=O(N)2026.02 | 95.1 | |
| 2F256HD (Hammouamri et al.)Exact Gradients=false, Continuous Spike Times=false, Parallelized=false, Compute Depth=O(T), Memory=O(T)2026.02 | 95.07 | |
| Bullet TrainsModel Architecture=2F256HD, Exact Gradients=true, Continuous Spike Times=true, Parallelized=true, Compute Depth=O(C log K), Memory=O(N)2026.02 | 94.78 | |
| Temporal Skips with delay learningRecurrent=false, Hardware=false, #Params=1.3M2026.02 | 94.7 | |
| Recurrent SNNRecurrent=true, Hardware=false, #Params=3.9M2026.02 | 94.6 | |
| Spectro-temporal graph on SoC FPGARecurrent=false, Hardware=true, #Params=272k2026.02 | 94.5 | |
| Spectro-temporal graphRecurrent=false, Hardware=false, #Params=217k2026.02 | 94.3 | |
| Yang+Hardware=Simulated, Number of nodes=3+2026.03 | 94 | |
| 2F512HD (Mészáros et al.)Exact Gradients=true, Continuous Spike Times=false, Parallelized=false, Compute Depth=O(T), Memory=O(N)2026.02 | 93.1 | |
| DL128-SNN-DlossRecurrent=false, Hardware=false, #Params=140k2026.02 | 92.6 | |
| STSCT=152022.10 | 92.4 | |
| STSC-SNNRecurrent=false, Hardware=false, #Params=2.1M2026.02 | 92.4 | |
| S5-RFRecurrent=true, Hardware=false, #Params=214k2026.02 | 91.9 | |
| TA-SNNRecurrent=false, Hardware=false2026.02 | 91.1 | |
| TA-SNNT=152022.10 | 91.08 | |
| TA-SNNProposals=This work (SOTA)2021.07 | 91.08 | |
| Adaptive SRNNT=2502022.10 | 90.4 | |
| Adaptive SRNNRecurrent=true, Hardware=false2026.02 | 90.4 | |
| Spectro-temporal graphRecurrent=false, Hardware=false, #Params=17.1k2026.02 | 90 | |
| RELU SRNNProposals=Yin et al. 20202021.07 | 88.93 | |
| Spectro-temporal graph on SoC FPGARecurrent=false, Hardware=true, #Params=8.6k2026.02 | 88.8 | |
| Cuba-LIFRecurrent=true, Hardware=false, #Params=139k2026.02 | 87.8 | |
| Fully configurable FPGA SNNRecurrent=true, Hardware=true2026.02 | 87.8 | |
| FF SNN with dendritic delaysRecurrent=false, Hardware=false, #Params=224k2026.02 | 87.6 | |
| EventProp-GeNNRecurrent=true, Hardware=false2026.02 | 84.8 | |
| SG-based SNNProposals=Zenke et al. 20212021.07 | 84 | |
| Hetero. RSNN2022.10 | 83.5 | |
| Convolutional RSNNRecurrent=true, Hardware=false, #Params=209k2026.02 | 83.5 | |
| Recurrent SNNRecurrent=true, Hardware=false2026.02 | 83.2 | |
| LIF RSNNT=20002022.10 | 73.3 | |
| RSNN on FPGARecurrent=true, Hardware=true2026.02 | 72.9 | |
| LIF RSNNProposals=Cramer et al. 20202021.07 | 71.4 |