Classification on SHD (test)
96.41AccuracySpikCommander
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SpikCommanderArchitecture=1L-8-128, Params.=760K2025.07 | 96.41 | — | |
| Sun et al.Trainable parameters=0.2M, Implementation type=Software SNN2026.05 | 96.26 | — | |
| EventSSMArchitecture=6L-64, Params.=1.6M2025.07 | 95.9 | — | |
| Schöne et al.Trainable parameters=0.4M, Implementation type=Software SNN2026.05 | 95.9 | — | |
| Baronig et al.Trainable parameters=0.45M, Implementation type=Software state space model2026.05 | 95.81 | — | |
| Hammouamri et al.#Params=200k, Features/Details=FF SNN with delay learning2025.03 | 95.1 | — | |
| Hammouamri et al.delays/conv.=true, hierarchy=true, # hid. layers=2, hidden size=256, # params=200k2024.07 | 95.1 | — | |
| Hammouamri et al.Trainable parameters=0.2M, Implementation type=Software SNN2026.05 | 95.07 | — | |
| Malettira et al.#Params=1.3M, Features/Details=Temporal Skips with delay learning2025.03 | 94.7 | — | |
| hCSNNdelays/conv.=true, hierarchy=true, data augment.=true, Test as Valid.=true, # hid. layers=4, hidden size=128, # params=345k2024.07 | 94.7 | — | |
| RadLIFResolution=14 ms, # Params=3.9M, Rec.=true, Del.=false2025.06 | 94.62 | — | |
| Bittar et al.#Params=3.9M, Features/Details=RSNN2025.03 | 94.6 | — | |
| Bittar and GarnerRec=true, Adapt=true, data augment.=true, Test as Valid.=true, # hid. layers=2, hidden size=1024, # params=1.8M2024.07 | 94.6 | — | |
| Hardware-Accelerated Event-Graph Neural Network#Params=272k, Features/Details=Spectro-temporal graph on SoC FPGA2025.03 | 94.5 | — | |
| Rafeldt et al.#Params=217k, Features/Details=Spectro-temporal graph2025.03 | 94.3 | — | |
| cAdLIFResolution=10 ms, # Params=38.7k, Rec.=false, Del.=false2025.06 | 94.19 | — | |
| hCSNNdelays/conv.=true, hierarchy=true, data augment.=true, # hid. layers=4, hidden size=128, # params=345k2024.07 | 94.1 | — | |
| DCLSArchitecture=2L-256, Params.=800K, Buff.=200, Reproduction=true2025.07 | 93.95 | — | |
| mGRADEArchitecture=2L-64, Params.=257.6K, Buff.=2002025.07 | 93.77 | — | |
| Nowotny et al.Rec=true, data augment.=true, # hid. layers=3, hidden size=256, # params=250k2024.07 | 93.5 | — | |
| Sun et al.delays/conv.=true, # params=110k2024.07 | 92.5 | — | |
| Yu et al.#Params=2.1M, Features/Details=RSNN with synaptic dynamics2025.03 | 92.4 | — | |
| Sun et al.#Params=140k, Features/Details=FF SNN with delay learning2025.03 | 92.4 | — | |
| CNNImplementation type=Software2026.05 | 92.4 | — | |
| hCSNNdelays/conv.=true, hierarchy=true, # hid. layers=4, hidden size=128, # params=345k2024.07 | 92.2 | — | |
| S5-RFResolution=4 ms, # Params=0.2M, Rec.=false, Del.=false2025.06 | 91.86 | — | |
| Yao et al.Att=true, # hid. layers=2, hidden size=1282024.07 | 91.1 | — | |
| D’Agostino et al.delays/conv.=true, # params=224k2024.07 | 90.7 | — | |
| Rafeldt et al.#Params=17.1, Features/Details=Spectro-temporal graph2025.03 | 90 | — | |
| ExpectationArchitecture=adLIF, Source temporal resolution (bS)=2, Target temporal resolution (bT)=12024.11 | 89.5 | — | |
| IntegralArchitecture=adLIF, Source temporal resolution (bS)=2, Target temporal resolution (bT)=12024.11 | 89.5 | — | |
| Deckers et al.Trainable parameters=256k, Implementation type=Software LSM2026.05 | 89.3 | — | |
| LSTMImplementation type=Software2026.05 | 89 | — | |
| Hardware-Accelerated Event-Graph Neural Network#Params=8.6k, Features/Details=Spectro-temporal graph on SoC FPGA2025.03 | 88.8 | — | |
| hSNNhierarchy=true, # hid. layers=4, hidden size=128, # params=150k2024.07 | 88.1 | — | |
| Dampfhoffer et al.#Params=139k, Features/Details=RSNN2025.03 | 87.8 | — | |
| Matinizadeh et al.Features/Details=Fully configurable FPGA SNN (QUANTISENC tool)2025.03 | 87.8 | — | |
| Matinizadeh et al.Implementation type=FPGA-based SNN2026.05 | 87.8 | — | |
| ExpectationArchitecture=RadLIF, Source temporal resolution (bS)=2, Target temporal resolution (bT)=12024.11 | 87.7 | — | |
| IntegralArchitecture=RadLIF, Source temporal resolution (bS)=2, Target temporal resolution (bT)=12024.11 | 87.7 | — | |
| D'Agostino et al.#Params=224k, Features/Details=FF SNN with dendritic delays2025.03 | 87.6 | — | |
| D’Agostino et al.Trainable parameters=224k, Implementation type=Resistive RAM-based analog SNN2026.05 | 87.5 | — | |
| ExpectationArchitecture=adLIF, Source temporal resolution (bS)=3, Target temporal resolution (bT)=12024.11 | 87.4 | — | |
| IntegralArchitecture=adLIF, Source temporal resolution (bS)=3, Target temporal resolution (bT)=12024.11 | 87.4 | — | |
| hSNNhierarchy=true, # hid. layers=4, hidden size=32, # params=25k2024.07 | 84.9 | — | |
| EulerArchitecture=adLIF, Source temporal resolution (bS)=2, Target temporal resolution (bT)=12024.11 | 84.6 | — | |
| FPGA-based quasi-analog LSM (Ours)Trainable parameters=86.26 k, Implementation type=FPGA-based quasi-analog LSM2026.05 | 84.5 | — | |
| ExpectationArchitecture=adLIF, Source temporal resolution (bS)=4, Target temporal resolution (bT)=12024.11 | 84.3 | — | |
| IntegralArchitecture=adLIF, Source temporal resolution (bS)=4, Target temporal resolution (bT)=12024.11 | 84.3 | — | |
| Rossbroich et al.#Params=209k, Features/Details=Convolutional RSNN2025.03 | 83.5 | — | |
| Cramer et al.Features/Details=Recurrent SNN (RSNN)2025.03 | 83.2 | — | |
| Cramer et al.Implementation type=Software SNN2026.05 | 83.2 | — | |
| KaimingNumber of hidden layers=32022.06 | 83.1 | — | |
| Fluctuation-driven initializationNumber of hidden layers=3, target fluctuation parameter (sigma_U)=12022.06 | 82.7 | — | |
| Perez-Nieves et al.Rec=true, # hid. layers=1, hidden size=256, # params=100k2024.07 | 82.7 | — | |
| ExpectationArchitecture=RadLIF, Source temporal resolution (bS)=3, Target temporal resolution (bT)=12024.11 | 81.6 | — | |
| IntegralArchitecture=RadLIF, Source temporal resolution (bS)=3, Target temporal resolution (bT)=12024.11 | 81.6 | — | |
| Fluctuation-driven initializationNumber of hidden layers=7, target fluctuation parameter (sigma_U)=12022.06 | 80.9 | — | |
| EulerArchitecture=adLIF, Source temporal resolution (bS)=3, Target temporal resolution (bT)=12024.11 | 77.8 | — | |
| Biswas et al.Trainable parameters=30k, Implementation type=Software LSM2026.05 | 77.8 | — | |
| Kaiming & Homeostatic PlasticityNumber of hidden layers=72022.06 | 77 | — | |
| EulerArchitecture=RadLIF, Source temporal resolution (bS)=2, Target temporal resolution (bT)=12024.11 | 76.8 | — | |
| Carpegna et al.Implementation type=FPGA-based SNN2026.05 | 72.99 | — | |
| Carpegna et al.Features/Details=RSNN on FPGA (Spiker+ framework)2025.03 | 72.9 | — | |
| ExpectationArchitecture=RadLIF, Source temporal resolution (bS)=4, Target temporal resolution (bT)=12024.11 | 72.8 | — | |
| IntegralArchitecture=RadLIF, Source temporal resolution (bS)=4, Target temporal resolution (bT)=12024.11 | 72.8 | — | |
| EulerArchitecture=adLIF, Source temporal resolution (bS)=4, Target temporal resolution (bT)=12024.11 | 71.3 | — | |
| ExpectationArchitecture=adLIF, Source temporal resolution (bS)=10, Target temporal resolution (bT)=12024.11 | 70.4 | — | |
| IntegralArchitecture=adLIF, Source temporal resolution (bS)=10, Target temporal resolution (bT)=12024.11 | 70.4 | — | |
| NoneArchitecture=adLIF, Source temporal resolution (bS)=2, Target temporal resolution (bT)=12024.11 | 58.1 | — | |
| EulerArchitecture=RadLIF, Source temporal resolution (bS)=3, Target temporal resolution (bT)=12024.11 | 57.5 | — | |
| Linear SVMImplementation type=Software2026.05 | 56 | — | |
| Time-const.Architecture=adLIF, Source temporal resolution (bS)=2, Target temporal resolution (bT)=12024.11 | 53 | — | |
| EulerArchitecture=RadLIF, Source temporal resolution (bS)=4, Target temporal resolution (bT)=12024.11 | 50.5 | — | |
| EulerArchitecture=adLIF, Source temporal resolution (bS)=10, Target temporal resolution (bT)=12024.11 | 49.7 | — | |
| ExpectationArchitecture=RadLIF, Source temporal resolution (bS)=10, Target temporal resolution (bT)=12024.11 | 45.4 | — | |
| IntegralArchitecture=RadLIF, Source temporal resolution (bS)=10, Target temporal resolution (bT)=12024.11 | 45.4 | — | |
| EulerArchitecture=RadLIF, Source temporal resolution (bS)=10, Target temporal resolution (bT)=12024.11 | 45.3 | — | |
| NoneArchitecture=RadLIF, Source temporal resolution (bS)=2, Target temporal resolution (bT)=12024.11 | 31.8 | — | |
| NoneArchitecture=adLIF, Source temporal resolution (bS)=3, Target temporal resolution (bT)=12024.11 | 30.3 | — | |
| NoneArchitecture=adLIF, Source temporal resolution (bS)=4, Target temporal resolution (bT)=12024.11 | 20.5 | — | |
| Time-const.Architecture=adLIF, Source temporal resolution (bS)=3, Target temporal resolution (bT)=12024.11 | 19.9 | — | |
| Time-const.Architecture=adLIF, Source temporal resolution (bS)=4, Target temporal resolution (bT)=12024.11 | 14.3 | — | |
| NoneArchitecture=adLIF, Source temporal resolution (bS)=10, Target temporal resolution (bT)=12024.11 | 9.6 | — | |
| Time-const.Architecture=RadLIF, Source temporal resolution (bS)=2, Target temporal resolution (bT)=12024.11 | 7.5 | — | |
| NoneArchitecture=RadLIF, Source temporal resolution (bS)=10, Target temporal resolution (bT)=12024.11 | 6.7 | — | |
| NoneArchitecture=RadLIF, Source temporal resolution (bS)=4, Target temporal resolution (bT)=12024.11 | 6.1 | — | |
| Time-const.Architecture=RadLIF, Source temporal resolution (bS)=3, Target temporal resolution (bT)=12024.11 | 5.6 | — | |
| NoneArchitecture=RadLIF, Source temporal resolution (bS)=3, Target temporal resolution (bT)=12024.11 | 5.3 | — | |
| Time-const.Architecture=RadLIF, Source temporal resolution (bS)=4, Target temporal resolution (bT)=12024.11 | 5 | — | |
| Time-const.Architecture=adLIF, Source temporal resolution (bS)=10, Target temporal resolution (bT)=12024.11 | 5 | — | |
| Time-const.Architecture=RadLIF, Source temporal resolution (bS)=10, Target temporal resolution (bT)=12024.11 | 5 | — | |
| KaimingNumber of hidden layers=72022.06 | 4.5 | — | |
| ConvSynSize of Hidden Layer=400-4002022.04 | — | 33 | |
| MAP-SNNSize of Hidden Layer=400-4002022.04 | — | 13 | |
| MSPSize of Hidden Layer=400-4002022.04 | — | 17.1 | |
| R-SNN (Cramer)2022.04 | — | 16.8 | |
| R-SNN (Zenke)2022.04 | — | 18 | |
| SNN-base2022.04 | — | 28.6 | |
| Spiking-MLP2022.04 | — | 52.5 |