Experiments Results
Accuracy · Accuracy–Energy Trade-off · Spike Counts
Results — Accuracy
Top-1 Accuracy (%) Hybrid TCN LSTM SpikingTCN SNN-only
0 20 40 60 80 100 87.78 Hybrid 85.66 TCN 71.23 LSTM 76.61 SpikingTCN 62.84 SNN

Key takeaway: The Hybrid TCN–SNN achieves the highest accuracy (87.8%), edging out TCN (≈85%) and SpikingTCN (≈76.6%).

Results — Trade-off
Accuracy vs Estimated Energy Hybrid TCN LSTM SpikingTCN SNN-only
Estimated energy → 0 20 40 60 80 100 SNN (61.7) SpikingTCN (76.6) Hybrid (87.8) TCN (85.0) LSTM (71.23) Best balanced-model Best for Neuroprosthetic Systems

Insight: SNN is lowest energy but accuracy-limited; TCN/LSTM are accurate but costlier; SpikingTCN achieves a balanced trade-off between accuracy and efficiency in neuroprosthetic settings; Hybrid sits near the accuracy frontier at low energy.

Spike counts (per sample) Hybrid SpikingTCN SNN-only
Model Encoding Total Spikes Mean Firing Rate (%)
SNN-onlyRate69513.57
HybridRate2805.47
SpikingTCNRate66512.98
SNN-onlyLatency96318.81
HybridLatency1082.11
SpikingTCNLatency94518.46
SNN-onlyDelta112221.91
HybridDelta801.56
SpikingTCNDelta137926.93

Spike stats: Hybrid shows the lowest spike activity (e.g., 40 @ delta), indicating strong efficiency potential.

Prev Next