Table 2 Baseline results for the keyword few-shot class-incremental learning task

From: The neurobench framework for benchmarking neuromorphic computing algorithms and systems

Baseline

Accuracy (Base / Session Avg)

Footprint (bytes)

Model Exec. Rate (Hz)

Connection Sparsity

Activation Sparsity

SynOps (per model exec.)

      

Dense

Eff_MACs

Eff_ACs

M5 ANN

(97.09% / 89.27%)

6.03 × 106

1

0.0

0.783

2.59 × 107

7.85 × 106

0

SNN

(93.48% / 75.27%)

1.36 × 107

200

0.0

0.916

3.39 × 106

0

3.65 × 105

  1. Base accuracy refers to accuracy on the 100 base classes after pre-training while session average accuracy is the average accuracy over all sessions for the corresponding prototypical baseline. The detailed accuracy per session for the different baselines are shown in Fig. 3.
Back to article page