Table 2 Evaluation of a spiking neural network with probabilistic metaplasticity and TACOS on the split-CIFAR-10 task in the domain-IL scenario. Probabilistic metaplasticity is evaluated considering multi-memristor weights \(n_{mem}\) = 7 and TACOS considers full-precision weights. The table lists the mean and standard deviation of the individual task accuracies and the mean accuracy across tasks over 5 runs after sequential training.
From: Probabilistic metaplasticity for continual learning with memristors in spiking networks
Task | Baseline | Probabilistic metaplasticity | TACOS11 |
|---|---|---|---|
Class 0,1 | 86.64 ± 1.86 | 93.52 ± 1.36 | 90.84 ± 0.36 |
Class 2,3 | 59.81 ± 3.69 | 73.11 ± 4.80 | 71.27 ± 1.08 |
Class 4,5 | 63.93 ± 3.08 | 80.87 ± 4.07 | 80.01 ± 0.93 |
Class 6,7 | 82.03 ± 5.75 | 84.90 ± 1.61 | 88.83 ± 0.57 |
Class 8,9 | 96.96 ± 0.80 | 94.72 ± 0.21 | 97.64 ± 0.23 |
Mean | 77.87 ± 2.77 | 85.42 ± 2.11 | 85.72 ± 0.55 |
