Table 1 Comparison of accuracy and uncertainty prediction performances
Conventional ANN (float32) | Bayesian (float32) | Bayesian Hardware (filamentary memristor experimental) | Bayesian Hardware (filamentary memristor simulation) | Bayesian Hardware (phase-change memory simulation) | |
|---|---|---|---|---|---|
Accuracy classification | best: 81% mean: 80% | best: 80% mean: 79% | 75% | best: 76% mean: 76% | best:73% mean: 73% |
Prediction confidence (aleatoric) [AUC] | best: 0.90 mean: 0.79 | best: 0.92 mean: 0.90 | 0.91 | best:0.91 mean: 0.89 | best:0.85 mean: 0.87 |
Anomaly detection (epistemic) [AUC] | 0.5 | best: 1 mean: 0.95 | 0.99 | best: 0.96 mean: 0.92 | best: 0.96 mean: 0.82 |
