Noise Robustness
Hybrid image-and-gate noise improved performance beyond the baseline in validation accuracy, patient accuracy, and AUC. Gaussian perturbation was used as a simpler approximation because built-in Cirq noise channels caused long simulations…
1 sources - 5 claims
Hybrid image-and-gate noise improved performance beyond the baseline in validation accuracy, patient accuracy, and AUC. Gaussian perturbation was used as a simpler approximation because built-in Cirq noise channels caused long simulations and sometimes kernel failures. The article interprets the hybrid noise result as stochastic regularization similar to dropout. Gate-level noise reduced patient accuracy to 0.51 and AUC to 0.56. Image-level Gaussian noise on T1Gd moderately reduced patient accuracy from 0.67 to 0.61 while AUC was 0.67.