Table 2 The area over perturbation curve of the subject-classification of ground reaction forces for different noise perturbation runs (AOPC)59.

From: Explaining the unique nature of individual gait patterns with deep learning

Model/Noise Gaussian
Random
σ = 0.5
Gaussian
Random
σ = 1.0
Gaussian
Random
σ = 2.0
Salt
Random
Pepper
Random
Salt+
Random
Shot
Random
Linear (SGD) 40.5 (6.0) 45.6 (3.0) 47.6 (1.8) 46.1 (2.7) 37.1 (7.0) 46.4 (3.1) 46.5 (2.9)
Linear (SVM) 4.0 (3.9) 18.6 (7.0) 35.4 (4.7) 31.3 (7.7) 2.5 (2.4) 32.8 (8.7) 32.4 (8.6)
MLP (3, 64) 16.2 (14.0) 27.6 (12.6) 37.8 (9.0) 43.5 (7.1) 14.4 (13.6) 43.3 (7.4) 43.3 (7.4)
MLP (3, 256) 8.9 (9.7) 20.8 (10.9) 34.9 (8.4) 41.7 (6.8) 7.7 (9.5) 41.2 (7.4) 41.1 (7.5)
MLP (3, 1024) 5.4 (7.0) 16.7 (9.4) 32.4 (7.8) 40.0 (6.6) 4.3 (6.1) 39.4 (7.2) 39.4 (7.2)
CNN-A 6.1 (6.8) 18.2 (9.4) 33.1 (8.0) 36.9 (6.6) 8.8 (8.5) 37.4 (7.5) 37.4 (7.6)
CNN-C3 12.2 (8.6) 27.5 (9.4) 38.0 (7.0) 31.0 (8.5) 11.3 (8.6) 37.9 (8.2) 38.1 (8.1)
  1. Values are reported as pairs of mean (standard deviation). Smaller values correspond to higher robustness.