Table 5 Night/day class and domain train ROCAUC of logistic regression on \(z^c\) and \(z^d\) using 50 train subjects.
From: Tackling inter-subject variability in smartwatch data using factorization models
Model | Latent space | Class accuracy | Domain accuracy | ||
|---|---|---|---|---|---|
MLP | \(z^c\) | 89.71 | 8.52 | ||
FAE | \(z^c\) | 82.42 | 10.21 | ||
\(z^d\) | 74.14 | 10.93 | |||
GFAE (\(m^d=0\)) | \(z^c\) | 90.52 | 14.11 | ||
\(z^d\) | 87.70 | 16.24 | |||
GFAE (\(m^d\ne 0\)) | \(z^c\) | 98.61 | 25.62 | ||
\(z^d\) | 97.32 | 36.94 | |||
TFAE | \(z^c\) | 98.43 | 10.23 | ||
\(z^d\) | 98.21 | 28.14 | |||
