Table 6 Partially quantized model performance on the test set. MCC: Matthews Correlation Coefficient; Acc: accuracy; \(F_1\): \(F_1\)-score; \(\mu\) mean of the metric over \(10^{5}\) bootstrap resampling; \(\rho\) radius of the corresponding 5%-95% CI
From: Endoscopy-based IBD identification by a quantized deep learning pipeline
Method | MCC | Acc | \(F_1\) | Unpredicted | |||
|---|---|---|---|---|---|---|---|
\(\mu\) | \(\rho\) | \(\mu\) | \(\rho\) | \(\mu\) | \(\rho\) | ||
PTQ | 0.82 | \(6\cdot 10^{-4}\) | 0.91 | \(3\cdot 10^{-4}\) | 0.91 | \(4\cdot 10^{-4}\) | 2 |
PACT | 0.86 | \(6\cdot 10^{-4}\) | 0.93 | \(3\cdot 10^{-4}\) | 0.93 | \(3\cdot 10^{-4}\) | 2 |
QPWL | 0.87 | \(5\cdot 10^{-4}\) | 0.94 | \(3\cdot 10^{-4}\) | 0.93 | \(3\cdot 10^{-4}\) | 1 |
QAT | 0.82 | \(6\cdot 10^{-4}\) | 0.91 | \(3\cdot 10^{-4}\) | 0.91 | \(3\cdot 10^{-4}\) | 1 |