3 ms·
fitting on noise in the training data is exactly what overfitting is. underfitting is smoothing out signal
by samrus 1y ago
fitting on noise in the training data is exactly what overfitting is. underfitting is smoothing out signal
- wongarsu 1y agoExactly. Underfitting would be if the model doesn't pick up on the fact that outro silence is labeled differently from regular silence and transcribes them the same
- xigoi 1y agoOverfitting implies a failure to properly generalize the training data. Here it generalized them correctly. Garbage in, garbage out.
- samrus 1y agoNo. Because there would have been indtances in the data where silence was labelled correctly. But the model couldnt handle the null case, so it over fit on the outros. But generally it fit on the random error in the label of the null feature. Which is what overfitting is