What is Label Quality Model?
Label Quality Model is a technique used to predict clean labels from noisy labels. This technique relies on the presence of rater features and a subset of training data with both clean and noisy labels, which is known as a paired-subset.
In real-life scenarios, it is sometimes difficult to avoid some level of label noise. LQM works as long as the clean label is less noisy than a randomly selected label from the pool. Clean labels can come from expert raters or from the aggregation of multiple raters.
How does Label Quality Model work?
The LQM technique requires a paired-subset which consists of clean and noisy labels. Rater features are also required, which are used as input during training. The model training takes place on the paired-subset with rater features and noisy labels as the input.
The model is then applied to the entire training corpus which results in a more accurate alternative to the noisy labels. Label Quality Model is an intermediate supervised task that predicts clean labels from noisy ones.
Why is Label Quality Model important?
Label Quality Model is important because it helps in cases where it is difficult to obtain clean labels for model training. The presence of noisy labels can adversely affect the model performance. By using the output of the LQM technique, which is more accurate than the noisy labels, the model can be trained with better quality data resulting in improved performance.
Furthermore, LQM can help in reducing the cost of training by reducing the amount of expert labeling required. By using the output of LQM, which is produced using a subset of already-labelled data, the need for expert labeling is reduced.
Label Quality Model is a useful tool for improving model performance by predicting clean labels from noisy ones. It is an intermediate supervised task that requires a paired-subset of clean and noisy labels and rater features. By using the output of LQM, the model can be trained with better quality data resulting in improved performance.
Furthermore, LQM can help reduce the cost of training by reducing the amount of expert labeling required. This is especially important in real-life scenarios where obtaining clean labels can be a challenging task.
