Confidence-Aware Calibration and Scoring Functions for Curriculum Learning
This work addresses model calibration and generalization issues in deep learning, particularly for classification tasks, but it is incremental as it builds on existing label smoothing and curriculum learning methods.
The paper tackled the problem of over-confident and miscalibrated deep neural networks by integrating model and human confidence into label smoothing and curriculum learning, resulting in improved model performance and calibration across image and text classification tasks with multi-rater datasets.
Despite the great success of state-of-the-art deep neural networks, several studies have reported models to be over-confident in predictions, indicating miscalibration. Label Smoothing has been proposed as a solution to the over-confidence problem and works by softening hard targets during training, typically by distributing part of the probability mass from a `one-hot' label uniformly to all other labels. However, neither model nor human confidence in a label are likely to be uniformly distributed in this manner, with some labels more likely to be confused than others. In this paper we integrate notions of model confidence and human confidence with label smoothing, respectively \textit{Model Confidence LS} and \textit{Human Confidence LS}, to achieve better model calibration and generalization. To enhance model generalization, we show how our model and human confidence scores can be successfully applied to curriculum learning, a training strategy inspired by learning of `easier to harder' tasks. A higher model or human confidence score indicates a more recognisable and therefore easier sample, and can therefore be used as a scoring function to rank samples in curriculum learning. We evaluate our proposed methods with four state-of-the-art architectures for image and text classification task, using datasets with multi-rater label annotations by humans. We report that integrating model or human confidence information in label smoothing and curriculum learning improves both model performance and model calibration. The code are available at \url{https://github.com/AoShuang92/Confidence_Calibration_CL}.