Angular Gap: Reducing the Uncertainty of Image Difficulty through Model Calibration
This work addresses the challenge of curriculum learning for researchers by improving difficulty estimation, though it is incremental as it builds on existing hyperspherical learning and calibration techniques.
The paper tackles the problem of unreliable image difficulty estimation in curriculum learning by proposing Angular Gap, a difficulty measure based on angular distances in hyperspherical learning, and shows it outperforms recent metrics on CIFAR10-H and ImageNetV2. It further applies this to unsupervised domain adaptation, combining it with Cycle Self Training to achieve state-of-the-art results on Office31 and VisDA 2017 datasets.
Curriculum learning needs example difficulty to proceed from easy to hard. However, the credibility of image difficulty is rarely investigated, which can seriously affect the effectiveness of curricula. In this work, we propose Angular Gap, a measure of difficulty based on the difference in angular distance between feature embeddings and class-weight embeddings built by hyperspherical learning. To ascertain difficulty estimation, we introduce class-wise model calibration, as a post-training technique, to the learnt hyperbolic space. This bridges the gap between probabilistic model calibration and angular distance estimation of hyperspherical learning. We show the superiority of our calibrated Angular Gap over recent difficulty metrics on CIFAR10-H and ImageNetV2. We further propose Angular Gap based curriculum learning for unsupervised domain adaptation that can translate from learning easy samples to mining hard samples. We combine this curriculum with a state-of-the-art self-training method, Cycle Self Training (CST). The proposed Curricular CST learns robust representations and outperforms recent baselines on Office31 and VisDA 2017.