Real-world data universally confronts a severe class-imbalance problem and
exhibits a long-tailed distribution, i.e., most labels are associated with
limited instances. The na\"ive models supervised by such datasets would prefer
dominant labels, encounter a serious generalization challenge and become poorly
calibrated. We propose two novel methods from the prior perspective to
alleviate this dilemma. First, we deduce a balance-oriented data augmentation
named Uniform Mixup (UniMix) to promote mixup in long-tailed scenarios, which
adopts advanced mixing factor and sampler in favor of the minority. Second,
motivated by the Bayesian theory, we figure out the Bayes Bias (Bayias), an
inherent bias caused by the inconsistency of prior, and compensate it as a
modification on standard cross-entropy loss. We further prove that both the
proposed methods ensure the classification calibration theoretically and
empirically. Extensive experiments verify that our strategies contribute to a
better-calibrated model, and their combination achieves state-of-the-art
performance on CIFAR-LT, ImageNet-LT, and iNaturalist 2018.

提出了两种新方法，一种是平衡取向的数据增强（Uniform Mixup），另一种是基于贝叶斯理论的 Bayes Bias，并通过理论和实验证明，这两种方法能够确保分类校准和提高性能。

面向长尾视觉识别的校准模型

Towards Calibrated Model for Long-Tailed Visual Recognition from Prior  Perspective

The field of Astronomy requires the collection and assimilation of vast
volumes of data. The data handling and processing problem has become severe as
the sheer volume of data produced by scientific instruments each night grows
exponentially. This problem becomes extensive for conventional methods of
processing the data, which was mostly manual, but is the perfect setting for
the use of Machine Learning approaches. While building classifiers for
Astronomy, the cost of losing a rare object like supernovae or quasars to
detection losses is far more severe than having many false positives, given the
rarity and scientific value of these objects. In this paper, a Linear Support
Vector Machine (LSVM) is explored to detect Quasars, which are extremely bright
objects in which a supermassive black hole is surrounded by a luminous
accretion disk. In Astronomy, it is vital to correctly identify quasars, as
they are very rare in nature. Their rarity creates a class-imbalance problem
that needs to be taken into consideration. The class-imbalance problem and high
cost of misclassification are taken into account while designing the
classifier. To achieve this detection, a novel classifier is explored, and its
performance is evaluated. It was observed that LSVM along with Ensemble Bagged
Trees (EBT) achieved a 10x reduction in the False Negative Rate, using the
Learning from Mistakes methodology.

本文探讨了使用线性支持向量机技术检测类星体的方法，并结合集成装袋树技术，在考虑分类不平衡和错误成本的基础上对其性能进行了评估，取得了 False Negative Rate 的 10 倍降低。