论文标题
长尾班级学习
Long-Tailed Class Incremental Learning
论文作者
论文摘要
在课堂增量学习（CIL）中，模型必须依次学习新课程，而不会忘记旧课程。但是，传统的CIL方法考虑每个新任务的平衡分布，这忽略了现实世界中长尾分布的流行。在这项工作中，我们提出了两个长尾的CIL场景，我们订购了LT-CIL。订购的LT-CIL考虑了以下情况，我们从收集的校课中学到的样本比较少的尾巴课程更多。另一方面，洗牌的LT-cil假定每个任务都完全随机长尾分布。我们在LT-CIL方案中系统地评估了现有方法，并且与传统的CIL方案相比，行为非常不同。此外，我们提出了一个两阶段的学习基线，具有可学习的重量缩放层，以减少由LT-CIL中长尾巴分布引起的偏见，这反过来又改善了由于有限的示例而导致的常规CIL的性能。我们的结果表明，我们在CIFAR-100和Imagenet-Subset上的方法的出色性能（平均增量精度最高为6.44点）。该代码可在https://github.com/xialeiliu/long-tailed-cil上找到
In class incremental learning (CIL) a model must learn new classes in a sequential manner without forgetting old ones. However, conventional CIL methods consider a balanced distribution for each new task, which ignores the prevalence of long-tailed distributions in the real world. In this work we propose two long-tailed CIL scenarios, which we term ordered and shuffled LT-CIL. Ordered LT-CIL considers the scenario where we learn from head classes collected with more samples than tail classes which have few. Shuffled LT-CIL, on the other hand, assumes a completely random long-tailed distribution for each task. We systematically evaluate existing methods in both LT-CIL scenarios and demonstrate very different behaviors compared to conventional CIL scenarios. Additionally, we propose a two-stage learning baseline with a learnable weight scaling layer for reducing the bias caused by long-tailed distribution in LT-CIL and which in turn also improves the performance of conventional CIL due to the limited exemplars. Our results demonstrate the superior performance (up to 6.44 points in average incremental accuracy) of our approach on CIFAR-100 and ImageNet-Subset. The code is available at https://github.com/xialeiliu/Long-Tailed-CIL
