本文提出了一种基于回放的类增量学习(Class-Incremental Learning, CIL)算法,通过利用类别激活图(CAM)生成掩码来压缩旧类别的示例图像,从而在有限的内存预算下保存更多的示例。关键步骤包括:1) 使用CAM生成0-1掩码以定位图像中的判别性像素;2) 通过类增量掩码(CIM)模型自适应地优化掩码生成过程;3) 利用双层优化问题...
近年来，图神经网络领域也出现了FSCIL应用。Tan等人在2022年提出了HAG-Meta，这是第一个GFSCIL（图少样本类增量学习）研究，基于前瞻性学习解决问题。Lu等人在2022年则提出了Geometer，基于类原型表示的GFSCIL方法，通过观察新类的几何邻近性、均匀性和可分性来调整基于注意力的原型。前方的挑战 尽管FSCIL取得了显著进...
TPL方法的核心思想是利用CIL(Class Incremental Learning)中的额外信息,如重放数据和已学习的任务信息,来设计一个更优化的任务ID预测方法。该方法通过以下几个关键技术实现: 似然比估计:TPL通过比较一个测试样本属于当前任务的概率与属于其他任务的概率来预测任务ID。这个比较是通过计算似然比来实现的,即将样本属于当前...
Class-incremental learning (CIL) aims to incrementally learn a unified classifier for new classes emerging, which suffers from the catastrophic forgetting problem. To alleviate forgetting and improve the recognition performance, we propose a novel CIL framework, named the topological schemas model (TSM)...
Class-incremental learning (CIL) has achieved remarkable successes in learning new classes consecutively while overcoming catastrophic forgetting on old categories. However, most existing CIL methods unreasonably assume that all old categories have the same forgetting pace, and neglect negative influence of...
Class-Incremental Learning (CIL) requires models to continually acquire knowledge of new classes without forgetting old ones. Despite Pre-trained Models (PTMs) have shown excellent performance in CIL, catastrophic forgetting still occurs as the model learns new concepts. Existing work seeks to utilize...
To maintain consistency with existing literature, we refer to this as graph Few-shot class incremental learning (GFSCIL). One of the pioneering studies in this field is the HAG-Meta method proposed by Tan et al. (2022), which incorporates the previously mentioned Prospective Learning concept. ...
1. Introduction Class-incremental learning (CIL) [40, 12, 5] has attracted appealing attentions recently by accumulating previous learned experience to learn new classes incrementally. It *Equal contributions. †The corresponding author is Prof. Yang Cong. ‡This work was...
Continual learning addresses this challenge as it aims to build models that can integrate new knowledge over time while preserving previously acquired knowledge. In the context of class- incremental learning (CIL), training a classification model is a sequential process where each step consists in...
In Class-Incremental Learning (CIL), different categories arrive in every phase, and the categories in these phases are disjoint with each other. There have been a great deal of efforts in CIL, and the currently existing methods can be broadly divided into 4 sections, i.e., rehearsal, archi...