生成式二元记忆：基于二值化嵌入的伪回放类别增量学习
Generative Binary Memory: Pseudo-Replay Class-Incremental Learning on Binarized Embeddings
AI总结：
本文提出 GBM 伪回放方法，利用 BMM 在二元潜在空间生成伪样本，并通过专用二值化器兼容常规 DNN 与 BNN，在多个 CIL 基准上提升准确率并显著降低内存。
摘要：
在新概念不断涌现的动态环境中，深度神经网络（DNN）必须通过学习新类别同时保留先前已获得的类别来实现适应。类别增量学习（CIL）正是应对这一挑战的方法。本文提出 Generative Binary Memory（GBM，生成式二元记忆），这是一种新颖的 CIL 伪回放方法，可生成合成的二元伪样本。GBM 依赖 Bernoulli Mixture Models（BMM，伯努利混合模型），在潜在二元空间中有效建模类别分布的多模态特征。借助专门设计的特征二值化器，该方法适用于任何常规 DNN。GBM 还原生支持 Binary Neural Networks（BNN，二元神经网络），可满足嵌入式系统中高度受限的模型规模需求。实验结果表明，对于配备所提二值化器的 ResNet-18，GBM 在 CIFAR100（+2.9%）和 TinyImageNet（+1.5%）上取得了高于最先进方法的平均准确率。在 CORE50 上，GBM 也优于新兴的面向 BNN 的 CIL 方法，最终准确率提升 +3.1%，内存占用减少至 1/4.7。
英文摘要：
In dynamic environments where new concepts continuously emerge, Deep Neural Networks (DNNs) must adapt by learning new classes while retaining previously acquired ones. This challenge is addressed by Class-Incremental Learning (CIL). This paper introduces Generative Binary Memory (GBM), a novel CIL pseudo-replay approach which generates synthetic binary pseudo-exemplars. Relying on Bernoulli Mixture Models (BMMs), GBM effectively models the multi-modal characteristics of class distributions, in a latent, binary space. With a specifically-designed feature binarizer, our approach applies to any conventional DNN. GBM also natively supports Binary Neural Networks (BNNs) for highly-constrained model sizes in embedded systems. The experimental results demonstrate that GBM achieves higher than state-of-the-art average accuracy on CIFAR100 (+2.9%) and TinyImageNet (+1.5%) for a ResNet-18 equipped with our binarizer. GBM also outperforms emerging CIL methods for BNNs, with +3.1% in final accuracy and x4.7 memory reduction, on CORE50.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
