📄 中文摘要
在动态环境中部署的神经网络面临灾难性遗忘的问题，即在适应新任务时覆盖先前学习的知识，导致早期任务的性能显著下降。提出了一种动态方法——选择性遗忘感知优化（SFAO），该方法通过余弦相似度和逐层门控调节梯度方向，实现了在保持可塑性与稳定性之间的平衡。SFAO通过可调机制有效地选择性地投影、接受或丢弃更新，采用高效的蒙特卡洛近似。实验结果表明，SFAO在标准持续学习基准上实现了具有竞争力的准确性，同时显著降低了内存成本。
通过选择性梯度投影减轻持续学习中的遗忘
📄 English Summary
Mitigating Forgetting in Continual Learning with Selective Gradient Projection
Neural networks deployed in dynamic environments encounter the challenge of catastrophic forgetting, where previously learned knowledge is overwritten when adapting to new tasks, leading to significant performance degradation on earlier tasks. This study proposes Selective Forgetting-Aware Optimization (SFAO), a dynamic method that regulates gradient directions through cosine similarity and per-layer gating, enabling controlled forgetting while balancing plasticity and stability. SFAO selectively projects, accepts, or discards updates using a tunable mechanism with efficient Monte Carlo approximation. Experiments on standard continual learning benchmarks demonstrate that SFAO achieves competitive accuracy with markedly lower memory costs.