基于 t-vFM 相似度实现持续学习的鲁棒特征学习
Towards Robust Feature Learning with t-vFM Similarity for Continual Learning
AI总结:
针对持续学习中数据不平衡导致表征学习受限的问题,本文在监督对比损失中以 t-vFM 相似度替代余弦相似度,以增强特征鲁棒性,并在 Seq-CIFAR-10 上取得优于近期基线的结果。
摘要:
持续学习已从特征学习的角度,利用标准监督对比损失得到发展。由于训练过程中存在数据不平衡,学习更优表征仍面临挑战。在本工作中,我们建议在监督对比损失中使用一种不同的相似度度量来替代余弦相似度,以学习更鲁棒的表征。我们在图像分类数据集之一 Seq-CIFAR-10 上验证了该方法,结果优于近期的持续学习基线。
英文摘要:
Continual learning has been developed using standard supervised contrastive loss from the perspective of feature learning. Due to the data imbalance during the training, there are still challenges in learning better representations. In this work, we suggest using a different similarity metric instead of cosine similarity in supervised contrastive loss in order to learn more robust representations. We validate the our method on one of the image classification datasets Seq-CIFAR-10 and the results outperform recent continual learning baselines.
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。