arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2204.00616cs.LGcs.CV
收藏

自监督学习和下游分类中的单纯形嵌入

Simplicial Embeddings in Self-Supervised Learning and Downstream Classification

Mila(米拉人工智能研究所);Université de Montréal(蒙特利尔大学);National University of Singapore(新加坡国立大学)

Samuel Lavoie, Christos Tsirigotis, Max Schwarzer, Ankit Vani, Michael Noukhovitch, Kenji Kawaguchi, Aaron Courville

摘要:

单纯形嵌入(SEM)是通过自监督学习(SSL)学习的表示,其中表示被投影到$V$维度的$L$单纯形中,每个都使用softmax操作。该过程在预训练期间将表示条件化到受约束的空间上,并赋予组稀疏性的归纳偏差。对于下游分类,我们正式证明了SEM表示导致更好的泛化比未规范化的表示。此外,我们通过经验证明,使用SEM训练的SSL方法在自然图像数据集(如CIFAR-100和ImageNet)上的泛化能力有所提高。最后,当在下游分类任务中使用时,我们表明SEM特征表现出紧急语义一致性,其中小组学习特征明显预测语义相关类。

英文摘要:

Simplicial Embeddings (SEM) are representations learned through self-supervised learning (SSL), wherein a representation is projected into $L$ simplices of $V$ dimensions each using a softmax operation. This procedure conditions the representation onto a constrained space during pretraining and imparts an inductive bias for group sparsity. For downstream classification, we formally prove that the SEM representation leads to better generalization than an unnormalized representation. Furthermore, we empirically demonstrate that SSL methods trained with SEMs have improved generalization on natural image datasets such as CIFAR-100 and ImageNet. Finally, when used in a downstream classification task, we show that SEM features exhibit emergent semantic coherence where small groups of learned features are distinctly predictive of semantically-relevant classes.

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。

↑