Core Concepts
Effective contrastive losses in SRL depend on gradient dissipation, weight, and ratio components.
Stats
"Sentence Representation Learning (SRL) is a crucial task in Natural Language Processing (NLP), where contrastive Self-Supervised Learning (SSL) is currently a mainstream approach."
"Contrastive SSL outperforms non-contrastive SSL significantly in SRL."
"Two questions arise: First, what commonalities enable various contrastive losses to achieve superior performance in SRL? Second, how can we make non-contrastive SSL, which is similar to contrastive SSL but ineffective in SRL, effective?"
Quotes
"We propose a unified gradient paradigm for effective losses in SRL, which is controlled by three components: the Gradient Dissipation, the Weight, and the Ratio."
"Our work advances the understanding of why contrastive SSL can be effective in SRL and guides the future design of new optimization objectives."